In recent years, investments have transitioned towards genuine machine learning, which enables artificial intelligence systems to swiftly analyze vast amounts of data while enhancing their performance through this analysis. The New York-based firm Rebellion Research, established by the grandson of baseball Hall of Famer Hank Greenberg, utilizes a method known as Bayesian networks. This approach involves a select number of machines to forecast market movements and identify specific trades. At the same time, companies like Aidyia and Sentient are leveraging AI technologies that operate across hundreds or even thousands of machines, employing methods inspired by evolutionary biology alongside deep learning capabilities. This technology is now commonplace in image recognition, speech identification, and various other functionalities in major tech companies such as Google and Microsoft.
The goal of these systems is to autonomously detect market fluctuations and adapt in ways that traditional quantitative models cannot replicate. Ben Carlson, author of A Wealth of Common Sense: Why Simplicity Trumps Complexity in Any Investment Plan, notes, "They’re trying to see things before they develop," referencing his decade-long experience with an endowment fund that diversified across numerous asset managers.
It’s important to differentiate AI-driven fund management from high-frequency trading, as the former focuses on identifying optimal trades over a longer timeline—ranging from hours to weeks—rather than executing rapid trades for immediate profit. More importantly, the strategies are generated by machines, not humans.
Evolving Intelligence
While the company has not actively promoted its fund, Sentient’s CEO Antoine Blondeau reveals that official trades have been made using funds from private investors since last year, following an extended test trading phase. A report from Bloomberg highlighted the organization’s collaboration with JP Morgan Chase’s hedge fund division to develop AI trading technologies. However, details about these partnerships remain undisclosed, with Blondeau emphasizing that their fund is fully managed by artificial intelligence.
According to Babak Hodjat, the chief science officer and a key figure in the creation of Siri prior to its acquisition by Apple, the system allows for adjustments to risk parameters but largely operates independently of human intervention. “It creates its own strategies and instructs us,” Hodjat explains. “It provides commands like: ‘Purchase this volume now, utilizing this instrument and order type.’ It also indicates when to exit positions or modify exposure.”
Hodjat claims that the system harnesses surplus computational power from "millions" of processors located in data centers, internet cafes, and gaming centers across various regions, including Asia. Its software framework is driven by evolutionary computation—a principle also at the core of Aidyia’s strategy.
Essentially, this process generates a diverse group of digital stock traders, tests their efficacy against historical data, and iterates upon the most successful ones. This cycle of selection and evolution continues until the system develops a capable digital trader. “Over many generations, trillions of virtual beings compete, leading to a population of adept traders ready for deployment,” Blondeau states.
Deep Investing
While evolutionary computation currently underpins the system, Hodjat sees potential in enhancing deep learning algorithms, which have shown significant capabilities in tasks like image recognition and natural language processing. Just as deep learning can identify specific traits in a photo of a cat, it could similarly discover the characteristics of stocks that yield profits.
Goertzel, who also leads the OpenCog Foundation—a venture aimed at establishing an open-source framework for general artificial intelligence—holds a differing view. He believes that deep learning algorithms have become too mainstream. “If a technique is widely adopted, its insights will likely be reflected in market pricing,” he argues. “To succeed, you need an unusual approach.” He also asserts that while deep learning excels at analyzing data with identifiable patterns—such as images and language—these patterns may not exist in financial markets or, if they do, might not provide a significant advantage since they could be readily discovered by others.
However, Hodjat remains focused on advancing deep learning technologies. This might include combining it with evolutionary computation, a method he refers to as neuroevolution. “You can evolve the weights applied to the deep learner, but you can also enhance the structure of the deep learning model itself.” Companies like Microsoft are already pursuing deep learning systems inspired by natural selection, even if they aren’t explicitly using evolutionary computation.
Pricing in AI
Regardless of the techniques employed, skepticism persists regarding the potential for AI to thrive on Wall Street. There is a concern that if one fund achieves success through AI, it could lead to replication of that system by others, ultimately diminishing its advantages. Carlson expresses doubt: “If someone discovers a profitable strategy, it’s likely that other funds will adopt it, leading to the capital inflow that complicates envisioning sustained success without arbitrage rendering it ineffective.”
Goertzel acknowledges this risk, which is why Aidyia employs a comprehensive array of technologies beyond just evolutionary computation. If others start to mimic its approaches, he argues, the company will pivot to integrate additional forms of machine learning. The objective is to pursue strategies that no other human or machine currently employs. “In finance, it’s not just about intelligence,” Goertzel concludes, “but about being uniquely intelligent compared to others.”