(ORDO NEWS) — Neural networks learn from people to paint pictures and compose music, and it still looks like a fun machine learning trick. But this is just a prelude.
It will end when neural networks begin to learn how people take risks, make choices, and how they use morality. Such studies are already underway, and we create the training data ourselves, often without even knowing it.
The program, created by researchers at the University of Toronto, Cornell University and Microsoft Research, guesses who played a set of chess games.
It calculates the author by moves and can distinguish him from the games of thousands of other chess players who regularly play on the popular Lichess server . In essence, it determines the player’s inherent decision-making style.
Chess fans have long known that grandmasters have their own recognizable play style. Someone plays aggressively and is not afraid to take risks, while someone is cautious and waits for the opponent’s mistakes.
There are those who are strong in openings, while others, on the contrary, are especially dangerous in endgames, when there are already few pieces on the board.
In a word, every chess player is unique, and there is something in his choice of moves that distinguishes him from all others. It is as unique as a fingerprint, a kind of “imprint” of style.
This is exactly what the program catches, only it does not care who makes the moves, a master or a novice amateur. She easily recognizes everyone.
The decisive contribution here belongs to machine learning – the authors took the records of the games of players who played at least a thousand times on Lichess, and selected sequences of up to 32 moves from these games.
Each move – a change of position – they encoded in the form of numbers and transmitted to the neural network, and she represented any game as a point in multidimensional space.
For a neural network, all games of a chess player are a cluster of dots (or a cluster). She was taught to maximize the density of each player’s cluster and the distance between clusters of different players.
So the neural network has learned to distinguish people by the way the moves of their parties converge into clusters.
This cluster is the individual style of a particular player, which is not always explicitly expressed, but the machine sees it. Moreover, it distinguishes players with a high rating, even if it is trained only on amateur games, and vice versa. The program really captures individuality.
The authors of the study believe that the same can be done with poker. Or, they say, given the right data, such a program could identify people by the way they drive a car or the time and place they use their cell phone.
In a word, instead of a set of moves in chess, there can be any digital traces. Any sufficiently long (digitized) history of behavior potentially contains data for training such programs.
A person is recognized by characteristic chains of actions simply because we are different, and each of us, even in small things, is somehow different from the others.
And if earlier, in order to confuse the traces, it was possible to try to distort the handwriting or voice, then changing the style of decision-making is much more difficult, it’s like replacing your psyche. Moreover, it is not known in advance what signs the network highlights and what exactly needs to be masked.
The authors of the program are concerned that their approach is suitable not only for chess, but is also easily transferred to other areas: a neural network can be trained on any available data, and not everything is going smoothly with ethics.
After all, not only scammers, but also ordinary people often want to remain anonymous, and not necessarily with malicious intent. Machine learning will make them visible.
In theory, this means that entering the network under someone else’s IP will no longer help – any person can be calculated by their unique style, no matter what it is expressed in.
True, this is still only in theory. In practice, everything is not so simple: to train artificial intelligence, you first need to collect labeled data, that is, separately record the digital traces of each of the many millions of people present on the Internet, and preferably within months.
And then constantly track them over the network. This requires serious computing power, and they, in turn, require additional energy.
Finally, people walk around different sites, and it is possible to link their stories only if these sites actively exchange data with each other, which is hardly feasible (with the exception of a number of exceptions – sites that are part of large corporations, like Instagram and Facebook, for example).
Large sites with a huge audience can provide themselves with such data. They will mainly collect digital footprints of visitors, but not primarily to reveal their identities: it is much more promising to use these footprints to study and predict behavior (for example, for marketing reasons). Social networks are a suitable testing ground for this. But the best is massive online games.
The game effectively reveals the properties of the psyche. During the course of the game, people make many decisions and interact with other players in a complex and rapidly changing situation. They have to think tactically and strategically.
They have to learn and gain experience. In some games, people play for years, which means they accumulate a rich history of their actions. More importantly, these games are played by millions of users. All this creates huge amounts of statistics, it is more than enough for machine learning.
Those who started playing as a teenager can continue many years later, evolving along with the gaming universe. During this time, their unique decision-making style will be deeply studied and defined, and such information can later become very valuable.
Sometimes former teenagers become business leaders, high-ranking officials, politicians, high-ranking military leaders. A machine trained on a large amount of data will not only know how they think and act, it will help to make predictions about them.
Of course, the accuracy of the forecast also depends on whether people retain their decision-making style over decades.
This question is difficult to answer definitively, but longitudinal studies show that core personality traits are fairly stable from adolescence through adulthood. If a young girl is prone to reflection, she will delve into herself even in old age. If a young man is overly impressionable, then with age he will not lose this property.
Nuances can be smoothed out or developed, but the core of the psyche is difficult to change. It can be reasonably bet that the peculiarities of thinking and perception, as well as temperament, people will carry with them all their lives. And if programs learn to catch these features, it promises profound consequences.
After all, the power of neural networks is not only that they find hidden patterns in a data set, they can also reproduce these patterns.
A chess program created at the University of Toronto is able to play the way people play, predict the moves of a particular chess player, and even anticipate typical mistakes that he will make in a game. She knows what mistakes players make at different skill levels and can indicate the level at which people stop making them.
In other words, the program does not look for the best move for a given position – it suggests moves that a person would make. It models the decision-making process of chess players. This is what forecasting is.
One should not hope that the matter will be limited to the artificial environment of chess. Last year, Princeton psychologists published an article in the journal Science titled “Using Large-Scale Experiments and Machine Learning to Uncover Theories of Human Decision Making.”
The authors trained the neural network on a large database collected by various scientists over many years. It contains the results of psychological experiments on how people make risky choices, including gambling – in total more than 10,000 different situations in which the subjects made certain decisions.
It turned out that trained neural networks are able to imitate human decisions with high accuracy, and they are significantly superior to risky choice models previously proposed in psychology.
So machine learning helped psychologists create a new, more effective theory of behavior that could not be developed before.
And this is not surprising: in trying to explain the choice of people, experts put forward hypotheses and rely on their intuition, but it is limited by the experiments that the human mind can grasp.
No psychologist is able to shovel through a huge database, which contains the decisions of hundreds of thousands of participants in thousands of different situations of choice.
How about moral issues? The Moral Machine project has already collected about 40 million decisions from people in more than 200 countries.
This is the largest ever online moral dilemma experiment. Participants are asked to identify a traffic situation where a self-driving car might swerve in one direction or another. The test subject must decide whom to save and whom to sacrifice.
The picture can have different characters (for example, a man, a child, a female doctor, a dog) and different environments. Millions of unique challenges for moral choice.
Such a multidimensional space of decisions is beyond the power of the human mind. Instead, a neural network trained on this data allowed psychologists to build “an informative, interpretable psychological theory that defines a set of moral principles that underlie people’s judgments.”
They write that this theory is superior to those that have been invented before, and thanks to it they have identified three new effects.
As a result, due to machine learning, you can both look for a decision-making style and model what decision a person will make. But will it be possible to move even further – to influence his choice?
Not excluded. Last year, IBM Research AI introduced the Project Debater standalone computer system , which is capable of debating with people in real time. The creators of Project Debater even tried to make the system’s voice mechanical so that viewers would not confuse it with a human.
Large language models “understand” texts and plots better and better, distinguish causes and effects better and more successfully reason within the framework of common sense.
Recently, Google AI made another breakthrough by training the PaLM model to not only give correct answers in logic problems, but also to explain why they are correct.
She can even chew on the meaning of a joke that she sees for the first time. Judging by the pace of progress, we are on the threshold of systems capable of inventing strong arguments on any issue and topic.
People play – neural networks learn
If we combine the technologies discussed above, we will get a machine that reveals the decision-making style of a particular individual, builds probable scenarios for his future decisions and, taking into account his unique style, selects a system of arguments – seeks to convince him to think or act in the right direction.
Who might be interested in such a machine?
Personal data from major social networks has already been used to create psychological profiles and targeted political campaigning in election campaigns around the world. The Cambridge Analytica story made headlines after the 2016 US presidential election.
And while everyone publicly denounces such methods, including Facebook, whose data was borrowed, the important thing is to acknowledge that machine learning schemes work.
Online games are even better for collecting data on how people make decisions. This data is, of course, actively collected.
The actions of the players are recorded, then the programs look for patterns in the telemetry sets and build behavior models based on them. For example, using machine learning, researchers at the University of Aalen in Germany studied the behavioral data of 700,000 users across 3,300 games on the Steam platform.
First of all, the program divided the players into categories according to the style of play, then delved into the search for individual styles in the categories found.
Behavioral analysis from gaming telemetry is a growing area of research. From a game design point of view, this will help make games even more fun and at the same time increase their monetization, that is, encourage participants to spend money more willingly. To do this, it is desirable to be able to predict their behavior and characteristic decisions in different situations.
It is impressive that such analysis is already being done in real time. For example, you can calculate anomalies in the behavior of the player right during the battles in MOBA. The anomalies point to a likely scam – for an esports industry already valued at more than $2 billion, this is a serious topic.
About 3 billion people play online games on different platforms , and this is not the limit. Over time, games will become more realistic, intense, interactive and immersive. And if today the gaming environment is programmed from and to, in the future it will be able to construct neural networks right during the game, depending on the behavior of the players.
It is likely that the gaming universe will become part of a more global meta-universe that IT giants are planning to build. At least half of the world’s population will visit it regularly.
Each time, acting within the virtual world, making choices and making decisions, people will increase their individual trajectories of digital footprints.
They can build this trajectory for years – playing, having fun, working and collaborating online. Therefore, digital universes will increasingly gravitate towards the collection of personal data. Thanks to machine learning, this makes perfect sense.
From creating pictures to scaling talent
So, we are entering an era where many people, especially those who regularly play online, will have a rich and long history of their behavior, their decisions and interactions with other people and AI agents.
Advanced algorithms will allow you to deeply analyze these statistics. Programs will not only reveal a person’s unique style, but also create his model – behavioral and psychological.
The model implies prediction. She can build different scenarios and rank them by probability. The task is to guess which choice a particular person is inclined to in a given situation.
Or what mistake he is most likely to make at some point. Next, artificial intelligence, a descendant of Project Debater and large language models, will come into play, which will communicate with a person individually, taking into account his psyche and thinking style.
In general, for large masses of people, all this worked for a long time, long before the rise of machine learning. Advertising, politics, propaganda stand on this basis, but the result has always been achieved “on average”.
A feature of the new time is personal tracking. Now, in some aspects, the machine will know the person better than he himself.
All this reasoning is not here to scare you, the reader, with a dystopia, but to probe the vulnerabilities and risks of future technologies.
The creators of the chess program from the University of Toronto immediately realized the danger of their offspring, around which they had a discussion with colleagues. It’s a paradox, but they created the program not at all in order to deanonymize the players.
Their original goal was to develop intelligent systems that would be easier for people to interact with. And for this, AI must be able to behave like people and calculate their behavior. The ability to recognize a chess player by his playing style is just a side effect of such a noble mission.
In the same way, the technology of “decision-making style” recognition can bring many benefits, including in science, opening up new spaces for solving complex problems. And the key here will be the ability of generative neural networks to generate new data, inspired by learning.
A vivid example of such a skill is the artificial intelligence system DALL·E 2 from OpenAI, which can create images from text descriptions. For example, she easily copes with the task of painting “a fox sitting in a field at sunrise, in the style of Claude Monet”, and the picture looks very convincing.
DALL·E 2 accurately captures stylistic traits in human-created images and then reproduces those traits in images it creates. This allows not only to quickly get a lot of new paintings a la Monet or Warhol, but also to combine and combine styles.
And this property is decisive – it opens the way to the unknown.
Like DALL·E, neural networks of the future will be able to search for original thinking strategies by combining decision-making styles from different people.
Or apply the style of a particular person to problems that he did not encounter. Such combinations may sound crazy, like Kasparov-style trying to solve quantum gravity problems, but therein lies the possibility.
Einstein could not have been closely involved in organic chemistry, materials design, banking reform, city design and a host of other necessary things in parallel with the theory of relativity.
A century later, such a trick can work out – the neural network will borrow part of the intelligence of geniuses, learning from their trajectories. It is as if she rents their best strategies and uses them where the geniuses have not reached.
And who knows, maybe in the foreseeable future no one will be surprised by a scientific discovery or an economic miracle (or, for example, a military operation) where the style of a multiplayer champion was applied.
Any powerful tool is always double-edged. The main decision – for what purpose to use it – will still remain with the people.
Contact us: [email protected]