Swarm Intelligence and the Global Brain

In nature, when multiple individual intelligences communicate with one another, it often results in super-intelligences. The Internet allows people to be much more connected, thereby expediting the exchange of information. Companies such as Facebook and Twitter harness machine learning methodologies, and gather information from this. It is possible to predict and learn behaviours through the analysis of large volumes of data. AIs are becoming an ever more important part of our communication. This hybrid creature could become a global conscience for humankind.

If a group of people were asked to guess the number of marbles in a jar, they would give very different answers. However, by taking a mathematical average of these results, the outcome would be astonishingly close to the correct answer. The wisdom of the crowd is an old invention.

Undoubtedly, anyone who has been to Rome has seen the swarming starlings that swoop as an amoebic mass, and create fantastic patterns in the sky. Biologists have discovered that each individual bird ultimately only pays attention to a few direct neighbours and “calculates” its own behaviour on this basis. We see similar mechanics at work in schools of fish, all of which can be reduced to very simple physical forces of push and pull.

However, what is intelligent about movement, even if it is coordinated in such impressive fashion? The ultimate relevance for the individual is the knowledge of how much longer or more secure its life might be as part of the group. Even in the case of simple life forms, such as ants or bees, it is possible to observe fantastic cognitive achievements that are only possible because of the collective approach. No central authority (and certainly not the queen) is responsible for the actions of the individuals. Instead, the dynamic concatenation of interactions between creatures results in something greater than the sum (or mean) of the individual parts. Whether a colony, swarm or herd – a collective is a super-being of sorts: one that is brought to life through the self-organisation of individuals (it is often more intelligent, as well).

People can also surpass individual performance when they are part of a swarm. Swarm intelligence is the speciality of the company unanimous.ai. It tested the average accuracy of Super Bowl predictions from a “swarm” of more than 400 football fans against those of around 30 randomly selected fans. The human swarm were able to collectively “bet” on an outcome using an app. This used a virtual arena, in which each user could attract a metal ball using a virtual magnet. The potential outcomes – or which team would win the game – were positioned in the corners of the arena. Therefore, where the individual placed the magnet influenced the direction in which the metal ball moved.

Differences of opinions within the swarm meant that things became interesting. A number of essential swarm characteristics became apparent, as did differentiation from the average vote. First of all, time became a factor: throughout the interaction, the metal ball could change direction multiple times. Secondly, each individual influenced the others: with where they chose to position their magnets, participants influenced the strength of the attraction and its direction. Thirdly, there were personality differences: not everybody exhibited the same behaviour, even when they were predicting the same answer. For example, if the ball deviated from the chosen direction, some stubborn participants maintained their contrary position – even if it was obvious that they could not win. Others tried to choose a third alternative to move the ball in the direction of a compromise.

As a result of the complexity of human personalities and various soft factors, such as the motivation to “win” or the individual technical skills of the users, the swarm reached a surprising conclusion. When predicting football results, the swarm was correct in 70 percent of its predictions: 20 percent better than the method of averages, and on around 40 percent better than lone individuals.

In the age of the Internet, people can communicate much faster and with greater reach. Information flows through countless channels. Those who often take public transport can experience the complexity of the communications network at close quarters. It positively chatters, vibrates, and flashes from mobile phones; some people might use an analogue mouth-to-ear connection, whilst others might read a book or newspaper; if somebody stares out the window, their gaze is sure to be captured by the next billboard. It is impossible not to communicate.

A great deal of what our society has created is ultimately the result of complex, constantly circulating; strengthening or weakening information flows in the human communication network. The success of a new product – be it acceptance of organic waste bins or the outcome of a democratic election – is decided by the swarm.

As a result of machine learning methodologies, a better understanding of these complex processes has become possible in recent years. The two most important factors are already intuitive: on the one hand, individual differences or personality define how somebody responds to information. Secondly, it is the structure of the social network that determines how the group processes information.

Look more closely at a swarm, and each person within the network represents a computing unit that translates sensory input into output: for example, a behaviour. Such processing is very complex in humans, primarily because we have comparatively good and long memories. However, ants and bees also have brains and they do not all react in the same way, even when presented with the same impulse. Therefore, we can reduce individuals to the same stimulus-response-characteristic. The outputs of each individual might be of a communicative nature – such as a tweet or a call – or, in turn, they might be subtler, in which case the counterpart can decode them without a smartphone. These signals are then transported through the social network (analogue and digital), to represent an input component at each connected node.

A network can display very different connection patterns. In a bird swarm, the direct neighbours are typically always connected. However, in the case of humans, the communication is global and social networks are far more complex and variable than in a bird swarm. In the Internet age, a large proportion of our communications (especially long-range communications) can be registered using technology; allowing us to study the social network in (almost) real-time.

Companies of the digital economy, led by Google and Facebook, have used machine learning to create added value for the user. The primary basis for this is to “learn” our individual response characteristics from the large amount of data we generate through our behaviour in the digital world. If it is known, for example, which films a user enjoys, it is possible to predict the extent to which he or she might like another film they have not yet seen. If the prediction is correct, the user will be pleased with the outcome.

These so-called recommendation systems are based on the sensible assumption that if a similar user likes a product, you will like it as well. User similarities are determined by harnessing similarities in their product ratings. A matrix is constructed internally, which then assigns a row to each user. Columns are dedicated to a particular product, listing values that reflect user ratings. Undoubtedly, there will be a variety of reasons why user 1 enjoyed film A. It could be due to certain actors the user likes, favourite directors, particular subject, etc. Such hidden reasons are known as latent variables, and users also have latent – or invisible – traits that they share with one another, or not, as the case may be. Using non-negative matrix factorisation, an existing dataset can be harnessed to determine these latent variables. These matrices can be “learned” step-by-step, using a well-known machine learning process called gradient descent. When there is sufficient data for these optimisation processes, the matrices can be used to predict a film rating for a certain user. If the predicted rating is high, it is very likely that the user will enjoy the film and, consequently, it is recommended.

So, artificial intelligences are slowly beginning to integrate themselves into our communication network. They filter and bundle human communication, thereby shaping the flow of information through the network. AI methods can generate artificial but very realistic images or soundtracks that are capable of fooling people. And in the meantime, AIs have not just become interfaces or tools for human communication. Within the digital world, advancements in natural language processing mean they themselves can become useful communications partners (link to article).

However people also interact without apps and AIs. The swarm is very capable of incredible stupidity. Countless examples of collective behaviour – better left unmentioned – can be found on a daily basis on Twitter or Facebook. People are quick to collectively jump to conclusions, prejudices, and judgements that often turn out to be embarrassingly wrong a day later, in light of new information. Recent major democratic elections – the American presidential election and the British referendum on “Brexit” – demonstrated all too well that targeted campaigns, planned using precise target group analysis, can have a significant impact on swarm dynamics.

Regardless of whether we consider the election of Trump or the vote for Brexit to be intelligent or not, we need to increasingly look upon ourselves as a hybrid swarm. This process will become more physically apparent in a couple of years at the latest, when autonomous vehicles start sitting together in traffic alongside human drivers.

AI will increasingly be used to predict human behaviour.

Not only to better target advertisements or promote political agendas, but also to prevent accidents and save lives. And perhaps it will also be an AI that finally helps us to overcome divisive barriers such as languages and culture.

Add a comment

your browser is not up to date
to enjoy this website you will need to install a modern browser.
we recommend to update your browser and to install the latest version.

iOS users, please male sure you're running at least iOS 9.

Mozilla Firefox Google Chrome Microsoft Edge Internet Explorer