This is Neil Erwits. I'm here with Paul Shari, who is the new director of our Technology and National Security Program, and one area where he has already been and will be focusing much more is artificial intelligence and how it's going to affect the battlefield. Paul, thanks for joining us. Thanks, Neil, for having me.
So we've heard how artificial intelligence is going to reshape the economy, that it could have tremendous effects on traffic and what it can do to the trucking industry and so many other things. You hear less about the effect that it's going to have on the battlefield. What effect will it have? Well, I think that's a good question.
It's one that we're interested in exploring in the research and our technology program. We've been doing work for a number of years on robotics, automation, and inhabited systems in a particularly military context. There will be a battlefield but also for important back-end functions like logistics and networking and communications medical evacuation, for example, as an opportunity. But there are these broader issues, both in terms of other national security applications beyond just battlefield use but also broader technological implications.
We're really at the beginning of what many people are describing as a new industrial revolution, this one powered by artificial intelligence. And this is really because of some very exciting advances in the past several years in a type of artificial intelligence technology called deep learning. And these deep learning systems are very, very powerful. They're already seeing them being applied to things like healthcare, finance, transportation.
People are talking about massive disruption in these sectors on the order of trillions of dollars, looking at major job dislocation, jobs being created, jobs being destroyed, probably unfortunately not the same, jobs are the same people, which will be challenging. But there's just been much less thinking done on one of the national security and global security implications of this technology. For talking about something on the scale of an industrial revolution, well, we saw in past industrial revolutions that that shifted not only the balance of power between countries but really the nature of power itself. So coal and steel producing nations became more powerful, oil became a strategic resource that superpowers battle for influence over around the globe.
What is that in an era of artificial intelligence? So we're going to see a battle for Silicon Valley? I mean, well, I think the one of the questions is what is the strategic resource? How do companies and nation states, and I think it's really important to realize here that nation states don't necessarily have even the lead in terms of power with this technology, right?
That while in the past, nation states were made very powerful by private private industrial revolutions here. Now we're looking at all the really exciting innovations are happening in the private sector. So who has the power? Where does power come from?
Some of it's clearly data building up proprietary data sets that can be leveraged in for machine learning. But some of us also human capital, it's finding the right talent and being able to then leverage things. I think something's going to be about speed. It's going to be about processes that allow companies and governments to leverage this technology and innovate quickly.
So you mentioned both the balance of power and speed. Does the US have a head start here? Are we going to be able to control the rules of the road for a better term, or is this something that could really knock the US off its perch? Well, and I think the question is what do we mean by the United States?
If we mean the country in the broadest sense and it includes US companies and individuals, the United States is as a nation is very much a leader in artificial intelligence. A lot of the top companies, not all of them, but a lot of them around in the AI field, Google, IBM, Facebook, Apple, Amazon, Microsoft, they're the US based multinationals. So of course they don't work exclusively in the United States, but they have a lot of US researchers. But China is doing quite a bit in artificial intelligence.
There's a lot of really exciting Chinese companies like Bidey, they're doing very cutting edge work. There's a lot of great Chinese researchers. There's work going on in the UK and companies like DeEin, which is owned by Alphabet. So I think, and what's important is all of that in the United States is happening in the private sector that's not necessarily tied into government.
So even though these US companies might be really at the leading edge, that doesn't necessarily mean that the government, particularly for national security implications, is really well positioned to take advantage of this because there is, in some circles, the AI community, quite a bit of hesitation about necessarily engaging with people in national security space. Which starts about misusing the technology. Now, a lot of people when they think about artificial intelligence and particularly artificial intelligence and the warfare sphere for a lot of a better term, again, are thinking about iRobot or Terminator. Is there a major downside here for weaponizing AI or is there safeguards we really should be or need to be putting in?
There's been a lot of concern among AI researchers about some of the harmful applications of AI. In a number of ways, one, there are some basic concerns about simply AI safety. A lot of the AI tools today are very powerful, but they have a number of vulnerabilities that can be vulnerable to spoofing attacks that can feed them false data and then lead to harmful outcomes that are not consistent with the operators' intentions. But there are also a suite of safety problems that are lumped under what may refer to as a control problem.
And this isn't really an AI becoming self-aware and rising up against humanity like a science fiction. Something much more practical about AI systems that maybe do precisely what they are programmed to do or what they learn to do based on guidance they've been given, but turns out to not really be what you meant at the time. So one of my favorite examples is AI bot that was playing old classic Nintendo games from the 80s and learned playing Tetris to hit the start button to pause the game right before the very last brick fail because then it would never lose. And that was, yeah, that was allowed under those rules it was given.
That was shrewd. It was pretty clever, right? And that was legit. Right, of course.
Sounds like something my older brother would do while I was waiting to play. Yeah, I mean, it's kind of like a move kind of a cue-able. But so that's, you know, that's kind of humorous. It is when you're talking about Tetris.
When you think about something analogous to that in the real world, when you have AI systems doing really important functions like transportation and finance or healthcare or national security, maybe things like cyber security, that may have some very serious rule of implications. And so that's something that researchers are grappling with and trying to think about how to make this thing safe. And that's a concern we've seen TSA talking about using neural networks for image recognition and some other scanners at airports. While a lot of these image recognition algorithms that use deep learning systems are vulnerable to these spoofing attacks where they can be fed false data.
And they buy it. They buy false data. And there's not really any known ways to correct that vulnerability right now. And then there are many researchers worried about malicious applications of AI, which unfortunately we live in a world that's going to happen, just like people have created malware and viruses and worms and intelligent horses and other kinds of computer software that's harmful.
People will certainly use AI to those ends. And so some researchers will sort of think about what that might look like and ways to counter it now, which is really important to get ahead of that. Well, this is going to be fascinating over the next few years. Thanks, Paul.
Yeah, thank you.