Vasant Dhar
Vasant Dhar is the Howard J. Heyman Professor at NYU's Stern School of Business and a faculty member at the Center for Data Science. His research spans machine learning, predictive analytics, and the design of AI systems for decision-making in financial markets. Across three decades of work, Dhar has built a research program around the question of where machine intelligence augments human judgment and where it must defer to it. In 2024, Dhar and his team built DBOT — a large language model fine-tuned on every blog post, lecture, valuation, and book Aswath Damodaran had published — and tested whether the bot could replicate Damodaran's valuation work. The experiment's result became the founding empirical case for Damodaran's "Beat Your Bot" framework: DBOT successfully imitated Damodaran's prose but systematically failed at the framing decisions that produce credible valuations.
In the AI Story
Dhar is one of a small number of academics whose career has straddled finance and machine learning long enough to test rather than merely speculate about the relationship between the two. His earlier work on algorithmic trading and predictive systems for hedge funds gave him direct experience with the failure modes of automated decision-making in financial contexts. The DBOT experiment was, in this sense, a controlled test of how far the failure modes had retreated under modern LLM architectures.
The collaboration with Damodaran is illuminating because Damodaran's published corpus represents perhaps the largest, most internally consistent body of valuation writing by a single author. If any expert's analytical work could be successfully replicated by training on text, his could. The fact that DBOT failed at the analytical layer despite succeeding at the prose layer was, for Dhar, an empirical confirmation of a hypothesis he had long held: that the cognitive operations producing expert judgment leave traces in text but cannot be reconstructed from text alone.
Dhar's broader research program has implications beyond the DBOT experiment for how investors should evaluate AI-driven services across industries. His work on prediction machines and decision systems argues consistently that AI augmentation produces the largest gains where the underlying problem is well-specified, the feedback signal is clean, and the cost of error is bounded. Where any of these conditions fails — as they typically do in valuation — augmentation works less reliably, and the human judgment layer remains load-bearing.
Origin
Dhar joined NYU Stern in 1988 after earning his PhD in artificial intelligence from the University of Pittsburgh. He directed the Center for Business Analytics at Stern and has held visiting positions at multiple research institutions.
Key Ideas
Empirical experimentalist. Dhar's career is distinguished by testing rather than asserting claims about AI capability in financial contexts.
DBOT as controlled experiment. The construction was not commercial but diagnostic — an attempt to locate empirically the boundary between AI-replicable and human-essential cognitive work.
Augmentation theory. Dhar's research consistently argues that AI augmentation succeeds where problems are well-specified and fails where judgment is essential.
Bridge between disciplines. Dhar's career embodies the cross-domain dabbling Damodaran prescribes as a moat-building practice.
Appears in the Orange Pill Cycle
Further reading
- Vasant Dhar, Data Science and Prediction, Communications of the ACM (2013)
- Vasant Dhar, Various papers on machine learning in finance (NYU Stern, 1990s-2020s)
- Vasant Dhar, "When to Trust Robots with Decisions, and When Not To," Harvard Business Review (2016)