42AI: An Innovative Approach to Human Behavior Prediction Using a "World Model"

Israeli startup 42AI enters the scene with significant investment and the ambition to predict human actions through a unique behavioral "world model." Their technology aims to understand users' inner motivations and psychological tendencies.

42AI: An Innovative Approach to Human Behavior Prediction Using a "World Model"

In the era of artificial intelligence, the question arises: what if AI systems could predict our actions before we even decide, even in completely novel situations? With this ambitious vision, Israeli startup 42AI emerged from stealth, having developed a "behavioral world model." The company also announced securing $62.5 million in seed funding and is already deploying its predictive systems with selected financial brokerage firms.

At the core of 42AI's philosophy is the conviction that human behavior is not deterministic but constantly probabilistic. Even as circumstances change, people retain fundamental psychological tendencies that influence their decisions. It is these deeply rooted inclinations that the startup seeks to measure and translate into data that the model understands. The goal is thus to uncover individuals' inner motivations, rather than merely extrapolating patterns from past actions.

To achieve this, 42AI analyzes how people react to seemingly unrelated events. For an investor or trader, for example, these could be processes like identity verification (KYC), registration, customer support interactions, money transfers, or trading itself. The model tracks various data points – from sentiment and voice intonation during support calls to contact frequency and subsequent trading transactions. It strives to understand how an individual perceives the world, because for the model, each individual is their "world."

These insights are then converted into mathematical representations of characteristics, which are fed into a transformer "world model." This model then establishes probabilities of possible decisions in various situations. The company uses a separate language model to generate explanations for these predictions. During training, the model is configured to maintain a person's fundamental tendencies; for example, if a risky model suddenly started treating a cautious person as a risk-taker, it would indicate an error.

The development of this complex approach took 42AI three years and involved a combination of psychological research, neuroscience, genetics, and consulting from experts in psychology and post-traumatic stress disorder, including Nobel laureate in chemistry Dan Shechtman. The team initially studied thousands of people using psychological assessments and behavioral scenarios, to which they later added government and biobank datasets covering over 2 million people. Today, customer datasets have expanded to nearly 30 million individuals.

The system analyzes existing customer records, including transactions and support calls. The company emphasizes that the model does not require personally identifiable information and "never knows who the person is," although it creates individual behavioral profiles from linked customer activities. In real-world market trading deployments, 42AI reports 98% accuracy, which, however, rather means that trading flows fall within the predicted range, not an exact prediction of every trade. For trading direction and volume, the model achieves an F1 score of 0.8 to 0.85, a metric balancing correct predictions with false positives and missed events.

It is important to note that the model has its limitations. There is a natural "ceiling," as the system does not have access to all of a person's experiences. The system's accuracy also decreases when it lacks important situational context, for example, if someone decides to change their trading strategies based on advice from a friend or colleague.

Dominik Medal

Planning a new website or app?

Tell me what you have in mind — I will take a look and advise on the best way forward.

42AI currently collaborates with ten large brokerage firms that use its model to manage liquidity, margin requirements, and exposure hedging. The company generates annual revenues exceeding seven figures. Customers pay either a flat fee or a percentage of the value the platform helps them achieve.

Beyond finance, 42AI is exploring applications for its model in the gaming industry, fraud prevention, and cybersecurity. In the future, it aims to help robots and AI agents better understand people, although for each new domain, the model will need to be trained on specific rules and contexts.

With every advanced AI technology come potential ethical questions. If a brokerage firm can predict its customers' behavior, what prevents the misuse of this knowledge to the company's benefit at the individual's expense? 42AI's founders acknowledge this concern and refer to regulations, monitoring, and an internal ethics committee. However, they themselves declare that their technology is part of the solution and helps address these risks.

Dominik Medal

Let's talk about your project

Tell me what you need and together we will work out the best way forward.

Stop scrolling, call me

+420 735 505 585