Israeli startup 42AI says it has built an artificial intelligence system that can estimate what a person is likely to do before they do it, even in circumstances the person has never faced. The company has come out of stealth with $62.5 million in seed funding, and it is already selling the technology to financial brokerages.
What is 42AI's behavioral world model?
The company calls its product a “behavioral world model.” Its chief executive and co-founder, Amit Gal-Or, told TechCrunch that human behavior is probabilistic rather than fixed. People carry underlying psychological tendencies that shape their choices even when their circumstances change, he said, and 42AI is trying to measure those tendencies and turn them into data a model can use.
“A week from now, you might be a different person, but your tendencies will always play out in certain areas to a probability or distribution,” said Ehud Ezra, the company’s chief technology officer and co-founder. In practice, he said, the model produces the probability that a person will make a particular decision.
How does 42AI predict what people will do?
Rather than extrapolating from a person’s previous actions, 42AI says its model tries to uncover internal motivations. It does this by watching how people respond to events that seem unrelated to the decision at hand. For a trader or investor, those events can include completing know-your-customer checks, onboarding, support calls, money transfers and stock trades.
Ezra described how the process works. When a customer contacts support or sales during onboarding, the system looks at sentiment, voice and the number of interactions, including how often the person calls. Those data points feed into the model. Later, the system examines how the customer buys and sells and when, and it factors in market conditions at that point.
The company converts these observations into mathematical representations of personal traits. Those representations go into a transformer-based world model that produces probabilities for possible decisions under different conditions. A separate language model generates the explanations for its predictions.
Ezra said the model is trained to keep a person’s core tendencies intact during simulations. If someone is generally risk-averse but the model suddenly treats them as a risk-taker, he said, “the target is off.” He said that is how the training rewards are set up.
The company says it spent three years on the approach. It combined psychological research with neuroscience and genetics, and it drew on advice from experts in psychology and post-traumatic stress disorder. Dan Shechtman, who won the 2011 Nobel Prize in Chemistry, also advised the team.
Gal-Or said the team first studied thousands of people through psychological assessments and behavioral scenarios. It later added government and biobank datasets covering more than 2 million people. The company’s customer datasets have since grown to cover nearly 30 million people, he said.
In live deployments, the model analyzes existing customer records, including transactions and support calls. Gal-Or said the system does not need personally identifiable information. “The model never knows who the person is,” he said. Even so, the system builds individual behavioral profiles from linked customer activity, which raises questions about how far that distinction goes.
How accurate is 42AI's trading prediction model?
42AI claims its model reaches 98% accuracy when predicting market trades. Gal-Or clarified that the figure measures whether trading flows fall within a predicted range, not whether each individual trade was called correctly. He also said the model consistently achieves an F1 score of 0.8 to 0.85 for predicting trading direction and volume. An F1 score measures how well a classification model balances correct predictions against false alarms and missed events.
Gal-Or acknowledged the limits. “There’s a natural ceiling because you really don’t have access to everything that a person experiences,” he said. Ezra said accuracy drops when the model misses important context, such as when a trader changes strategy after advice from a friend or colleague.
Gal-Or said 42AI works with 10 large brokerages, which use the platform to manage liquidity, margin requirements and hedging exposure. He said the company is signing more customers and is generating more than “seven figures” in revenue, though he declined to give a specific figure. Customers pay either a flat fee or a percentage of the value the platform helps them gain. He said one brokerage generated about $30 million in additional value through dynamic hedging.
The $62.5 million came from the founders and Infinity Group over three years, Gal-Or said, and was not raised entirely from outside investors. Infinity was founded by his father, Amir Gal-Or, who is also 42AI’s chairman and a co-founder. The company plans to raise more money.
Beyond finance, 42AI is exploring uses for the model in gaming, fraud prevention and cybersecurity. The company eventually wants to help robots and AI agents understand people, though Ezra acknowledged that the model must be trained on the rules and context of each new domain before it can work there.
The technology also raises an obvious concern. If a brokerage can anticipate what its customers will do, what stops the brokerage from using that knowledge for its own benefit? Ezra called the concern “a valid concern.” The founders pointed to regulation, monitoring and an internal ethics committee. They did not explain what, if anything, would prevent brokerages from using the technology to gain an edge over the customers it profiles.
Gal-Or said he believes the technology can help address those risks. “We are very proud that actually we are part of the solution; we are the solution to this problem,” he said.
Questo articolo è stato scritto con l'assistenza dell'IA.
News Factory APP - notizie agentiche per potenziare il tuo SEO e AEO.