Skip to content
National Competition in Artificial Intelligence episode artwork

Sep 26, 2017 9 min

National Competition in Artificial Intelligence

from CNAS Podcasts Hosted by Center for a New American Security

Vladimir Putin recently said that whoever leads in artificial intelligence "will be ruler of the world." Elon Musk has warned that national competition in AI is the "most likely cause of WW3." Are we at risk of an AI arms race?

Episode details come from the podcast's RSS feed.

Embed this episode

Ready to play

National Competition in Artificial Intelligence

0:00 9:26
of matches

Transcript Auto-generated, may contain errors

This is Neil Erwits with the Center for New American Security. We're here today with Paul Shari, the Director of our Technology and National Security Program, who's going to be talking to us about artificial intelligence and a lot that's happened. Paul, thanks for joining us. Thanks, Ryan.

So most recently Putin, Vladimir Putin said that whoever leads in artificial intelligence will rule the world. Is he overstating the power involved in artificial intelligence? Or is he basically right? I think he's on point on this one.

What we're seeing is a tremendous wave of machine intelligence or artificial intelligence. He's very powerful. I look at it as something akin to another industrial revolution, something that was heard from a number of people looking at this space. And for a country to lag behind in that, what I think would be strategically disastrous.

So, yeah, if we're looking at something like another industrial revolution, whoever comes out on top and uses this technology best is really going to benefit them. So Iwan Musk, more or less in response to Putin's comments, said that artificial intelligence is the most likely cause of World War III. Was that an overstatement or the risks here, the downside risks really that acute? Well, we've talked before about some of the concerns that Musk has raised about artificial intelligence, maybe running a mock, and it costs some problems.

I think something worth keeping in mind here is that when we talk about another war, that's not just about a safety risk. It's also looking about security risks that could come from people deliberately using artificial intelligence for harmful means. Like what? Well, like maybe cyber weapons, right?

If we could use AI to analyze big data, certainly governments and military could use that for all sorts of harmful purposes. Now, on the flip side, the head of AI for Google said, slow down. It's irresponsible to talk about it as though we're going to hit I-Robot in the next six or seven minutes here. Is he right?

Does he have a point? Or do we really need to be focusing on something that could spiral out of control? Yeah, I think what's hard in these conversations about AI risk is you have people coming from a variety of different perspectives and they're talking about sometimes different things, right? When Musk is generally speaking about what he talks about AI risk, it's really long-term challenges.

Not something that's in the near term, but what might be possible maybe decades down the road if AI continues to advance at the pace that it is now, which might not, right? It might be around. That's definitely possible. And a lot of AI scientists have been very dismissive about this saying, look, in the near term, you know, I've seen those movies too, but they're not going to, AI's are not going to rise up against us.

Machines are going to turn on us. And I think that's a bit of a cartoon oversimplification, you know, like any new technology. There are risks to thinking about how to use it and employ it safely. And there are some near term safety problems with artificial intelligence or machine learning, sometimes learning the wrong things.

When you have, you know, a machine that's being used in a really controlled setting, it's playing a game on a computer, and it learns something clever and creative and maybe not what you want, it's amusing and entertaining. It might not be amusing if a country develops a cyber weapon and then moves control of it or does something unexpected. And I think that's an important component to add to this mix, which is, look, Google was thinking about this and other AI companies from the respective of, they want to use the technology to benefit humanity and maybe make driving safer or analyze healthcare records better. They're looking at it at things that are just really beneficial.

But you've got to keep in mind that there are people like Vladimir Putin out there who are going to use this technology to do things like, you know, like, he's not using it just to analyze medical records? Yeah, I mean, I don't think Putin's out there just like sort of save lives and make them all the better places. This isn't really open technology area, right? Anyone can get their hands on this if you're going to apply this.

So, you know, Russia may not be the leader in inventing new AI concepts in research, but they certainly have the capacity to take this technology and adapt it. Just like people adapted, you know, the internal combustion engine to build tanks and submarines and airplanes. Let me ask a kind of related question then, is the concern that we would get into a sort of automatic tit for tat before any human was able to put a safety break on this thing that RAI would respond to a perceived hostility from Russia's AI, even though Russia didn't intend to be hostile and in 10 minutes were at war, even though no human looked at it? I mean, I think when you start to look at this technology in the national security context, yes, those are the things that you have to start to think about because you're in a competitive dynamic where countries are trying to get an edge and there might be an advantage of being a first mover or reacting quickly.

We've already seen in stock trading that this has led to, you know, a race in speed and machines making trades in milliseconds, companies doing everything they can to shave off the time it takes them to execute a trade down to microseconds, and the result has been that you can have interactions among these stock trading bots that happen so quickly and it means that people can't possibly react like flash crashes. The way that the stock traders and the financial rate there's a deal with this is they create the circuit breakers to stop trading if the price moves too quickly on a stock, but there's no, you know, there's no referee to call time out in war. And so if you have Russia or other countries using cyber weapons, offensive and defensive, when you start automating things because you say, well, you know, I need to when we will respond, you could, I mean, it's conceivable, I think, to imagine, particularly in domains like cyberspace, we have things happening at machine speed in time-forms, I say, in milliseconds, faster than human reaction. So you can have these interactions that begin to spiral out of control or happen in ways that humans did not expect.

And I think we need to keep that into account as we think about how this technology is going to be used and implemented. It's not only going to be for beneficial purposes, and we've got to really be prepared for that. Speaking of beneficial purposes or lack thereof and competition, and this will be the last question, you wrote a piece not long ago talking about how it's going to be very hard for the international community to regulate these, that with the use of AI, it's very easy to imagine a race to the bottom. Do you still have that fear?

I think, you know, this is one of the, I think, the unknowns in this space is right now there are some unsolved safety problems. And so most of the AI researchers going towards making the technology better, and we see in the past couple of years, more money to expand, and most of the particular really has opened up this pocketbook and is a funder of a organization called OpenAI. He's put a 10 million dollars towards AI safety research. He's, you know, despite what people think about his claims one way or the other, I mean, he's really tried to push the boundaries, push the ball forward on funding AI safety research, but there's still much more money going towards making AI better, right?

So this would be kind of like people focusing all their attention towards making cars faster, but no one's paying attention to things like seat belts and airbags, right? And those are important things. So it's like the 60s. Yeah, right?

And we went through an era like that with Arnold Beals, which was kind of dangerous, right? And then we ended up with things like seat belts and regulation and standards. And so I think, you know, we're at an early stage of the technology's development with AI. So there's going to have to be some equivalent of that, right?

Some equivalent of this seat belt in the airbag. But it takes time to figure out what that is. And we need to add that in too. Now the danger is in a national security context where you may have domestic regulation, but not international regulation.

And Putin doesn't care about seat belts, right? And so maybe that's fine if the seat belt doesn't make the car slower, right? And this is one of the unknown challenges is are there some safety restrictions that might make AI safety safer, but maybe less capable in some way, right? And then you might say, look, I've got to take my cars up to the track and raise my competitor, but he's taking, he's stripped out of the airbags and seat belts.

And he gets a little edge. And so I've got to do the same. And this concept of a race to the bottom of the AI safety is, I think, a valid concern to think about in the national security space. Well, thanks, Paul.

Here's hoping we continue to talk about this under the right circumstances, not the wrong ones. Yes, thanks very much.

Frequently asked questions

How long is this episode of CNAS Podcasts?

This episode is 9 minutes long.

When was this CNAS Podcasts episode published?

This episode was published on September 26, 2017.

Is there a transcript available for this episode?

Yes, a full transcript is available for this episode. You can read the complete transcript on the episode page.

Can I download this CNAS Podcasts episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.