Ajay Agrawal

Senior VP of Global Services & Chief Business Development Officer, Carrier Global

Ajay Agrawal - AI Is Turning Everything Into a Prediction Problem

🎥 Feb 14, 2025 📺 The Lavin Agency ⏱ 1m 👁 81 views

The biggest shift in AI isn't just better predictions—it's redefining what counts as a prediction problem. Ajay Agrawal is an ...

Watch on YouTube

About Ajay Agrawal

Ajay Agrawal, a professor at the University of Toronto's Rotman School of Management and founder of the Creative Destruction Lab, has continued to discuss the economic implications of artificial intelligence in a series of appearances. He has described AI as "computational statistics that does prediction," arguing that its power comes from reframing problems—such as driving, email replies, and inspection—as prediction problems. Agrawal contrasted "point solutions," where AI is used to improve an existing process, with "system redesigns," such as Uber's use of navigational AI to create a new transportation model. He has also discussed the development of a humanoid robot capable of performing a variety of tasks, stating that such projects are progressing quietly and that the capabilities demonstrated by ChatGPT are being replicated in other domains. Agrawal has addressed the societal implications of AI, including the potential for machines to eventually perform all work. He has argued for a balanced approach to AI regulation, stating that while investment in safety and policy is needed, slowing down AI development would be a mistake due to its potential benefits in areas like healthcare and climate. He has also noted that the cost of prediction is falling, leading to increased use of AI for both familiar and novel applications.

Source: AI-verified profile updated from Ajay Agrawal's recent appearances. Browse all interviews →

Transcript (1 segments)

Ajay Agrawal0:00

We took all these types of problems that we didn't use to think of as prediction problems and we turned them into prediction. So when I said that there's two ways that we start using more prediction, one is everywhere that we were already doing prediction we use more of it, but also we start taking problems that we never thought of as prediction problems and we turn them into prediction to take advantage of the new AI. So things like driving, I suspect many people here five years ago would not have characterized driving as a prediction problem. If you use email on your phone, let's say Google Inbox or something, and those little recommendations in blue at the bottom when you hit reply, that's Google has read your email and it's predicted how you might want to reply. Every time you click one of those, you're training the AI and it's a prediction. Google has read your email, predicted how you want to reply. We've turned email replies into a prediction problem. Inspection has become a prediction problem. Now we put cameras, pressure sensors, various things, feed that into an AI, and an AI predicts when there might be a failure. So we've turned inspection into a prediction problem so that AIs can do it rather than, let's say, weekly or quarterly, continually 24/7, AI monitoring everything, making predictions.