How do we get the societal benefits of AI while guarding against its risks? In a recent op-ed, today's guest argues that winning in the AI era is not a choice between innovation and safety. I'm Allison Nathan, and this is Goldman Sachs Exchanges.
Joining me is David Solomon, Chairman and CEO of Goldman Sachs. We'll talk about how AI can be advanced with discipline, what AI means for jobs, how businesses are using it, and who should set the rules. We'll also discuss the firm's new commitment to help small businesses adopt AI. David, thanks for joining us.
Thank you. Thanks for having me.
It's great to have you here in the studio. You and I usually have a conversation at the start of the year to talk about what lies ahead, but we're sitting down today again to focus on the topic that seems to be on everyone's mind: AI. So, in your op-ed that was published last Friday, you argued that the AI debate is too focused on a choice between moving fast without guardrails or heavy-handed restrictions that could slow progress. So, what's that debate missing?
Well, I think the debate is missing nuance. Most things that are complex and important aren't so binary. And so, the narrative that's developed that if for some reason we don't let the innovation be unleashed and run at whatever speed with no guardrails, no protections, no risk and controls, we somehow inhibit our competitive position, so therefore, the choice is to regulate, slow down, and stop and weaken our competitive position. I just don't think it's a binary choice. I think it's a much more nuanced discussion. I think we absolutely can spur great innovation with this technology, but we have to have rules of the road. There have to be guardrails. There has to ultimately be a structure of accountability around the deployment of this technology into society in a way where we can really hold people accountable for the end result and what happens, and I think we can do that. I don't think it's different with this technology versus others, but it requires a more thoughtful, nuanced discussion than some of this binary, let it run or let's regulate or control it, stop it, slow it down.
You say the responsibility for control first sits with governments, second with the companies that are building these AI models, and third with the businesses using them, and it's really that third area that is getting less attention. Why are companies deploying AI as important as the other two layers of responsibility?
What I tried to do in the op-ed is I was trying to create a simple framework to enhance the conversation. And so, I think very simply, and this is a simple summary of what's in the op-ed, government needs to play a role and set some rules of the road. The people that are developing these models and the infrastructure that allow the technology to exist and ultimately be deployed in the economy have to take responsibility for what they're building and that it works and that it doesn't do harm. And then the companies that take those products and embed them in their enterprises and their service have to be responsible for the way they're deployed and the impact that they have. And the third is not getting as much attention, but it's equally important. Goldman Sachs has been using AI-type models for trading platforms for almost 20 years, but there's human accountability attached to every one of those platforms. I could never imagine someone at Goldman Sachs saying to me, 'The trading platform went off and did something awful on its own. We couldn't stop it.' We wouldn't deploy it if we didn't feel that we had the control to manage it appropriately, and that there was a human being that was ultimately accountable for anything that goes wrong. That's just one example of what needs to happen, but the enterprises have to be responsible for what they deploy into their businesses.
The financial services sector uses risk management principles and tools that could help developers and deployers. Tell us more about that. What approach do you think makes the most sense here?
There's been some talk of replicating certain practices that we've seen develop over time with financial services regulation, but these are different things. I do think, though, there's a lot that can be learned from the experience that large financial institutions have had, particularly around risk management. Financial institutions, particularly the large financial institutions, have developed great risk and control practices, great audit practices, great testing practices, over a long period of time. Given some of the experiences 20 years ago in the financial crisis, these institutions have learned how to really build up the right infrastructure to test, control, and earn trust. And ultimately, that needs to happen here. And there are businesses that are going to deploy this technology into their enterprise that have great experience in doing those kinds of things, and there are others that need to learn that it's a new muscle group. They have to have risk and control processes of the same level that are going to be necessary for this technology, and they're going to have to figure that out.
So, if we kind of put this all together, David, what does winning in the AI era look like from a safety and innovation perspective?
AI is a tremendous technology that I think is going to have enormous societal benefits. But like with anything else that's new and is creating change, there are going to be risk issues, there are going to be speed bumps, and it's our responsibility as a society to figure out how to implement that and put the proper controls into place. And if we do that, we'll reap the benefits. And while at times things will go wrong or there'll be friction, we will generally control them. And by the way, that's true with almost any technology that accelerates growth, productivity, economic activity. And so, we can do that successfully. Winning, if you want to use that expression, is that we get the benefits of what this technology can do in terms of solving problems in science and wellness and health, drug discovery, disease. There are all sorts of things that you can think about where we can reap enormous benefits. Winning is that we get those benefits, but we have built a system where there's trust and there's a belief that we can manage the risks associated with the technologies it's deployed, and I think we can accomplish that. I'm optimistic we can accomplish that.
Generally optimistic about the potential for the technology and what it can achieve. Also optimistic in terms of the implications for jobs, which has obviously been very much in focus. You've described two camps among business leaders, those expecting a job apocalypse and those expecting a great leap forward. Where do you lean, and what needs to happen for it to play out?
I'm more in the second camp than the first camp with some caveats. Whenever you create new technology, new technology disrupts and automates, and this technology is going to disrupt and automate, and there's just no question that we're going to see that. The question is at what pace? And the question also that doesn't get enough attention is, because of this technology, what new businesses spring up? What new jobs are created? So, remember, we have a very, very dynamic economy in the US. We create and destroy on average 30 million jobs every year. We generally create a million more than we destroy on average. A lot of those jobs come from small businesses. By the way, there's a great opportunity for small businesses to use this technology to accelerate their ability to grow, bring the cost down in the context of infrastructure they need to be successful. And so, broadly speaking, while there'll be disruption, I think there'll also be counterbalances to that. Now, here's a question that I think we've got to watch. If the pace of this change is happening quickly, the bridge between the benefits and the jobs that are created and jobs that are being automated could be quicker. We might need some policy actions to soften that disruption. And I also think you have to realize that a lot of what this technology does is it makes smart people more productive. So, when people talk about white-collar jobs, just because the technology makes them more productive, it doesn't mean the job goes away. It means that the smart people can do more productive things while getting the benefit of the technology in their workflows. So, there's going to be a lot of automation, a lot of workflow change, changes in the things that people do. But by the way, you can look back and notice that this has been going on for decades and decades, and this is the next phase of that evolution. I think about when I started, how I spent my time as a young analyst, and what the productivity of our young analysts are today. We don't have less of them. They just do different things, and they're much more productive than they were 40 years ago. And so that's a good lens. There are lots of jobs that have gone away because of technology. And as we do that, we create new jobs in our economy. I'm confident and optimistic that that will happen over the next decade or two. I think one of the things we have to watch carefully is in the short term, is there a little bit more disruption? And does that require some policy actions or some steps to smooth that out?
