Friday, 12 December 2014

NETWORKED MINDS: Where human evolution is heading

by Dirk Helbing  [1]
Having studied the technological and social forces shaping our societies, we are now turning to the evolutionary forces. Among the millions of species on earth, humans are truly unique. 
What is the recipe of our success? What makes us special? How do we decide? How will we further evolve? What will our role be, when algorithms, computers, machines, and robots are getting ever more powerful? How will our societies change?

In fact, humans are curious by nature – we are a social, information-driven species. And that is why the explosion of data volumes and processing capacities will transform our societies more fundamentally than any other technology has done in the past.

We continue FuturICT’s essays and discussion on Big Data, the ongoing Digital Revolution and the emergent Participatory Market Society written since 2008 in response to the financial and other crises. If we want to master the challenges, we must analyze the underlying problems and change the way we manage our technosocio- economic systems. Last week we discussed: SOCIAL FORCES: Revealing the causes of success or disaster.

Philosophers and technology gurus are becoming increasingly worried about our future. What will happen if computer power and artificial intelligence (AI) progresses so far that humans can no longer keep up? While a century ago some companies maintained departments of hundreds of people to perform calculations for business applications, for decades a simple calculator has been able to do mathematical operations quicker and more accurately than humans. Computers now beat the best chess players, the best backgammon players, the best scrabble players, and players in many other strategic games. Computer algorithms already perform about 70% of all financial trades, and they will soon drive cars better than humans. 

Will we have artificial super-intelligences or super-humans?

Elon Musk, the CEO of Tesla Motors, recently surprised his followers with a tweet saying that artificial intelligence could "potentially be more dangerous than nukes." In a comment on "The Myth of AI," he wrote:[2]

"The pace of progress in artificial intelligence (I'm not referring to narrow AI) is incredibly fast – it is growing at a pace close to exponential. The risk of something seriously dangerous happening is in the five year timeframe. 10 years at most. Please note that I am normally super pro technology, and have never raised this issue until recent months. This is not a case of crying wolf about something I don't understand.

I am not alone in thinking we should be worried. The leading AI companies have taken great steps to ensure safety. They recognize the danger, but believe that they can shape and control the digital super-intelligences and prevent bad ones from escaping into the Internet. That remains to be seen..." 
So, what will be the future of humans? Will we be enslaved by super-intelligent robots or will be have to upgrade ourselves to become super-humans? Will we be technologically enhanced humans, so-called cyborgs? While all of this sounds like science fiction, given the current stage of technological developments such scenarios can't be fully excluded. In general, it's pretty safe to say that everything that can happen is actually likely to happen sooner or later.[3] However, in the following, I would like to point out another scenario, which I believe is of much greater importance: a scenario of collective inteligence, enabled by the emergence of shared information flows.
It's certainly true that digital devices and information systems are increasingly changing human behaviour and interactions. Just observe how many people are staring at their smartphones while walking in town or even when hanging out with their friends. So, if we want to understand better how the digital revolution might change our society, we must identify the various factors that influence our decision-making. In particular, we need to find out how growing amounts of information and the increased interconnectedness of people may change our behaviour. 

One of the best-known models of human decision-making so far is that of the “homo economicus.” It is based on the assumption of perfect egoists, i.e. selfish, rational, utility-maximizing individuals and firms, where the "utility function" is imagined to represent payoffs (i.e. earnings) or stable individual preferences. Related to this, any behaviour deviating from such selfishness is believed to create disadvantages. It is straightforward to conclude that humans or companies who aren't selfish ultimately lose the evolutionary race with selfish ones. So, natural selection should eliminate other-regarding behaviour as a consequence of the principle of the "survival of the fittest." So we should all act selfishly and optimize our payoff.

The hidden drivers of our behaviour

Surprisingly, empirical evidence is not well compatible with this perspective (see Information Box 1). Therefore, I am offering here a novel, multi-dimensional perspective on human decision-making: I claim that self-regarding rational choice is just one way of decision-making people are capable of and that human decisions are often driven by other factors. Specifically, I argue that people are driven by different incentive systems, and that their number increases with human evolution. 

The so-called neocortex is typically considered to be responsible for rational decision-making and the last important brain area that has developed. Before, other parts of the brain areas (such as the cerebellum) were in control – and may still be from time to time... So, I claim that there are many other drivers that govern people's behaviours, too.

It is clear that, first of all, our body has to make sure that we take care of our survival, i.e. we look for water and food. For this, our body comes up with the feelings of hunger and thirst. If one hasn't had water or food for a long time, it will be pretty difficult to focus on mathematical calculations, strategic thinking, or maximizing a payoff function. 

A similar thing applies to sexual desires. There is obviously a natural incentive to promote reproduction, and for many people long-term abstinence can lead to sexual fantasies occupying their thinking. Trying to find sexual satisfaction can be a very strong drive of human behaviour. This explains some pretty irritating behaviours of sexually deprived people, which are often discussed away as "irrational." 

Sex, drugs and rock 'n roll

Similar things can be said about the human desire to possess. Our distant ancestors were hunters and gatherers. Accumulating food and other belongings was important to survive difficult times, to enable trade, and to gain power. This desire to possess can, in some sense, be seen as the basis of capitalism. 

But besides the desire to possess things, some of us also like to experience adrenaline kicks. These were important to prepare our bodies for fights or for fleeing from predators and other dangers. Today, people watch crime series on TV or play shooter games to get the thrill. Like sexual satisfaction, the desire to possess and adrenaline kicks come along with emotions: greed and fear. Financial traders know this very well.

Hunger for information

Intellectual curiosity is a further driver of our behaviour that comes primarily into play, when the previously mentioned needs are sufficiently satisfied. Curiosity serves to explore our environment and to reveal its success principles. By understanding how our world works, we can manipulate it better to our advantage. A trade-off between exploration and exploitation is part of all long-term reward maximizing algorithms. Individuals who only rely on known sources of rewards are quickly be outcompeted by those who explore and find richer sources to exploit. To make sure that we make sufficient efforts to study our environment, our brain rewards insights by hormone flashes, for example, dopamine-based ones. The effect of these hormones is excitement. In fact, as intellectuals and other people know, thinking can create great pleasure. 

Lessons learned

In summary, our body has several different incentive and reward systems. Many of them are related with intrinsic hormonal, emotional, and nervous processes (the latter including the Amygdala brain area and the solar plexus). When neglecting these factors, I claim, human behaviour cannot be well understood. Hence, a realistic description of human decision-making must take knowledge from the sciences studying brain and body into consideration.

For example, why do many people spend much time and energy on sports to an extent that has little material or reproductive benefits? Why do people buy fast and expensive cars that do not match their stated preferences? Why do people race or fight, ride rollercoasters or do bungee jumping? It's the adrenaline kicks that can explain it! This is also the reason why the principle of "bread and games" is so effective in satisfying people. 

The above observations have important implications: humans cannot be simply grasped as payoff maximizers, but as individuals who have evolved to maximize their success in many dimensions, which are often incompatible. They are driven by a number of different incentive and reward systems. In the evolutionary game of survival, reproduction, spreading of ideas, and other things that matter, different strategies can co-exist. Thus the influence of each of these reward systems is likely to be different from one person to the next. This implies different preferences and personalities ("characters"). While some people are driven to possess as much as they can, others prefer to explore their intellectual cosmos, and again others prefer bodily activities such as sex or sports. If nothing grants satisfaction for a long time, the consequence might be to use drugs, get sick, or even die 

Suddenly, "irrational behaviour" makes sense

In other words, when going beyond the concept of self-regarding rational choice, it suddenly becomes clear why there are intellectuals, sportsmen, vamps, divas and other extremely specialized people. In such cases, one reward system dominates the others. For most people, however, all drives are important. But they just don't sum up to define a personal utility function that is stable in time. Instead, each drive is given priority for some time, while the others have to stand back. Once the prioritized drive has been satisfied, another desire is given priority etc. We may compare this a bit with the way different traffic flows are served at an intersection – one after another. Once a vehicle queue has been cleared, another one is prioritized by giving it a green light. Similarly, when one of our drives has been satisfied, we give priority to another one, until the first drive becomes strong again and demands our attention. 

We can also understand what happens, if people are deprived, i.e. cannot satisfy one of their drives for one reason or another. In such cases, it makes sense that they try to get satisfaction from other kinds of activities, which is called compensation. Such a situation applies, for example, to people in poor economic conditions. 

If unable to experience intellectual pleasures (due to lack of education), to satisfy the desire to possess (by consumption), and to gain social recognition, adrenaline kicks will become relatively more important. Therefore, these people might engage more in violence, crime, or drug consumption, as they lack alternatives to find satisfaction. Such deprivation may also explain crime statistics or hooliganism in sports. Therefore, understanding human nature will enable entirely new cures of long-standing social problems, and it allows us all to benefit, too! 

Multi-billion dollar industries for each desire

It turns out that our societies have organized our whole lives around the various incentives driving human behaviour. In the morning, we have breakfast to eat and drink. Then, we go to work to earn the money we want to spend on shopping, thereby satisfying our desire to possess. Afterwards, we may do sports to get our adrenaline kicks. To satisfy our social desires, we may meet friends or watch a soap opera. At the end of the day, we may read a book to stimulate our intellect and have sex to satisfy this desire, too. In conclusion, I dare to say that, most of the time, people's behaviours are not well described by strategic optimization of one utility function that is stable in time.[4] Therefore, the basis of our currently established decision theory is flawed. Nevertheless, our economy is surprisingly well fitted to human nature!

Interestingly, we have created multi-billion-dollar industries around each of our drives, but so far, scientists haven't mostly seen it this way. We have built a food industry, supermarkets, restaurants and bars to satisfy our hunger and thirst, shopping malls to satisfy our desire to possess, stadia to get adrenaline kicks by watching our favourite sports team or by doing sports ourselves. We have a porn industry and perhaps prostitution to help satisfy sexual desires. And we read books, solve riddles, travel to cultural sites, or participate in interactive online games to stimulate our intellect and satisfy our curiosity. This is what our media and tourism industries are for.

Note, however, that there is a natural hierarchy of desires, and this explains the order in which these industries came up. Therefore, each newly emerging industry also changes the character of our society: it gives more weight to desires that were previously in the background. So, what are the drives that will determine our future society? 

The currently fasted growing economic sector is Information and Communication Technology. So, after all our other needs have been taken care of, we are now building a new industry to satisfy the desires of the "information-driven species" that we are. This trend will give everything related to information a much higher weight. In other words, the digital society to come will be much more determined by ideas, curiosity and creativity. But not only this...

Being social is rewarding, too

Humans are not only driven by the above mentioned reward systems. We are also social beings, driven by social desires. In fact, most people have empathy (compassion) – they feel with others. Empathy is reflected by emotions and expressed to others by mimics. It even seems that humans all over the world share a number of facial expressions (anger, disgust, fear, happiness, sadness, and surprise). According to Paul Ekman (*1934), these expressions are surprisingly universal, i.e. independent of language and culture. However, our social desires go further than that. For example, we seek social recognition. 

I argue that the increasing networking of people, supported by Social Media such as Facebook, Twitter and WhatsApp, have the potential to fundamentally change our society and economy. Such social networking through information and communication systems can potentially stimulate our curiosity, strengthen our social desires, and enable collective intelligence, if the information systems are well designed. The main reason for this is that, nature created us as social beings and "networked minds.” 

The evolution of "networked minds"

It's an interesting question to ask, why we are social beings at all? Why do we have social desires? And how is this compatible with the previously mentioned principles of selfishness and survival of the fittest? To study this, we developed a computer simulation describing interactions of utility-maximizing individuals, exposed to the merciless forces of evolution. Specifically, we simulated interactions of individuals facing a so-called "Prisoner's Dilemma" – a particular social dilemma situation, where it would be favourable for everyone to cooperate, but where non-cooperative behaviour is tempting and cooperative behaviour is risky. In Prisoner's Dilemma interactions, the selfish "homo economicus" would never cooperate, as non-cooperative behaviour creates more payoff. This, however, destabilizes cooperation and produces an outcome that is bad for everyone. Although nobody wants this, the desirable state of cooperation breaks down pretty much as free traffic flow breaks down on busy roads – each agent seeks small advantages to themselves that collectively make everyone worse off. The result is a "tragedy of the commons." In other words, the favourable outcome of cooperation does not occur by itself, and instead, an undesirable outcome results. 

In our computer simulations of the Prisoner's Dilemma interactions, we distinguished the actual behaviour – cooperation or not – from the preferred behaviour. We assumed that the preferred behaviour results from a trait determining the degree of other-regarding preferences, which we called the "friendliness." Our computer agents, which represented the individuals, were assumed to decide according to a best-response rule, i.e. to choose the behaviour that maximized their utility function, given the behaviours of their interaction partners (their neighbours). This assumption was mainly made to be acceptable to mainstream economics. The utility function was specified such that it allowed to consider not only the own payoffs. It was possible to give some weight to the payoffs of their interaction partners, too. This weight represented the "friendliness" and was set to zero for everyone at the beginning of the simulation. So, initially the payoff of others was given no weight, and everyone was unfriendly.

Furthermore, the friendliness trait was assumed to be inherited to offspring (either genetically or by education). In our computer simulations, the likelihood to have an offspring increased exclusively with their own payoff, not the utility. The payoff was set to zero, when a co-operating agent was exploited by all neighbours (i.e. if none of them cooperated). Therefore, such agents never had any offspring. 

Finally, if agents earned payoffs and had offspring, the inherited friendliness value tended to be that of the parent, but there was also a certain natural mutation rate, which was specified such that it did not promote friendliness. 

So, what results did our computer simulations produce? The prevailing outcome of the evolutionary game-theoretical computer simulations was indeed a self-regarding, payoff-maximizing "homo economicus," as expected. However, this applied only to most parameter combinations of our simulation model, not all of them (see figure above). When offspring tended to live close to their parents (i.e. intergenerational migration was low), a friendly "homo socialis" with other-regarding preferences resulted instead! Interestingly, this fits the conditions under which humans actually raise their children. 

This evolution of other-regarding preferences (not just other-regarding behaviour, i.e. cooperation) is quite surprising. Even though none of the above model assumptions promotes cooperative behaviour or other-regarding preferences in separation, in combination they are nevertheless creating socially favourable behaviour. This can only be explained as result of interaction effects between the above rules. Another interesting finding is the evolution of "cooperation between strangers," i.e. the occurrence of cooperation between genetically non-related individuals. Video illustrating this (see also the related figure below).

Making mistakes is crucial 

How can we understand the surprising evolution of other-regarding preferences? We need to recognize that random mutations generate a low level of friendliness by chance. This slight other-regarding preference creates conditionally cooperative behaviour. That is, if enough neighbours cooperate, a "conditional co-operator" will be cooperative as well, but not so if too many neighbours are uncooperative. 

Unconditionally cooperative agents with a high level of friendliness are born very rarely, and only by chance. These "idealistic" individuals will usually be exploited, have very poor payoffs, and no offspring. However, if born into a neighbourhood with enough agents, who are sufficiently friendly to be conditionally cooperative, an unconditionally cooperative "idealist" can trigger a cooperative behaviour of neighbours in a cascade-like manner.[5]

In the resulting cooperative neighbourhood, high levels of friendliness are passed on to many offspring such that other-regarding preferences spread. This holds, because greater friendliness now tends to be profitable, in contrast to the initial stage of the evolutionary process, when friendly people were rare outliers and lonely outsiders. In the end, co-operators earn higher payoffs on average than non-cooperative agents: if everyone in the neighbourhood is friendly, everyone has a better life. Therefore, while the "homo economicus" earns more initially, the finally resulting "homo socialis" eventually beats the "homo economicus" (see figure above). In the end, the friendliness levels are broadly distributed (see figure below). This explains the heterogeneous individual preferences that are actually observed: in reality, everything from selfish to altruistic preferences exists.

Note that in the situation studied above, where everyone starts as a non-cooperative "homo economicus," no single individual can establish profitable cooperation, not even by optimizing decisions over an infinitely long time horizon. It takes several "friendly" deviations in the same neighbourhood to trigger a cascade effect that eventually changes the societal outcome. One can show that a critical number of interacting individuals is needed to be friendly and cooperative by coincidence. Therefore, the "homo socialis" can only evolve thanks to the occurrence of random "mistakes" (here: the birth of "idealists" who are initially exploited by everyone). However, given suitable feedback effects, such "errors" enable better outcomes. Here, they eventually produce an "upward spiral" to cooperation with high payoffs. Thereby, idealists make it possible to overcome the "tragedy of the commons." 

"Networked minds" require a new economic thinking

The most important implication of the evolution of other-regarding preferences is that, by considering the payoff and success of others, decisions become interdependent. Therefore, while methods from statistics for independent, un-correlated events may sometimes suffice to characterize decisions of the "homo economicus," we need complexity science to understand the interdependent decision-making of the "homo socialis." In fact, the "homo socialis" may be best characterized by the term "networked minds."

In agreement with the findings of social psychology, the "homo socialis" is capable of empathy and often puts himself or herself into the shoes of others. By taking into account the perspective, interests, and success of others, "networked minds" consider externalities of their decisions. That is, the "homo socialis" decides differently from the "homo economicus." While the latter would never cooperate in a social dilemma situation, the "homo socialis" is conditionally cooperative, i.e. tends to cooperate if enough neighbours do so as well. Therefore, the "homo socialis" is able to align competitive individual interests and to make the individual and system optimum better compatible with each other. 

This makes the "homo socialis" superior to the "homo economicus," even if we measure success in terms of individual payoffs. While the Invisible Hand often doesn't work for the "homo economicus" in social dilemma situations, as we have seen, the "homo socialis" manages to make the Invisible Hand work by considering externalities. Therefore, while increasing the individual utility, the "homo socialis" manages to create systemic benefits, too, in contrast to the "homo economicus." Interestingly, the successful cooperative outcome emerging for the "homo socialis" is not the result of an optimization process, but rather of an evolutionary process. 

All the above calls for a new economic thinking ("economics 2.0"), and even enables a better organization of economy, as I will discuss it in the next chapter (see also Information Box 2). I strongly believe that we are heading towards a new kind of economy, not just because the current economy will not provide enough jobs anymore in many areas of the world, but also because information systems and social media are opening up entirely new opportunities. Moreover, to cope with the increasing level of complexity of our world, we need to enable collective intelligence, fostering not just the brightest minds and best ideas, but also learning how to leverage the hugely diverse range of experiences and expertise of people in parallel. And this again needs "networked minds."

The wisdom of crowds

Since the "wisdom of the crowd" was first discovered and demonstrated, people have been amazed by the power of collective intelligence. The "wisdom of the crowd" reflects that the average of many independent judgments is often superior to expert judgments. A frequently cited example first reported by Sir Francis Galton (1822-1911) is the estimation of the weight of an ox. Galton observed villagers trying to estimate the weight of an ox at a country fair, and noted that, although no one villager guessed correctly, the average of everyone’s guesses was extremely close to the true weight. Importantly, today the wisdom of crowds is considered to be the underlying success principle of democracies and financial markets. Of course, an argument can also be made for the "madness of crowds." In fact, when people influence each other, the resulting group dynamics can create very bad outcomes. When individuals copy each other, misjudgements can easily spread. For the wisdom of crowds to work, independent information gathering and decision-making are crucial. The design of the decision mechanism determines, whether the result of many decisions will be a success or failure (see Information Box 3).

The Netflix challenge

One of the most stunning examples for collective intelligence is the outcome of the Netflix challenge. Based on movie ratings by their customers, Netflix was trying to predict what movies they would love to see. But the predictions were frustratingly bad. So, back in 2006, Netflix offered a prize of 1 million US dollars to the team that was able to improve their own predictions of user-specific movie ratings by more than 10 percent. About 2,000 teams participated in the challenge and sent in 13,000 predictions. The training data contained more than 100 million ratings of almost 20,000 movies on a five-star scale by approximately 500,000 users. Netflix' own algorithm produced an average error of about 1 star, but it took three years to improve it by more than 10 percent. 

In the end, "BellKor's Pragmatic Chaos team" won the prize, and a number of really remarkable lessons were learned: First, given that it was very difficult and time-consuming to improve only 10 percent over the standard method, Big Data analytics isn't that powerful in predicting people's preferences and behaviours. Second, even a minor improvement of the algorithm by only 1 percent created a significant difference in the top-10 ranked movies that were predicted for the users. In other words, the results were very sensitive to the method used (rather than stable). Third, no single team was able to achieve a 10 percent improvement alone. 

A step change in performance was only made when the best-performing predictions were averaged with predictions of teams that weren't as good. That is, the best solution is actually not the best – averaging over diverse and independently gained solutions beats the best solution. This is really counter-intuitive: nobody is right, but together with others one can do a better job! The mechanism for this is subtle and extremely important for collective wisdom. Although each of the top teams had made almost a 10% improvement over the original algorithm, each used different methods that were able to find different patterns in the data. No single algorithm could find them all. By averaging the predictions, each algorithm contributed the knowledge it was specialized to find, and the errors of each algorithm were suppressed by the others. Thus, when complex tasks must be solved, specialization and diversity are key!

Actually, things were even more surprising than that: when giving better predictions a higher weight, it typically didn't improve the predictions. Researchers have argued that this is because weighting more successful algorithms more highly only works if at least one algorithm is correct. But in this case no single algorithm was perfect, and an equal combination was better than any solution alone or a weighted average that considered the relative ranks of the algorithms. This is probably the best argument for equal votes – but equal votes for different solution approaches, not for people! In other words, one should not favour majority solutions. Compared to our way of decision-making today, minority votes would need to have a higher weight – such that they enter the decision-making process. That would correspond to a democracy of ideas rather than a democracy of people. In other words, to take the best possible decisions, we would have to say good-bye to two approaches that are common today: first, the principle that the best expert takes the decision in a top-down way; second, the principle of majority voting. Therefore, if we want to take better decisions, we must question both, the concept of the "wise king" (or "benevolent dictator") and the concept of democracies based on majority opinions. This is shocking!

How to create collective intelligence

So, how could we create a better system? How can we unleash the power of "collective intelligence"? First, we have to abandon the idea that our reality can be well described by a single model – the best one that exists. In many cases, such as traffic flow modeling, there are several similarly performing models. This speaks for a pluralistic modeling approach. In fact, when the path of a hurricane is predicted or the impact of a car accident is simulated in a computer, an average of several competing models often provides the best prediction. 

In other words, the complexity of our world cannot be grasped by a single model, mind, computer, or computer cluster. Therefore, it's good if several groups, independently of each other, try to find the best possible solution. These, however, will always give an over-simplified picture of our complex world. Only if we put the different perspectives together, then we can get a result that approximates the full picture well. We may compare this with visiting an artfully decorated cathedral. Every photograph taken can only reflect part of its complexity and beauty. One photographer alone, no matter how talented or how well equipped, cannot capture the full 3D structure of the cathedral with a single photograph. A full 3D picture of the cathedral can only be gained by combining many the photographs representing different perspectives and projections.

Let's discuss another complex problem, namely the challenge to find the right insurance for you. It will certainly be impossible for any consumer to read the small print and detailed regulations of all available insurances. Instead, you would probably ask your colleagues and friends what experiences they have made with their insurances, and then evaluate the most recommended ones in detail to find the right insurance for you. Insurance companies that provide bad coverage or service create bad word-of-mouth reviews, making them less likely to be chosen by others. In other words, we evaluate insurances collectively, thereby mastering a job that nobody could do alone. In the Internet age, this word-of-mouth system is increasingly replaced by online reviews and price comparison websites, which widens the circle of people contributing additional information and improves the chances for each individual to take better decisions.

While this approach is able to create additive knowledge, science has found ways to create knowledge that is more than just a sum of all knowledge. In fact, when experts discuss with each other or engage in an exchange of ideas, this often creates new knowledge. The above examples illustrate how collective intelligence works: one needs to have a number of independent teams, which tackle a problem in separation, and after this, the independently gained knowledge needs to be combined. When there is too much communication in the beginning, each team is tempted to follow the successes of others, reducing the number of explored solutions. But when there is too little communication at the end, it's not possible to fully exploit all the solutions that have been found.

At this place, it is also interesting to discuss how "cognitive computing" works in IBM's Watson computer. The computer scans hundreds of thousands sources of information, for example, scientific publications, and extracts potentially relevant statements. But it can also formulate hypotheses and seek evidence for or against them. It then comes up with a list of possible answers and ranks them according to their likelihood. All of this is done using algorithms based on the laws of probability: how probable is this hypothesis given the observed data? These laws codify precisely and mathematically the type of reasoning humans informally perform when making decisions. However, Watson loses less information due to cognitive biases, or through the limited time and attention span humans have. For example, when used in a medical context, Watson would come up with a ranked list of diseases that are compatible with certain symptoms. A doctor will probably have thought of the most common diseases already, but Watson will also point the attention to rare diseases, which may otherwise be overlooked. 

Importantly, to work well, Watson should not be fed with consistent information. It must get unbiased information reflecting different perspectives and potentially even contradictory pieces of evidence. Watson is then trained by experts to weight evidence and sources of information in ways that are increasingly consistent with current wisdom. In the end, Watson may be doing better than humans. The power of Watson is in the sheer number of different sources of information it can scan, and the number of hypotheses it can generate and evaluate, both orders of magnitude above any single human. Humans tend to seek and attend to information that confirms their existing beliefs; Watson is largely immune to this bias. Humans tend to weight evidence more highly when it confirms their beliefs; Watson evaluates every piece of information algorithmically, according to the laws of probability.

Let us finally address a question that thrills many people these days: Using the future Internet, could we create something like a globe-spanning super-intelligence? In fact, the Google Brain project may want to establish such a super-intelligence, based on Google's massive data of our world. However, what we have discussed above suggests that it is important to have different independent perspectives – not just one. So, having many brains is probably superior to having one super-brain. Remember, the "wisdom of crowds" is often outperforming experts.[6] This implies a great potential of citizen science. Collective intelligence can beat super-intelligence, and a diversity of perspectives is key to success. Therefore, to master the complex challenges of the future, we need a participatory approach, as I we will discuss it in the next chapter.
There is more to come: New dimensions of life

To conclude, diversity is a major driving force of evolution, and has always been. Over millions of years, diversity has largely increased, creating a growing number of different species. Diversity drives differentiation and innovation, such that new dimensions of life are created. Eventually, humans became social and intelligent beings, and cultural evolution set in. The slow evolution of genetic fitness was then complemented by an extremely fast evolution of ideas. One might therefore even say that, to a considerable extent, humans have emancipated themselves from the limitations of matter and nature. The spreading of ideas, of so-called "memes," has become more important than the spreading of genes. Now, besides the real world, digital virtual worlds exist, such as massive multi-player on-line games. So, humans have learned to create new worlds out of nothing but information. The multi-player online games Second Life, World of Warcraft, Farmville, and Minecraft are just a few examples for this. 

It is equally fascinating that, with these digital worlds, new incentive systems have evolved, too. We are perhaps not so surprised that some people care about their position in the Fortune 500 list of richest persons, because it reflects their financial power in our real world. But people feel not just competitive about money. Tennis players and soccer teams strongly care about their ranking. Actors live on the applause they get, and scientists care about citations, i.e. the number of references to their work. 

So, people do not only respond to material payoffs such as money, and the various other drives we have discussed before. It turns out that many people also care about the scores they reach in gaming worlds. Even though some of these ranking scales don't imply any immediate material or other real-world value, they can motivate people to make an effort. It's pretty surprising how much time people may spend on increasing their number of Facebook friends or Twitter followers, or their klout score. Obviously, social media offer new opportunities to create multi-dimensional reward systems, as we need them to enable self-organizing socio-economic systems. 

There is little doubt: we are now living in a cyber-social world, and the evolution of global information systems drive the next phase of human social evolution. Information systems support "networked minds" and enable "collective intelligence." Humans, computers, algorithms and robots will increasingly weave a network that may be characterized as "information ecosystem," and therefore one question becomes absolutely crucial: "How will this change our socio-economic system?"

INFORMATION BOX 1: How selfish are people really?

Our daily experience tells us that many people do unpaid jobs for the benefit of others. A lot of volunteers work for free, some organize themselves in non-profit organizations. We also often leave tips on the restaurant table, even if nobody is watching and even if we'll never return to the same place (and that's true also for countries, where tips are not kind of obligatory as in the USA). Furthermore, billionaires, millionaires and normal people make donations to promote science, education, and medical help, often in other continents. Some of them do it even anonymously, i.e. they will never get anything in return – not even recognition.

This has, indeed, puzzled economists for quite some time. To fix the classical paradigm of rational choice based on selfish decision-making, they eventually assumed that everyone would have an individual utility function, which reflects personal preferences. However, as long as there is no theory to predict personal preferences, the concept of utility maximization does not explain much. Taking rational choice theory seriously, it claims that people, who help others, must have fun doing so, otherwise they wouldn't do it. But this appears to be a pretty circular conclusion.

Ultimatum and Dictator Games

In order to test economic theories and understand personal preferences better, scientists have performed ever more decision experiments with people in laboratories. Their findings were quite surprising and totally overhauled previously established economic theories. In 1982, Werner Güth developed the "Ultimatum Game" to study stylized negotiations. In related experiments, for example, 50 dollars are given to one person (the "proposer"), who is asked to decide how much of this money he or she would offer to a second person (the "responder"). If the responder accepts the amount offered by the proposer, both get the respective share. However, if the responder rejects the offer, both walk home with nothing.
According to the concept of the self-regarding "homo economicus," the proposer should offer not more than 1 dollar, and the responder should accept any amount – better get a little money than nothing! However, it turns out that responders tend to reject small amounts, and proposers tend to offer about 40 percent of the money on average. A further surprise is that proposers tend to share with others in all countries of the world. Similar experimental outcomes are found when playing for a monthly salary. To reflect these findings, Ernst Fehr (*1956) and his colleagues proposed inborn principles of fairness preferences and inequality aversion. Others, such as Herbert Gintis, assumed a genetic basis of cooperation ("strong reciprocity").
There is also a simpler game, known as "Dictator Game", which is in some sense even more stunning. In this game, one person is asked to decide, how much of an amount of money received from the experimenter he or she wants to give to another person – it can also be nothing! The potential recipient does not have any influence on the outcome. Nevertheless, many people tend to share – on average about 20 percent of the money they receive from the experimenter. Of course, there are always exceptions in positive and negative direction – some people actually don't share.

The surprising overall tendency to share could, of course, result from the feeling of being observed, which might trigger behaviours complying with social norms. So, would such sharing behaviour disappear when decisions are taken anonymously? To test this, we made a Web experiment with strangers who never met in person. Both, the proposer and responder got a fixed amount of money for participating in the experiment. However, rather than sharing money, the proposer and responder had to decide how to share a certain work load: together, they had to do several hundred calculations! In the worst case, one of them would have to do all the calculations, while the other would get money without working! But to our great surprise, the participants of the experiment tended to share the workload in a rather fair way. Thus, there is no doubt: many people decide in other-regarding ways, even in anonymous settings. They have a preference for fair behaviour.

INFORMATION BOX 2: A smarter way of interacting, not socialism 

In contrast to today's re-distribution approach based on social benefit systems, the "homo socialis" should not be considered as a tamed "homo economicus", who shares some payoff with others. As we have discussed before, the "homo economicus" tends to run into "tragedies of the commons," while the "homo socialis" can overcome them by considering the externalities of decisions. So, the "homo socialis" can create more desirable outcomes and higher profits on average. Therefore, when taking decisions like a "homo socialis," we will often be doing well.

In social dilemma situations the "homo economicus," in contrast, tends to produce high profits for a few agents who exploit others, but poor outcomes for the great majority. Therefore, redistributing money of the rich can't overcome "tragedies of the commons" and can't reach average profit levels that are comparable to those of the "homo socialis".[7] In conclusion, an economy in which the "homo socialis" can thrive is much better than an economy, in which the "homo economicus" dominates and where social policies try to fix the damages afterwards. Therefore, the concept of the "homo socialis" has nothing to do with a re-distribution of wealth from the rich to the poor.

Let me finally address the question, whether friendly, other-regarding behaviour is more likely when people have a lot of resources and can "afford" to consider the interest of others, or whether it occurs under particularly bad conditions. In fact, in the desert and other high-risk environments, people can only survive by means of other-regarding behaviour. However, such behaviour can create benefits also in low-risk environments, where people can survive by themselves. This is so, because the consideration of externalities of the own behaviour brings the system optimum and the individual user optimum into harmony. In other words, when considering externalities, as the "homo socialis" does, the socio-economic system can perform better, creating on average higher advantages for everyone. Even in a world with large cultural differences across cities, countries and regions, it seems that countries and cities with a particularly high quality of life are those that manage to establish other-regarding behaviours and take externalities into account. As I said before, the emergence of friendly, other-regarding behaviour is to the own benefit, if just enough interaction partners behave in this way. It is, therefore, desirable to have institutions that protect the "homo socialis" from exploitation by the "homo economicus." Reputation systems are one such institution. They can promote desirable outcomes in a globalized world.

INFORMATION BOX 3: Crowds and swarm intelligence

In the past years, the concept of crowds and swarm intelligence has increasingly fascinated the public and the media. At the time of Gustave Le Bon (1841-1931), the idea came up that people had something like a shared mind. However, the attention was put on "the madness of crowds," on mass psychology that can create, for example, a rioting mob. This was seen to be a result of dangerous emotional contagion, and it is still the reason why governments tend to feel uneasy about crowds. But crowds can have good sides, too.
Today, we have a more differentiated picture of crowds and swarm intelligence. We know much better, when crowds perform well, and when they cause trouble. Simply put, if people gather information and decide independently from each other, and the information is suitably aggregated afterwards, this often creates better results than even the best experts can produce. This is also more or less the way, in which prediction markets work. These have been surprisingly successful in anticipating, for example, election outcomes or the success of new movies. Interestingly, prediction markets have been inspired by the principles that ants or bees use to find the most promising food sources. In fact, such social insects have always amazed people for their complex self-organizing animal societies, which are based on surprisingly simple interaction rules, as we know today.
In contrast to the above, it often undermines the "wisdom of crowds," when people are influenced while searching information or making up their minds. This is best illustrated by the Asch conformity experiments, in which an experimental subject had to publicly state, which one of three lines had an identical length as another line that was shown. However, before answering, other subjects were giving wrong answers. As a consequence, the experimental subject typically gave a wrong answer, too. Moreover, recent experiments I performed together with Jan Lorenz, Heiko Rauhut, and Frank Schweitzer show that people are influenced by opinions of others even when no group pressure is put on them.
What conclusions can we draw? First, one shouldn't try to influence others in their information search and decision making, if we want the "wisdom of crowds" to work. Second, good education is probably the best immunization against emotional contagion, and can therefore reduce negative effects of crowd interactions. And third, we must further explore, what decision-making procedures and institutions can maximize collective intelligence. This will be of major importance to master the increasingly difficult challenges posed by our complex globalized world. 

[1] I would like to thank Richard Mann for his useful comments on this chapter. 

[2] The comment on "The Myth of AI" was originally posted at but apparently deleted in the meantime. 

[3] The smaller the probability of the event, however, the longer it will usually take until it happens. 

[4] But note that nonetheless, people make informed trade-offs, such as avoiding to spend too much money on parties, if they have the goal to possess something. 

[5] One might think that this is what happened, for example, in the case of Jesus Christ. He preached to "love your neighbour as yourself," i.e. other-regarding behaviour weighting the preferences of others with a weight of 0.5. His idealistic behaviour was painful for himself. He ended on the cross like a criminal and without any offspring. But his behaviour caused other-regarding behaviour to spread in a cascade-like way, establishing a world religion. 

[6] While Google can easily implement many different algorithms, the very nature of a large corporation with its self-regarding goals, uniform standards, hiring practices and communications means that the teams developing these will be more prone to observe and follow each other’s successes, think in similar ways and thus produce less diverse opinions. 

[7] when the latter interact among each other

Thursday, 4 December 2014

SOCIAL FORCES: Revealing the causes of success or disaster

by Dirk Helbing

We have seen that self-organizing systems can be very effective and efficient, but their macro-level behavior crucially depends on the interaction rules, interaction strength, and institutional settings. To get things right, it's important to understand the factors that drive the dynamics of the system. 
In physics, many phenomena can be understood by means of forces, and it takes suitable measurements to reveal them. I show that in socio-economic systems, too, success and failure depend on hidden forces, which are not directly accessible to our senses. But now, the data about our world increasingly allow us to measure the forces underlying socio-economic change, empowering us to take better decision and more effective actions in the future.

We continue FuturICT’s essays and discussion on Big Data, the ongoing Digital Revolution and the emergent Participatory Market Society written since 2008 in response to the financial and other crises. If we want to master the challenges, we must analyze the underlying problems and change the way we manage our technosocio- economic systems. Last week we discussed: HOW SOCIETY WORKS:Social order by self-organization
Societies around the world are suffering from financial crises, crime, conflicts, wars, and revolutions. These "societal diseases" do not occur by chance, but for a reason. The fact that they are happening time and again proves that there are hidden regularities behind them, but that we haven't understood them well. This is why we fail to master them. But in the future, I argue, we will be able to understand societal diseases and cure them. One might imagine this to work in a similar way as the x-ray discovered by Wilhelm Conrad Röntgen (1845-1923), which has helped to cure diseases of billions of people by revealing where something is broken in our bodies, or wrong.

In fact, the growing amounts of data about our world will allow us to develop entirely new measurement methods to see hidden patterns in the activities of our global techno-socio-economic-environmental system, very much like microscopes and telescopes in the past have enabled us to discover and understand the micro- and macro-cosmos around us. As we have built elementary particle accelerators to discover the forces that keep our world together, we can now create Socioscopes to reveal the principles that make our society succeed or fail.

Given that the loss of control over a system often results from a lack of knowledge about the rules governing it, it is important that we learn to measure these hidden forces, which govern our economy and societies. This will also put us into a position to use these forces such that we can overcome systemic instabilities and crises. To achieve this, we don't have to save all the data in the world in one database. It's much better to perform tailored measurements as needed for the respective question or purpose. But how can we proceed?

Measuring the world 2.0


Some of the greatest discoveries in human history were made by measuring the world. We have discovered new cultures and new continents. Reaching out for the skies, we have explored our universe and discovered black holes, dark matter, and entirely new worlds. Now, the Internet is offering novel opportunities to quantify our world. Performing a sentiment analysis of blogs, Facebook posts, or tweets, we can visualize emotions of people such as happiness (see, for example, We Feel Fine, Hedonometer, and Sentistrength. We can also get a picture of the social, economic, and political "climate," by determining the subjects that people publicly communicate about see the GDELT project . Mining data on the Web, one can further determine the gross domestic product per capita,[1] violence, and crime[2], as a function of location and time, Se Gapminder . One can even reconstruct the three-dimensional world around us from the photos that people upload on platforms such as Flicker all the time, see Video .

Furthermore, we can create Financial Crisis Observatories to detect the likelihood of financial bubbles and crashes See . We can build RiskMaps and CrisisMaps, to support first aid teams in a disaster-struck area See . And we can map innovations, the spreading of knowledge, and scientific concepts, as I have started to do it with Amin Mazloumian, Katy Börner, Tobias Kuhn, Christian Schulz and others (see pictures below and the Living Science app . The spreading of culture in the world over the centuries, as studied by Maximilian Schich together with Laszlo Barabasi, myself and others, is particularly fascinating (Video 1 , Video 2 , Video 3 . Inspired by Wikipedia and OpenStreetMap, we could also create an OpenResourcesMap visualizing the resources of the world, and who uses them, to help reduce undesirable shortages. And we could produce an OpenEcosystemsMap to depict environmental change and who causes it.

Very soon, we will have not only maps accumulating past activities, but also systems delivering real-time answers. When asking a question, this will trigger tailored measurements telling us, for example, what is the traffic situation in London's Oxford street? What's the weather in Moscow, and how might this affect investor decisions and consumer choices? How happy are people in London today, and how much money will they spend shopping? Why did people in Switzerland vote against "mass migration"? What worries people in Paris at the moment? How many people are up between 3 and 4am on Sunday nights around Manhattan's Central Square, and is it worth selling pizza at that time? How noisy is it to live in the quarter of town I am considering to move to? What's the rate of flu (or Ebola) infections in the place I want to spend my holidays in? Where are the road bumps in my city located? And when did we have the last earthquake greater than 4 on the Richter scale within a range of 500 kilometers?

Information like this can create an increased awareness of the world around us, and empower us to take better decisions and more effective actions. But how will we get all this real-time information?

Creating a Planetary Nervous System as a Citizen Web 

The sensor network underlying the emerging "Internet of Things" will enable us to perform real-time measurements of almost everything. It will be possible to get all the information needed to establish real-time feedbacks in complex dynamical systems. This will allow us to support a favorable self-organization – not just of traffic lights and production, but basically of everything that requires proper real-time adaptation.

We can use the "Internet of Things" also to build an intelligent information platform called the "Planetary Nervous System" (PNS). This was proposed by the FuturICT project . The Planetary Nervous System would have three main functions: First, to configure sensor networks in order to answer specific queries and provide real-time measurements; second, to create awareness of problems and opportunities around us; and third, to measure the hidden forces underlying socio-economic change and important intangible factors such as trust, reputation, public security or other quantities depending on interactions in social networks.

In fact, my team has already started to develop such an information platform. We will build the Planetary Nervous System as a Citizen Web together with the emerging nervousnet community committed to the further development of the project. This approach will give the citizens control over their personal data, in accordance with their right of informational self-determination, and create new opportunities for everyone. Nervousnet will be engaged in protecting privacy while offering everyone the possibility to contribute to the measurement of our environment. In some sense, the project may be envisaged to be a real-time Wikipedia of our world. Like OpenStreetMap it will build on data provided by many volunteers. A large share of this data will be Open Data, and you will be able to use it to develop your own business. So, why don't you join the nervousnet community today?[3]

The Planetary Nervous System will soon be a large-scale distributed information platform providing real-time social mining services as a public good. While existing Big Data systems threaten social cohesion as they are designed to be closed, proprietary, privacy-intrusive and discriminatory, we will rather create an open, privacy-preserving and participatory platform designed to be collectively built by citizens and for citizens. This will establish a novel social mining paradigm: Users are provided with freedom and incentives to share, collect and, at the same time, protect data of their digital environment in real-time. In this way, social mining turns into a knowledge extraction service for the public good. In perspective, it will provide a public information service for everyone, and perhaps become a foundational public institution for the emerging digital societies of the 21st century. But it takes more than data to understand the world and its problems...

Sociophysics: Revealing the hidden forces governing our society


As previously discussed, analyzing and visualizing data should be only the first step. In the chapter on the Crystal Ball, I have pointed out that data mining alone usually doesn't deliver a good understanding of a complex dynamical system. In order to make sense of data, it's important to have explanatory models, which allow one to make predictions for situations that haven't occurred before.

In the last chapter, we have discussed interaction rules ("social mechanisms") that influence human behavior in a similar way as the gravitational force determines the motion of the planets – there is just more diversity and randomness. Social roles, i.e. behaviors that social norms expect from us, are further examples illustrating the existence of such rules. While the scientific approach of "agent-based simulations" specifies these rules by computer codes, the research field of Sociophysics expresses them by mathematical formulas. In this chapter, I will discuss the powerful concept of "social forces," which allows one to construct a link between the micro-level interactions between individuals and the often unexpected macro-level outcomes in socio-economic systems resulting from them.

The concept of forces is one of the main pillars of physics. In order to discover it, one had to replace the worldview assuming the Earth to be the center of the universe by a worldview with the planets moving around the sun. This new interpretation of planetary motion data allowed Isaac Newton (1642-1727) to formulate a simple and plausible model based on the concept of gravitational forces. By now, most parts of physics are formulated in terms of forces and the way they influence the world. The predictive power of the respective models is striking. It has been impressively demonstrated by the moon shot of the Apollo program and many other examples.

A further aspect that made physics so successful is its long tradition in building instruments to measure things that are otherwise not accessible to our senses. This reaches from the early stages of our universe to the exploration of elementary particles up to the study of processes in biological cells. Therefore, we should ask ourselves how to build "Socioscopes" that can reveal the hidden forces behind the self-organization of socio-economic systems. In this way, we will eventually learn to understand the counter-intuitive behaviors of complex systems. I believe we will soon be able to diagnose emergent "diseases of society" such as financial crashes, crime, or wars, before they happen. This would enable us to avoid or mitigate these problems in a similar way as instruments for medical diagnosis have helped us to cure diseases. Isn't that an exciting perspective?

Social forces between pedestrians


To demonstrate the feasibility of this approach, let me first give an example for the usefulness of force models in the social sciences relating to pedestrian and crowd dynamics. Starting in 1990, when I wrote my diploma thesis, I noticed that pedestrian paths around obstacles looked similar to streamlines in fluids. So, I decided to formulate a fluid-dynamic theory of pedestrian flows, and I derived it from an individual-based pedestrian model, which was inspired by Newton's force model.

This "social force model" assumes that the acceleration, deceleration and directional changes of pedestrians can be approximated by a sum of different forces, each capturing a different desire or interaction effect. For example, the adaptation of the actual pedestrian velocity to the desired speed and preferred direction of motion of a pedestrian can be modeled by a simple "driving force," describing a gradual adaptation of the velocity within a typical time period. Moreover, the desire to avoid collisions and to respect a certain "territory" around others is reflected by repulsive interaction forces between pedestrians with a strength that exponentially decays with distance. Repulsive interactions with walls or streets can be captured by similar forces. The attraction toward tourist sites can be described by attractive forces, and the reason for family members to stay together as well. Finally, a random force may reflect behavioral variability.

It is exciting that computer simulations of this model match many empirically observed phenomena surprisingly well despite its simplicity. For example, it is possible to understand the emergence of river-like flow patterns through a standing crowd of people, the wave-like progression of individuals waiting in queues, or the lower densities on dance floors as compared to the people standing around.

Self-organization of lanes of uniform walking direction


There are also various self-organization phenomena that lead to fascinating collective patterns of motion. For example, when pedestrians are entering a corridor on two sides, we observe the formation of lanes of uniform walking direction, Video. That is, the opposite flows are automatically coordinated in a way that produces an efficient separation of counter-flows. One might see the Invisible Hand at work, here. But we can actually explain how social order is created and how a collectively desirable outcome results from local interactions. Whenever an encounter between two pedestrians occurs, the repulsive interaction force between them pushes the pedestrians a bit to the side. Importantly, these interactions are more frequent between opposite directions of motion, due to the higher relative velocity. This is the main reason that causes opposite directions of motion to separate from each other. A preference of pedestrians to walk on, say, the right-hand side is not needed to explain the phenomenon. From the point of view of complexity science, lane formation is a "symmetry breaking" phenomenon that occurs when a mixture of different directions of motion is unstable.

Walking through a "wall" of people without stopping


Surprisingly, the very same force model also reproduces a number of other interesting findings in pedestrian crowds, such as oscillatory changes of the flow direction at bottlenecks. This results from an alternating pressure relief in the crowd and has inspired the self-organized traffic light control discussed in a previous chapter. Another example of self-organization is the amazing phenomenon of "stripe formation," which allows pedestrians to cross a pedestrian flow without having to stop (see illustrations below). It's almost as if one could walk through a wall! Using the social force model, it's possible to understand how this is possible. The formation of stripes – which occur for similar reasons as the lanes discussed before – allows pedestrians to move forward with the stripes and sideward within the stripes. Taken together, this enables the continuous collective motion through a crossing flow, Video.

Measuring forces


In physics, forces are experimentally determined by measuring the trajectories of particles, especially changes in their speeds and directions of motion. It would be natural to do this for pedestrians, too. At the time when we developed the social force model for pedestrians, I could not imagine it would ever be possible to measure social forces experimentally. But a few years later, we actually managed to do this. Around 2006, with the advent of powerful video cameras and video processing, my former PhD student Anders Johansson was able to extract pedestrian trajectories. We furthermore adapted the parameters of the social force model to optimally reproduce the trajectories with computer simulations of the model. In 2006/07, such tracking methods became also important for the analysis and avoidance of crowd disasters.

Then, in 2008, Mehdi Moussaid and Guy Theraulaz set up a pedestrian experiment in Toulouse, France, under well-controlled lab conditions. This finally empowered us to do data-driven modeling. While before we had to make assumptions on the functional form of pedestrian interactions, it now became possible to determine the functional dependencies directly from the wealth of tracking data generated by our pedestrian experiment. After fitting the social force model to individual pedestrian data, it was finally used to simulate flows of many pedestrians. To our excitement, the computer simulations yielded a surprisingly accurate prediction of the pedestrian flows observed in a wide pedestrian walkway.

So, pedestrian modeling can be considered a great success of Sociophysics. Over time, pedestrian studies had evolved from a social to a natural science, bringing theoretical, computational, experimental and data-driven approaches together. This even led to practical and partly surprising lessons for the better design of pedestrian facilities and the planning of safer crowd events such as the annual pilgrimage in Mecca.

Most pedestrian facilities are inefficiently engineered        

Back in 1994/95, when comparing different designs of pedestrian facilities, Peter Molnar and I surprisingly found that obstacles in the way, if properly placed, can make pedestrian counter-flows more efficient (see figure below). In fact, all the conventional design elements of pedestrian facilities – corridors, bottlenecks, and intersections – turn out to be not well designed. They can be substantially improved! In many cases, "less is more," i.e. providing less space for pedestrians produces better flows. This surprising discovery can be best understood for bottlenecks such as doors. Here, a funnel-shaped design can reduce disturbances in the pedestrian flow, which result when the directions of motion are not well aligned (e.g. when some people approach the door from the front and others from the side).

In busy bi-directional pedestrian flows, the efficiency of motion can be improved by a series of pillars in the middle. It turns out that these pillars help to stabilize the interface between the opposite flows, thereby reducing disturbances. The effectiveness of the design becomes particularly obvious in subway tunnels, where pedestrians move both ways and pillars exist for static reasons.

Finally, an obstacle in the middle of a pedestrian intersection may also improve the flow. When Peter Molnar and I discovered this, it took us a long time to understand. But eventually we noticed that, at intersections, many different collective patterns of motion can emerge, for example, clockwise or counter-clockwise rotary flows, or oscillatory patterns of the crossing flows. The problem is that the different collective patterns of motion destroy each other after a short time, such that none of them is stable. Putting a column in the center can increase the likelihood of rotary flows and thereby increase the overall efficiency. But a further improvement can be reached by replacing an intersection of four flow directions by four intersections of two flow directions each, as it may be reached by suitably located railings. In this way, a rotary flow pattern is supported, and disturbances can be drastically reduced.

Crowd disasters


Unfortunately, pedestrian flows don't always self-organize in an efficient way. Sometimes, terrible crowd disasters happen, and dozens or hundreds of people may die, even though everyone has peaceful intentions and does not behave in a ruthless or otherwise improper way. How is this possible?

When I got interested in the problem in 1999, crowd disasters were often treated as God-given or natural disasters that are beyond human control. But the root cause for the breakdown of social order in pedestrian crowds has something in common with the occurrence of traffic jams. If the density gets too high, pedestrian flows turn unstable. The resulting crowd dynamics can be uncontrollable for individual people, and even for hundreds of security forces. But I will show below that crowd disasters can nevertheless be avoided, when their reasons are understood and when proper preparations are undertaken.

Crowd disasters have happened since at least Roman times. That’s why building codes were developed for stadiums, as exemplified by the Coliseum in Rome. The Coliseum had 76 numbered entrances and could accommodate between 50,000 and 73,000 visitors, who would exit through the same gate through which they had entered. With these rules and its generous provision of exits, the Coliseum could be evacuated within 5 minutes. Modern stadiums, which generally have a smaller number of exits, can rarely match this figure.

Despite the frequent and tragic occurrences of crowd disasters in the past, they still continue to happen. In other words, they are not properly understood. Media reports often suggest that crowd disasters occur when a crowd panics, causing a stampede in which people are crushed or trampled. The implication is that crowd disasters would be the result of unreasonable or aggressive behavior, with some individuals pushing others relentlessly as they try to escape. But why would people panic? Referring to crowd disasters, Keith Still once said to me: "People don't die because they panic, they panic because they die."

In fact, in my studies with Illes Farkas, Tamas Vicsek, Anders Johansson, Wenjian Yu and others, we revealed that many crowd disasters have physical rather than psychological causes. They may occur even if everybody behaves reasonably and tries not to harm anyone else. Therefore, the view that crowd disasters are mostly a result of panic is outdated. Alternatively, one might suspect that people are crushed when the inflow into a spatially constrained area exceeds the outflow for an extended period of time. Certainly, the density can become life-threatening under such conditions, as more and more people accumulate in too little space. But when in 2006 another crowd disaster happened during the Hajj – the annual Muslim pilgrimage around Mecca – it occurred on a large plaza.

Being experts in crowd dynamics, Anders Johansson and I were asked to evaluate videos showing the accident area. In the beginning, when we played the videos, we saw nothing informative. Due to the high pedestrian density, people just moved very slowly, some centimeters per second. However, when I asked Anders to play the videos 10 or even 100 times faster, we were totally surprised!

The accelerated videos showed some striking phenomena. First we discovered an unexpected, sudden transition from smooth pedestrian flows to stop-and-go flows (see the long-term photograph above and the Video. In contrast to freeway traffic, however, these stop-and-go flows were previously unknown and unlikely to result from delayed adaptation. We discovered that they were caused by a competition of too many pedestrians for too few gaps in the crowd, i.e. by a coordination problem. The stop-and-go movement set in. When the overall flow suddenly dropped as a critical pedestrian density was crossed. As a consequence of the drop, the outflow from the area was drastically reduced, while the inflow stayed the same. So, the density increased quickly, but this was not the final cause of the tragedy!

To our further surprise, some minutes later we witnessed another unexpected transition – from stop-and-go flows to a phenomenon we call "crowd turbulence" (see picture above and the Video. In this situation, people were pushed around in random ways. Eventually, Anders Johansson and I discovered in 2006/07 that it was not the density, but the density times the variability of speeds, the so-called "crowd pressure," which determined the onset and location of the crowd disaster.

It turned out that, when the density crosses a critical threshold, any body movements – even unintentional ones – can create forces acting on another pedestrian body. These forces can add up from one body to the next, such that the resulting force quickly changes the strength and direction. Therefore, people find themselves pushed around in unpredictable and uncontrollable ways.

It is just a matter of time until someone loses the balance, stumbles, and falls to the ground. This produces a "hole" in the crowd, such that the forces acting on the surrounding people get unbalanced, as the counter-force from the front is missing. Therefore, the surrounding people tend to fall on top of previously fallen persons or are forced to step on them. The situation ends with many people piled up on top of each other, such that those on the ground have difficulties to breathe and die of suffocation. Similar observations were made in other crowd disasters, for example, the Love Parade disaster in Duisburg, Germany.

Countering crowd disasters


Can we use the above knowledge to avoid crowd disasters in the future? Yes, indeed! Some years back, together with several colleagues, I became involved in a project aiming to improve pedestrian flows during the annual Muslim pilgrimage to Mecca. We were asked to find a better way of organizing the crowd movements around the New Jamarat Bridge, a focal point of the pilgrim route See more. On and around the previous Jamarat Bridge, thousands of pilgrims had died in tragic crowd disasters over the years. How could one avoid them?

This was a challenge that required us to take into account not just technical matters such as crowd densities, but also dozens of religious, political, historical, cultural, financial, and ethical factors. Our use of crowd modeling led us to propose measures including the counting and monitoring of crowds through newly developed video analysis tools,[4] the implementation of time schedules for pilgrim groups, re-routing strategies for crowded situations, contingency plans for possible incidents, an awareness program to inform pilgrims in advance about the procedures during the Hajj, and an improved information system that had to guide millions of pilgrims speaking about 200 different languages. After implementing these proposals, the 2007 Hajj (in 1427H) was indeed safely performed.[5] Like for the traffic assistant system discussed before, the main underlying success principle was to gather real-time information and respond to it adaptively.

Since then, the principle of providing real-time feedback has widely spread. An interesting example for this is crowd sensing. Paul Lukowicz, a member of the FuturICT project, and a number of further scientists such as Martin Wirz and Ulf Blanke, recently developed an app for safer mass events. In a number of festivals in London, Vienna and Zurich, they used voluntarily provided GPS traces of visitors to determine the crowd densities and pedestrian flows, i.e. averaged quantities determined from the GPS data of many people. These data were then provided to the visitors of the mass events, to give them a better idea of over-crowded areas that they should better avoid.

Forces describing opinion formation and other behaviors


Of course, one might ask whether the concept of social forces can be also used to understand different social phenomena and to overcome other kinds of problems, too, such as crime or conflict. I am convinced of this! The success of force models in describing pedestrian flows is related to the fact that pedestrians are moving continuously in space. Therefore, the dynamics of a pedestrian can be represented by an equation of motion, which says that the change of its spatial position with time is given by the velocity. Complementary to this, the change of velocity with time, i.e. the acceleration, is modeled by a sum of forces. But can we understand opinion formation processes or other behavioral changes by social forces as well? Surprisingly, the answer is "yes," if we have more or less gradual changes on a continuous opinion scale or in a continuous behavioral space. Otherwise, one must use generalized models, which exist as well.

After formulating the social force concept for pedestrians in Göttingen, Germany, in 1990 I joined the team of Professor Wolfgang Weidlich at the University of Stuttgart. He was probably the only physicists at this time working on modeling socio-economic processes and systems. In some sense, Professor Weidlich might be seen as grandfather of Sociophysics. My plan at this time was to learn, how opinion formation and decision making could be modeled. Since my work on pedestrians, I had the idea that both, the individual and collective behavior of people could be understood through social forces, and I managed to formulate a corresponding theory (see Information Box 1).

Interestingly, social force models can be formulated for migration processes, too, when people are assumed to relocate within a certain (not too large) radius. In one of my models formulated in 2009, I assume success-driven migration, where individuals try to avoid locations, in which they expect bad outcomes, but seek locations that appear to be favorable. Bad neighborhoods (those, where people were uncooperative) turn out to have a repulsive effect, while good neighborhoods (where people were cooperative) have an attractive effect. It is even possible to calculate the direction and strength of the repulsion or attraction effect, i.e. the force describing the average direction and speed of motion in a certain location.

A great advantage of using the concept of "social forces" is that it can help us to get a better imagination of complex processes underlying social change. Movements towards some subject or object are reflected by attractive forces, movements away by repulsive forces. It is also important to recognize that such forces may not be attributed to individuals, but rather to groups of individuals, companies, or institutions. In other words, social forces may be a collective effect. Group dynamics, or "group think" as a result of some emergent group identity, is probably a good example for this. There, the interaction of individuals creates a collective "group" perspective, which in turn changes the behavior of individuals. In fact, the theory of social milieus knows that the behavior of an individual is largely influenced by its environment. This can now increasingly be quantified and put into mathematical formulas with predictive power. But what is more powerful: physical or social forces?

Culture: More persistent than steel


It has often been claimed that war is the mother of civilization, and whoever has better weapons will rule the world. However, I don’t buy this. Even though war may have spread civilization, I believe the underlying mechanism is migration and the exchange of goods and ideas. Today, with the Internet, civilization can spread in ways that does not have to cost human lives.

But what is the basis of civilizations? It’s culture! Culture is largely a collection of rule sets, such as procedures and social conventions, norms, values and roles. These determine the success and failure of societies and guide their evolution over sometimes thousands of years. Just take religious values, which can determine the behaviors of people over thousands of years. It is therefore not exaggerated to say that culture is more persistent than steel. And culture is more relevant for success than weapons. In other words, social forces can be stronger than physical ones. A good example is the ancient Greek culture, which managed to spread to their Roman occupants, since it was more advanced.

However, while we all learned about physical forces at school, only very few people have an explicit knowledge of the social forces determining the behavior of socio-economic systems. This has to change, if we want to overcome or at least mitigate socio-economic problems. As the last chapter has shown, it's now possible to identify the interaction mechanisms that promote social order. Information Box 2 on social capital further illustrates that the success and failure of societies largely depends on invisible factors. In a sense, social norms are the fabric of our society, and social capital acts like a catalyst of socio-economic success.

Avoiding conflicts


Conflicts, wars and revolutions, too, can be understood with a social force approach. They relate to forces that destabilize a system and may break it into pieces. There are at least three types of conflict situations: (1) An encounter (say, between two countries) causes losses on both sides. This might be avoided by better advance awareness of the likely outcomes of such an encounter. (2) The encounter is beneficial for one side and unfavorable for the other, while it causes an overall damage. Here, the second party needs to be protected from exploitation (e.g. by solidarity from third parties or by establishing an efficient separation of the conflict-prone parties). (3) The encounter is advantageous for one side and undesirable for the other, but this time the overall outcome is positive. Then, value exchange can make the interaction beneficial for both sides, i.e. it's possible to align the interests and create a win-win situation. Recently, I have proposed Social Information Technologies that aim to reduce the occurrence of conflicts between companies or people.

Would it also be possible to measure the forces creating conflict? I think so. We could build a ConflictMap, revealing regional and international tensions and how they come about. In fact, when working in my team, Thomas Chadefaux mined millions of news articles over a period of more than 100 years and performed a sentiment analysis for words indicating conflict. This allowed him to quantify the level of tension between countries in the world. Moreover, he could show that the level of tension allows one to predict the likelihood of war outbreaks in the next six or twelve months. Such advance warning signals can provide valuable time for diplomatic efforts to reduce political tensions before it's too late for a peaceful solution. Our analyses also revealed how tension spreads from one country to the next, as it happened after the war on Iraq, thereby destabilizing the entire region. Apparently, this has produced fertile ground for the rise of the Islamic State (IS).

Conflict in the Middle East


Another data-driven study analyzed a problem that worries the world since many decades, namely, the conflict in the Middle East. Why haven't we so far been able to stop this conflict? A classical Big Data approach, even if we knew all the trajectories of all bullets shot, couldn't really reveal the causal interdependencies underlying the conflict. Therefore, in a study with Ravi Bhavnani, Dan Miodownik, and Maayan Mor, Karsten Donnay, we developed instead an empirically grounded agent-based model. The validation procedure of our model suggests that intercultural distance is the main driving force of the conflict.

An analysis of the violent events reveals that they are correlated with each other. There is rather a responsive dynamics, where each side "pays back" for the previous attacks from the other side (see Video 1 and Video 2). For example, Palestinians retaliate violence on the Israeli side and vice versa. What does this tell us? Basically, both sides punish each other for the violence they were suffering from before. From a rational choice point of view, this should stop the chain of violence, as one event triggers another, usually bigger one, or even several ones. Therefore, the conflict is costly for both sides, and increasingly so over time. The Israeli movie "Gatekeepers," which interviews previous chiefs of secret service, therefore, comes to a remarkable conclusion: "We have won every battle, but we are losing the war." In other words: it does not pay off to be violent – on the contrary. It seems that each party tries to send a message to the other one: "Stop being violent to us – you will otherwise have to pay a high price!"

So, why does such counter-violence cause escalation rather than stopping the chain of violence? Because both sides think they are right in what they do. In fact, they apply the right principles, but to the wrong situation. It is very important to recognize that the reason that makes us punish others is related to the way we use to establish social order.

We have learned that punishment is a mechanism that can establish and stabilize social norms. Therefore, we punish those who do not follow our norms. However, such punishment is only effective, if the punished side accepts the punishment. Otherwise it will strike back and pay revenge, which gives rise to an escalating conflict. It is, therefore, important to recognize that punishment will only be effective, if people share the same values, norms, and culture.

Therefore, in a multi-cultural society, punishment may not be effective in creating social order. Under such conditions, a possible way to reduce the level of conflict would be to separate the opposing parties, i.e. to live in different areas. Another one is to develop a culture of tolerance, understanding and respect. In fact, as we have seen in the last chapter, there are many social mechanisms that support the creation of social order, for example, reputation mechanisms. I am, therefore, confident that the deeper understanding of the mechanisms and forces producing conflicts will eventually allow us to overcome or mitigate them. Personally, I would strongly advice to go away from a punitive culture and to engage in a differentiated, reputation-based culture appreciating diversity. This means, almost all of us would have to change the way we are treating others. It would require a global change of culture. But the Internet may (help to) bring it on the way.

Flu prediction, better than Google


Not just wars, but pandemics too are a major threat to humanity. Some of them have killed millions of people. The Spanish flu in 1918 was a shocking example of this. Such pandemics are, in fact, expected to happen again, as viruses mutate all the time, finding our immune systems more or less (un)prepared. The world has also been surprised by the recent outbreak of Ebola.

To contain epidemic spreading, the World Health Organization (WHO) is continuously monitoring emerging diseases. It takes about 2 weeks to collect the data from all the hospitals of the world, such that one typically gets an overview of the actual situation with a two weeks delay. Then, Google Flu Trends invented an approach called "nowcasting," which was celebrated as major success of Big Data analytics. It was claimed that it was possible to estimate the number of infections in real-time, based on the search queries of Google users. The underlying idea was that queries such as "I have a headache" or "I don't feel well" or "I have a fever" and so on might be indicative of having the flu. Recently, however, the Google Flu approach was found to be unreliable, partly because of Google steadily changes its search algorithms and also because advertisements may bias people's behaviors. Fortunately, there is a model-based approach using much less data, which considers the mechanism of disease spreading. It looks at infection data in a way that is augmented by a model based on air travel data. How did Dirk Brockmann and I discover this approach in 2012/13?

Independently of each other, we had been interested in modeling epidemic spreading processes already for a couple of years. In 2002, in the wake of the September 11 attacks in 2001, there were fears of bioterror using anthrax or other deadly germs, which threatened the USA and the rest of the world. At this time, I was proposing to Otto Schily, the then German Minister of Internal Affairs, to build a self-calibrating epidemic simulator to predict the spreading of pandemics. While infection and recovery rates are often not well known after a disease outbreak, the idea was that a self-adaptive calibration model would produce increasingly accurate predictions, as more data of infected people would become available. However, I received a letter that such an approach wouldn't be feasible. I did not really believe this, but it delayed progress for an entire decade, because I did not have any funding for such a study.

Dirk Brockmann, however, started to investigate the spatio-temporal spread of diseases by means of computer simulations, and also by analyzing the paths of dollar bills in his famous "Where is George?" study . However, when visualizing spatio-temporal spreading patterns of epidemics, the patterns looked frustratingly chaotic and unpredictable. The relationship between the arrival time of a new disease as a function of the distance from its origin location was so scattered that one could not make much sense of it. But it became increasingly clear that this problem resulted from the high volumes of air passenger travels. So, Dirk had the idea to define an effective distance based on the travel volumes between all airports in the world and to study the spreading dynamics as a function of this alternative distance measure.

Our collaboration finally emerged in 2011, when Germany was shocked to see the spreading of the deadly, food-borne EHEC epidemic. I got in touch with Dirk, because I thought we could combine his epidemic spreading model with a model of food supply chains and, thereby, help to identify the origin location of the disease, which was unknown at that time. Unfortunately, we could not get hold of proper supply chain data at that time. But our discussion triggered a number of important ideas. Particularly, our attention moved from predicting the spreading of diseases to detecting their origin locations.

In fact, looking at the empirical infection cases in an effective distance representation from the perspective of all airports in the world, we found that the most circular spreading pattern identified the most likely origin of the disease. But what is more important: once the origin location is known, combined with the circular spreading pattern in the effective distance representation, it becomes possible to predict the order, in which cities will be hit by a pandemic wave. See Video. Furthermore, it turns out that this technique can be successfully applied even if the epidemic spreading parameters such as the infectiousness and recovery rate are not well-known, which is typical the case after the outbreak of a new disease. The only data important for our analysis are the air travel volumes between all airports, which are needed to specify the effective distance.

Shortly later, Ebola broke out and, using our previously developed method, Dirk made early predictions of Ebola imports into other countries, which became the basis of international preparations to contain its spreading. See . I would also like to mention the team of Alessandro Vespignani and Vittoria Colizza, both partners of the FuturICT initiative, who have built a sophisticated simulator to predict diseases and test the effectiveness of political measures. See 

INFORMATION BOX 1: Social Fields and Social Forces


When I worked on the social force model, I soon discovered the book by Kurt Levin (1890-1947) on the social field concept. I like his idea a lot, even though a behavioral and theoretical foundation of the concept was missing. So I decided to develop such a foundation in my PhD thesis in 1992. This resulted in the derivation of Boltzmann-like and Boltzmann-Fokker-Planck equations from behavioral assumptions. 

These equations contain a quantity determining the systematic motion in the behavioral space, which can be interpreted as "social force" and often expressed as the slope of a "social field." Such a social field can be imagined like a mountain chain in behavioral space, where the steepest slope in a location determines the social force that a person with the corresponding behavior would feel. This social force describes the expected size and direction of the behavioral change. Valleys of the social field correspond to social norms. If complying with the norm, the social force is zero. But when deviating from a social norm, one will feel a social force, as it happens in reality.

Note, however, that the above discussed "mountain chain" and, with it, the corresponding social field is variable. It changes depending on the behavioral changes of others. Therefore, the social field influences the behavior of a person, but at the same time, it is potentially modified by that person's behavior and the behaviors of others. In other words, social norms may change over time as a result of social interactions.

INFORMATION BOX 2: Social Capital


Most of us have probably learned that money makes the world go round and all that matters is to have enough of it. Money is certainly a powerful invention, but there is more that contributes to economic development. This includes human capital (like education), but also social capital. 

So, what is social capital? I define it as everything that results from social network interactions and can potentially be turned into a benefit. Examples are cooperativeness, public safety, and culture of punctuality, reputation, trust, respect, and power. While our own actions influence our social capital, we can't fully control it, which is in contrast to money. In many cases, we can't buy social capital (or only to a limited extent), but social capital creates value added. Interestingly, by doing certain things, we are not automatically entitled to get a certain amount of social capital. As we know from reputation and respect, these things are given to us by others. They depend on interaction effects. 

Note that the amount of social capital also determines the resilience of a system, and its risk of failure. Social capital influences both, the probability and size of damage. This became clear to me in a seminar of ETH Zurich’s Risk Center. The Risk Center brings together experts in probability theory with experts in complexity and network theory. We discussed that large disasters have an over-proportional impact on public opinion. That's why plane crashes and terror attacks matter a lot to people, while they seem to feel less threatened by everyday risks such as car accidents or deaths of smokers. Therefore, it is often believed that "size matters," i.e. large disasters make people respond in an irrational, perhaps even panic way. 

However, having studied the phenomenon of panic for some time, I rather concluded it was more likely that "people respond to the fact that there has been more damage than just the physical one – namely damage to the social capital." For example, a large-scale disaster often damages the trust into the risk management of companies or public authorities, in particular when it was caused by unprofessional behavior or corruption. While people care about such things, no insurance company is covering this damage to social capital. Hence, we must protect social capital in a similar way as we protect economic capital or our environment. Social capital can be damaged and exploited, but this should be prevented. In order to get there, we must learn to measure social capital and to quantify its value. Quantifying the value of our environment also helped eventually to protect it better.

Trust and power


To stress the importance of social capital, it is important to recall that the financial crises resulted from a loss of trust: banks did not trust other banks anymore and did not want to lend their money; customers did not trust their banks anymore and emptied their bank accounts; banks did not want to give loans to companies anymore; people did not want to invest in financial derivatives anymore, etc. In the end, the resulting financial meltdown amounted to an estimated 20,000 billion US dollars. So, trust has a pretty high value, and when it gets lost, the economic losses are tremendous. To give another example: the recent loss of trust into US cloud storage companies after the NSA scandal was estimated to cause an economic loss of up to one third of the previous business volume.

Trust is also the basis of power and legitimacy. When I studied in Göttingen in Germany, one day a deadly car accident caused by a mistake of the police triggered a large public outcry and massive demonstrations. This was the first instance, when I noticed that public institutions can easily lose their public support, in other words: their social capital. This happens, if trust gets lost over something that the authorities should not have done according to the moral beliefs of the public. I made the same observation in Zurich, Switzerland, when there were many complaints about the work of the migration office. During this time, the windows of the migration office were broken time and again, but when the office director was replaced, the problem disappeared.

Interestingly, soft factors such as credibility and trust are the basis of power. For example, the loss of control during the England riots in 2011 occurred, after the London police shot a person without giving a sufficient justification to the public. A similar thing happened 2014 in the US city of Ferguson, Missouri. In fact, riots in many other countries, too, were triggered by events where public authorities hadn't done a proper job in the eyes of the people. The Arab spring, for example, started in Tunisia, after Mohamed Bouazizi burnt himself, because of police corruption and bad treatment.

In other words: legitimacy and power result from doing the right thing in the eyes of the people. When people don't offer their idealistic or practical support anymore, authority and power are gone. While one may buy weapons and, with this, destructive power, constructive power depends on the trust and support of the people – otherwise they won't provide support, and this basically means that one hasn’t got any power. Brutality does not create respect. It might create fear, but this can replace legitimacy only up to a certain degree. As the situation gets increasingly unacceptable, more and more people will lose their fear and start to resist the previously respected authorities actively or passively, or even become ready to sacrifice their lives. The problem of such extremism or even terrorism is well-known from freedom fighters, who have previously led normal family lives.

But even passiveness of its citizens can make a country fail within just a few years or decades. This could, for example, be seen in the former German Democratic Republic. I, therefore, believe that trust is the only sustainable basis of power and social order. It must, therefore, concern us all that, in many countries, politics and management are currently the professions with the lowest levels of reputation. In contrast, "social" professions that create public goods – firefighters, scientists, doctors, nurses and teachers – earn the highest levels of reputation. As we will see later, this has some important implications for the future.

[4] For example, one may play back videos of security cameras in an accelerated fashion, say, every 60 seconds, which allows the brain to notice advance warning signals of critical crowd conditions such as stop-and-go waves. Video post-processing can overlay colors representing the local density, or arrows representing the average flow in a certain location – important information that the naked eye cannot see.
[5] and, in fact, in the following years as well. However, as I moved to another university, where I focused on new tasks, I haven't been involved in the changes that have been made since 2007.