Nikhil Singh
Assistant Professor of Computer Science
Dartmouth College
Nikhil Singh is an assistant professor of computer science at Dartmouth College in Hanover, New Hampshire, where he directs the SAHAS Lab (Science and Art of Human-AI Systems). He and his collaborators test how AI shopping agents choose products, and found that prices, ratings and marketing nudges such as a "popular" tag sway those agents far more than they sway people.
Sending AI Agents Shopping
"The ability to make decisions autonomously under uncertainty is, for me, the hallmark of an agent," Singh told Dartmouth's computer science department in May 2026. An agent sent to find a backpack, he said, "has to translate the goal into a series of steps, find and navigate webpages, figure out how to compare products," and then pick one.
With Manuel Cherep, Pattie Maes and colleagues at the MIT Media Lab, Singh built ABxLab, a system that sits between a website and an AI agent, changes what the agent sees, such as a price, a rating or a persuasive label, and records what it buys. In A Framework for Studying AI Agent Behavior, presented at ICLR 2026, the team ran 17 LLM agents through more than 80,000 shopping experiments. Choices shifted predictably with product order, ratings, price and nudges drawn from consumer psychology, including authority cues, social proof and scarcity, and the agents were far more sensitive to those signals than people were. "It's clear to us that their decisions are biased in a systematic way that we should properly understand before we start delegating everything to them," Singh said.
Images That Persuade Machines
The group then turned to agents that read web pages as images. Using an image generation model, they searched for edits that make a picture of a person, place or product more persuasive without changing its subject, such as better lighting or added context. "We found that we can very reliably influence the choices of these agents, which means we were discovering the visual features that really strongly bias their decisions," Singh said. "And then, to our surprise, we found that visual tweaks also can work on people." The paper, Visual Persuasion: What Influences Decisions of Vision-Language Models?, was accepted at ICML 2026 and proposes ways to blunt those effects on agents.
From Berklee to Dartmouth
Singh is also a musician and composer. He earned his bachelor of music at Berklee College of Music in Boston, Massachusetts, in 2017, studying composition, jazz and computer music, and later taught electronic production and design there. He completed a master's degree and a PhD in media arts and sciences at MIT, where he held a LEGO Papert Graduate Fellowship, and interned at Netflix and the Allen Institute for AI.
He joined Dartmouth in January 2025 and launched the SAHAS Lab that September. He taught Dartmouth's first course on AI agents, co-directs its Digital Arts program, and co-organized the first Workshop on Agent Behavior at the COLM 2026 conference.