Is AI Accelerating Innovation — or Just Amplifying Its Real Barriers?

Using artificial intelligence to speed up innovation inside a company can look like an appealing fix. But researchers from Harvard Business School, Wharton, Kellogg, and Columbia — Julian De Freitas, Ayelet Israeli, Gideon Nave, Artem Timoshenko, and Olivier Toubia — argue in a new paper that while AI does accelerate idea generation, it accelerates the reinforcement of familiar, outdated ideas just as fast.

The researchers say the problem isn't the technology itself: 'Most barriers to innovation aren't technological, they're human — how people anchor to familiar ideas, how expert panels converge on shared notions of what counts as promising, and how consumers can't articulate what they want until they see it.' Because AI is trained on existing human-generated content, it reproduces the very tendencies that created those barriers in the first place — only now faster and at far greater scale.

The scholars walk through how AI deepens human barriers at every stage of the innovation process — ideation, screening, consumer insight, and market research. At the ideation stage, when a language model is used without a clear structured prompt, 'the model is statistically biased toward the typical response, because it's trained to predict the most likely next words and auto-complete text,' the researchers note. 'And once a person reads that typical response, they anchor on it — which crowds out the more unusual ideas they might otherwise have generated unprompted.' At the screening stage, organizations typically rely on a staged evaluation committee to fund new ideas; because evaluators are naturally biased against risky, novel concepts, 'AI-generated proposals score higher simply because they're inherently smooth and well-structured' — regardless of how good the idea actually is.

The researchers' bottom line: 'Our response isn't to slow AI adoption, but to make a deliberate choice about which parts of the process still require direct, unmediated contact with reality.' In other words, using AI in the innovation process isn't itself the problem — the problem is failing to clearly define where and under what oversight it should be applied.

The study offers a practical lesson for managers: if a company folds AI into processes like ideation, product testing, or market research without critical oversight, it risks locking the organization into safer, more average decisions rather than accelerating innovation. The real breakthrough lies in clearly distinguishing where to apply AI and where to preserve direct contact with raw, unfiltered human judgment and the market itself.