New research suggests that AI may not uplift everyone equally, pointing to a growing divide in how much people gain from working with advanced models.
Researchers from the University of Boston and University College London measured how humans perform alone versus when paired with different AI models, using thousands of test questions across domains like physics, math and moral reasoning.
They applied a two-stage Bayesian framework based on Item Response Theory to isolate each user’s solo ability, collaborative ability with AI and the difficulty of each task.
“By estimating the impact of AI on individual users while controlling for differences in task difficulty, user ability, and shrinking noisy estimates, our framework offers a deeper understanding of human-AI collaboration: it allows us to quantify human–AI synergy as the incremental value AI provides to individual users.”
The researchers find that people get very different boosts from AI, even when using the same model. Some users see their performance jump dramatically when they work with an AI partner, while others improve only a little.
On average, AI lifts everyone, but the gains are uneven. The study shows that weaker models still raise human performance a lot, and stronger models like GPT 4o raise it even more, shrinking the gap between human and machine when they work as a team.
They also find that the size of the boost depends on skills that have nothing to do with traditional intelligence. Users who can anticipate how the AI will respond, check its reasoning and adjust their approach mid-conversation tend to benefit far more. These traits act like collaboration multipliers, allowing some people to unlock much higher levels of synergy than others.
“We demonstrate that collaboration ability is distinct from individual problem-solving ability. Users better able to infer and adapt to others’ perspectives achieve superior collaborative performance with AI—but not when working alone.”
The team concludes that AI is not a uniform upgrade. It is a force amplifier, and how much it amplifies depends heavily on the user’s ability to interact with it as a thinking partner rather than a static tool.
“The presence of strong human-AI synergy demonstrates that response quality is not an inherent property of the model alone, but emerges from the interaction between human reasoning and AI capabilities.”
Disclaimer: Opinions expressed at CapitalAI Daily are not investment advice. Investors should do their own due diligence before making any decisions involving securities, cryptocurrencies, or digital assets. Your transfers and trades are at your own risk, and any losses you may incur are your responsibility. CapitalAI Daily does not recommend the buying or selling of any assets, nor is CapitalAI Daily an investment advisor. See our Editorial Standards and Terms of Use.