Will AI Make Markets Less Efficient?

• Osman Ali, Global Co-Head of Quantitative Investment Strategies (QIS) at Goldman Sachs Asset Management, leads a team with a 37-year track record dating to the late 1980s that utilizes AI and machine learning to analyze data across public markets and asset classes.
• QIS employs a combination of large and small language models to perform sentiment analysis on management disclosures and market trends, a capability that has evolved from rudimentary "bag-of-word" models introduced in 2008.
• The team specifically fine-tunes smaller, context-aware language models to extract nuances from management sentiment in various languages, such as analyzing Japanese executive disclosures for risk factors previously inaccessible via traditional techniques.
• Over the past 12 months, QIS estimates that more than 50 percent of equity returns are driven by market sentiment, themes, and trends rather than fundamental business metrics alone.
• Modern quantitative investors now analyze approximately 15,000 stocks daily, leveraging a shift from "broad but shallow" data analysis to deep, high-fidelity data processing made possible by advanced technology.
• Ali asserts that while technology is democratized, maintaining a market edge requires a unique combination of proprietary data (curated by Goldman Sachs over decades), custom technology infrastructure, and human experience to ask the right questions.
• Ali views investing as a zero-sum game where the increasing complexity of markets—driven by passive investors, retail euphoria, hedgers, and AI-driven actors—creates new sources of alpha rather than eliminating inefficiencies.
• The proliferation of large language models may increase market efficiency in under-analyzed segments like small-cap and emerging market stocks by identifying mispricings at scale.
• Conversely, widespread AI adoption may create new inefficiencies through crowding effects, where identical model outputs cause investors to pile into the same securities, leading to predictable price movements and deviations from fundamental value.
• QIS is actively modeling investor psyches to predict how retail and institutional agents using these tools make decisions, specifically to identify and capitalize on the predictable reversion caused by model-induced herd behavior.
• Ali advises that future careers in investing require a hybrid skillset combining data science and technology with traditional investing experience and context, favoring organizations that treat both as critical components.
• The QIS team size remains stable at approximately 100 employees globally despite technological advances, as automation handles hard data processing while human teams manage the "intimacy" of investment decision-making.
• Hosts Alison Nathan and George Lee conclude that the current market environment suggests AI may increase alpha opportunities by creating new, model-induced inefficiencies rather than converging toward perfect market efficiency.

The episode was recorded on May 1, 2026, and includes standard Goldman Sachs disclaimers regarding forward-looking statements and non-advisory content.