One of the biggest business headlines of 2024 was the surging growth of generative artificial intelligence across industries and sectors.
The numbers tell the story: A recent Wharton study found that only 37% of large firms used AI weekly in 2023, but that increased to 72% in 2024. Lynn Wu, a Wharton professor of operations, information and decisions, expects that upward trend will continue in 2025.
“I think this is very individual, for people and also for firms, to figure out what the best use case is going to be for you, specifically. That’s is going to be key to unlocking the benefit of AI,” she said during an interview with Wharton Business Daily on SiriusXM. (Listen to the podcast.)
Wu said the race to incorporate AI into daily operations shows no signs of slowing down. Yet the course will change as the tech evolves. That’s because it comes with some inherent challenges. First, there’s the cost.
While models such as ChatGPT and Microsoft Copilot are free to the public, Wu cautions that AI tools are “actually very, very expensive.” A basic law of supply and demand dictates that exponential growth generally makes technology cheaper as it becomes more available and competitive. But Wu doesn’t believe that will be the case with AI.
“It’s a highly concentrated industry with a few players at each level of the AI stack, so it’s never going to be that cheap because of the concentration of the industry,” she said.
The second significant challenge is the limitation of data. AI is still riding a growth curve because it’s being built on all of human knowledge. At some point, the models will exhaust the available knowledge.
“Using the current techniques, the performance gradually degrades the technology output because machine-generated data, when it’s fed into an AI algorithm, produces less good stuff than human data,” Wu said.
“It’s a highly concentrated industry with a few players at each level of the AI stack, so it’s never going to be that cheap….”— Lynn Wu
With AI, Practice Makes Perfect
Despite the dual forces of cost and data limitation pulling on AI, Wu thinks there’s still a lot of room for progress. She expects firms will spend the next year experimenting with AI to find more complementarity with humans. Firms that can figure out what tasks are best suited for employees, what’s best done by machines alone, and what needs a combination of the two will be able to deploy AI most effectively.
Of course, they still have to watch out for hallucinations — the propensity for AI to produce false information.
“One good rule of thumb is that if you’re already good at [something], AI is going to help even more. But if you’re not good already or have very rudimentary knowledge, I would use AI cautiously because it does hallucinate,” Wu said. “It will tell you things that sound very true but, in fact, are not true. And that’s where it’s difficult for somebody who does not have in-depth knowledge to disambiguate.”
Wu, whose research focuses on how technology affects innovation and the labor force, said AI offers tremendous benefits for entrepreneurs. It helps prototype products in a fraction of the time, and it provides a continuous feedback loop to improve them.
“One key thing we’ve discovered in our recent research is that AI really turbo-charges prototyping,” she said, using the example of creating an advertisement. “Before, you had to make an ad yourself, figure out the image, which actor to use. Now, these are all just generated automatically, so you can decide to have 10 different copies.”
Even in the excitement of turbo-charged innovation, Wu raises a red flag. It goes back to data limitation. About 80% of innovation is incremental, created by recombining or tweaking existing things, and that’s where AI shines, Wu said. But the other 20% is “radical innovation” that comes from developing something completely new. It is unclear what role AI will play, if any, in novel development.
“We actually need to protect that radical innovation more than before,” she said.
