AI in 2024: What Can We Expect?

Dan Loney: How do you process all that we have seen and talked about around AI this year?

Christian Terwiesch: It has been a crazy year. If you think back to the fall of 2022 when ChatGPT came out, it started off as something that we thought of primarily that kids would use to get their homework done. And it took the world by storm. At the same time, we had Dall-E, Midjourney, Stable Diffusion do amazing work in the text-to-image world, and with GPT4 launched, the large language models got even better. It’s been a hell of a ride.

Loney: It’s probably not a surprise to you to see how it has enveloped the thinking of so many businesses over the last year.

Terwiesch: I think technological progress doesn’t typically follow a continuum. There are what we call punctuated equilibria, these moments when there are big shifts. I think we have seen these massive shifts that go beyond what companies expected. Lots of companies we work with at the Mack Institute are now thinking about getting ready for the changes and retooling, now that the new tool is available.

‘Democritization of AI’

Loney: Is this going to be a path that all businesses will be able to take advantage of? Obviously, larger companies that have the resources will easily be able to do this. But what about the mid-sized and small businesses?

Terwiesch: I think it’s actually the opposite. I think this change that we have seen is going to lead to a democratization of AI. It used to be that you have to be a big company to have the big data set that you use to train your AI yourself. Thanks to the GPT technology, basically that training is pre-trained now. Right out of the box, a one-person startup can create a cool app and provide solutions to needs that are currently unmet. So, I think it helps the small guys.

The one thing that we’ve seen with the updates of ChatGPT and OpenAI’s work is that every time they come out with a new release, thousands of these small ventures that were about to take off are wiped away, because their functionality’s baked into the next generation. But I don’t think it is giving the big guys an unfair advantage.

Loney: It feels almost like update for our smartphone. It’s coming at that speed, almost.

Terwiesch: Unbelievable, right? You go from 3.5 to 4 to turbo. Dall-E gets integrated. There’s been progress in AI for many decades. But the last year, it seems to have hit the mainstream both on the availability of the technological solution capabilities and the awareness of what it can do in enterprises where we can use it to create value.

AI in Education

Loney: You and I talked many months ago when you had done research around the impact of AI on it trying to pass an MBA test. How has your thinking around the use of AI in education developed over that time?

Terwiesch: The first thing I think most of us do now is, if we have an exam that is a traditional exam, ban GPT. It’s too good. It passes every exam. You can’t figure that out or detect that. At the end of the day, we at the Wharton School want to give degrees to students, and not to software.

In my operations course, I just created the first assignment where I explicitly ask students to use GPT to help them sort through massive and complex data. In my innovation course, where we start new ventures and come up with ideas for new products and services, I urged my students to use ChatGPT to generate ideas. And even in something as benign as looking at my teaching evaluations, where we get lots of verbal comments, I just put this into GPT and say, “Summarize this and make three recommendations of what I should improve next year.” And it’s pretty good.

Loney: You don’t have a negative viewpoint on it at all. It’s very much a positive.

Terwiesch: Maybe that is wishful thinking. But I mean, we don’t have a choice, right? If we go back to a world where we did not have this technology, it would be a waste, right? This is a huge opportunity for many professions — education, health care, and many others — of really dealing with big problems. Folks are burning out. We had a meeting with the Mack Institute, our board, last night. Folks from Penn Medicine were there. The burnout rate of providers, of people working in medicine, is huge. We need to help these people. If you give me a magic wand that makes me 50% more productive, I take the wand.

Will AI Take Jobs?

Loney: One of the questions that has been brought up is, is AI going to take jobs away? But there has also been a focus on how AI can support the efforts of people in the workplace. Which way do you think it’s going to go?

Terwiesch: I think both. There will be jobs that will change. Basically, any job to be done, there is a customer who needs something. There is a provider that provides a solution, and they have to do this in a way that value is created. So that means the customer is willing to pay more than it costs to generate the solution.

Now the first-order effect is that you have AI generate the solution, which is much cheaper than the human-generated solution. So, lots of people will have to change their jobs. But second order, there are lots of things that right now make no economic sense, doing unmet customer needs that we can fulfill with AI. Which means that there are new jobs. And which way that goes, I don’t think anybody knows.

Loney: How does AI impact being an entrepreneur moving forward?

Terwiesch: I think one of the great uses of AI is as a brainstorming partner. We still see hallucinations in AI. We don’t have the trust in the technology yet where we would say, like, “Drive my car,” or “Manage my retirement portfolio.” The nice thing with entrepreneurship and idea generation is, you’re giving me 100 ideas. Even if 99 of those are horrible, if someone sees one good one in the middle, I take that portfolio. I think that startup aspect of generating ideas for dissertations, for cancer research, for entrepreneurship, that part of creativity we can enhance now. And that, to me, is a no-brainer. I’d take that anytime.

The Need for Regulation

Loney: There are calls to slow down the implementation of AI so that we truly understand what the impact is. Where do you fall on that?

Terwiesch: It’s probably in our lifetime the biggest game-changer that we’ve seen. I think you and I have been around for the internet becoming a big thing. Then mobile came. I think this is probably the biggest thing in our lifetime. So, it’s really hard to predict, right? You see people making the argument for super-intelligence, and that the robots will rule the world. There are lots of things we regulate in our society. You’re not allowed to build a nuclear bomb in your backyard, and it’s a good thing. So, it would be shocking if we would leave this space as totally unregulated.

I think what is especially concerning here is that it took the Manhattan Project over 100,000 people to build an atomic bomb. That’s something that you and I, even with evil intent, couldn’t do in our backyard. You jailbreak a smart AI system, it could do a horrible thing, it will only take one offender. I think some form of security, some form of regulation, is absolutely needed.

Loney: What is your expectation in the next couple of years?

Terwiesch: The first order of forecasting is extrapolation. You basically extrapolate the line, and that suggests that we’re going to see massive progress. It’s not going to stop tomorrow. This famous discussion of AGI (artificial general intelligence), superintelligence, how far we are away, nobody knows. Certainly, I don’t know. But I think that is more of a philosophical question.

Loney: It feels like it’s still the tip of the iceberg, doesn’t it?

Terwiesch: Yes, and I think it will not stop. If, even for just another one or two years, you apply some form of extrapolation of the trend that we’ve seen over the last years, we’re going to have J.K. Rowling-level writing come out of ChatGPT. We could imagine, the context window has just grown substantially. We could imagine much better science papers and dissertations being generated by ChatGPT. I don’t think it takes much of a crazy assumption to imagine a world in four or five years where a lot of these things become true.

A Driver of Productivity

Loney: Does this drive a greater focus on what productivity will be in the future? Because you will have this component as an assistant to be able to get a lot of this work done in a quicker fashion, but also potentially in a more concise fashion.

Terwiesch: I’m a big fan of productivity. It sounds totally boring, German operations management type of boring. But again, think about professions such as health care. Think about educators. We need to make them more productive to improve their quality of life and have healthier patients and smarter children. It’s up for us as a society what we do with these productivity gains. And I very much hope that we use them for better education and better treatments, as opposed to for lower cost of health care and schooling.

In the news, the stories that get reported are the capabilities of AI to help with the diagnosis. It can read an image to text. It can read X-rays. I’m working on a project on mental health right now where it can do pretty decent diagnostics assessments. I think we forget how much of the day in the life of a worker in health care, a nurse or provider, how much is spent writing notes, reports, keeping track of medical records. I think a lot of that can be freed up, and I think that’s much lower-hanging fruit. It’s less glorious, it’s less sexy to talk about, but I think that’s where the first line of attack is going to be.

Loney: I would imagine that health care is benefiting from the fact that it has already made a transition to technology, so this is the next step in the process.

Terwiesch: Absolutely. I come from Germany. My parents live in Germany, get care in Germany. Unbelievable how the system is still paper-based and years behind. We have had lots of people joke and complain about the transition to the electronic health record. It’s been the right thing, and I think we’re going to see massive rewards from that.

Looking Ahead

Loney: What are you most interested to watch or potentially see occur next year in and around this area?

Terwiesch: I think next year I want to see how we go from cool projects of, “Guess what? It can do that?” towards implementing systems — at the bedside or at the school bench — that lead to better outcomes. I think we urgently need it.

Loney: This is going to be that next phase of learning for younger generations. AI can complement a lot of what we do as we are growing up and going out into the world.

Terwiesch: Yeah. Many years ago, you and I had discussions about MOOCs — massive open online courses. We were saying how the MOOCs will disrupt everything. Well, what we couldn’t do with the video-based teaching is personalized tutoring. I think that is a major strength that we have in the technology now. Every kid can be picked up where he or she stands and be coached. I’m in no way implying that we don’t need human workers, teachers, to do this. But we can customize the education to an extent that we could not do before. And I very much hope that this overcomes a lot of the quality discrepancies around schooling. The OECD just this week released the PISA reports, with horrible report cards on math and science for many of the developed nations. Kids have not recovered from Covid in the classroom. The test results are horrible. We need that technology. We need that productivity boost. And we shouldn’t feel scared of it by putting it into this kind of “it’s going to steal our jobs” category.