Wharton professor Ethan Mollick says companies must make organizational changes if they want to benefit from AI

• Ethan Mollick says AI requires organizational changes to work well, not just individual use.
• Mollick is a professor at UPenn's Wharton School, where he runs an AI lab.
• Mollick's research shows that AI can boost productivity but requires strategic deployment.

Wharton professor Ethan Mollick believes that companies won't see gains from AI until they make changes at an organizational level.

"Until we change the organization, we won't get much benefit," he said at the MIT AI conference on Saturday.

Mollick has become a prominent figure in the AI revolution. He's the author of Co-Intelligence, a book about living and working with AI, in which he writes that the cost of understanding the potential of AI is "three sleepless nights."

His advice for companies is to start by rethinking the way they deploy artificial intelligence. Until the recent advances in AI, he said new technological innovations were introduced into companies under the assumption that they would be controlled by human workers — with intelligence and judgment. "That starts to change in AI, it's intelligence of a different source, but it's now intelligence deployable a different way," he said.

He pointed to a recent study he co-authored that examined the impact of AI on a group of BCG consultants. The study randomly assigned 758 consultants at the firm to one of three groups: no access to AI, access to ChatGPT powered by GPT-4, and access to ChatGPT, along with instructional videos and documents on prompt engineering strategies.

The results showed that AI creates a "jagged technological frontier" in which AI easily completes some tasks while others of similar difficulty are outside its current capability.

For tasks "inside the frontier," consultants using AI were "significantly more productive." However, those using AI to complete tasks "outside the frontier" were "19 percentage points less likely to produce correct solutions compared to those without AI."

The takeaway is that AI can be incredibly helpful to workers when deployed properly. But Mollick said right now AI integration is "all happening at the individual level and organizations are basically not learning any of this for a variety of really interesting reasons."

In his Wharton classes, he requires students to use ChatGPT for assignments. "There's a lot of positives about it. That doesn't minimize the fact that cheating and negativity are there, but those have been there for a long time," he told NPR.