Sounding the Alarm on AI: A Conversation Across the Political Divide | Civil Discourse | Clip 2

a kid gets on and says how do how do you commit suicide and an AI gives them different ways of of doing it terrible result and so the companies now are really working on uh ethical uh uh delivery of AI to make sure that those kind of harms aren't out there then there's the kind of harms that we see that that are uh you know nation state harms how does a how does a uh a Department of Defense uh mil milary use AI in in attacking um but then there are the unforeseen uh areas and that's really the the challenge uh that this commission has to deal with and I think all of us have to really be aware of that uh we don't know uh what what is what is out there um in in five or six years with AI in terms of of the the harm that can happen I want to touch a little bit you mentioned ethics a bit earlier and you know something that keep keeps me up at night and you touch on this in your you know ncle no autonomous nuclear launch uh bill that you introduced um we had during the Cold War there was uh a kind of escalation around nuclear weapons basically Mutual assur destruction Doctrine and so forth with AI we have a situation where we have U you in communist China there really are no kind of ethical boundaries by which the folks over there are going to be constraining themselves here we hopefully do have significant ethical boundaries by which we we will be constraining ourselves here yet I imagine there may be this kind of arms R accelerating arms race I don't I'd like to get both of your perspective maybe we can start with Congressman Le on this and how do we just deal with that reality that's an interesting question because in terms of AI you could actually use AI to counter other kinds of AI so right now there are companies working on essentially watermarking technology that can tell you if a picture you see is a original picture or if it's an altered picture so there are potentially ways for AI to be used and a way to help you um figure out if something is fake or if it's um not true but that's a difficult question because you're right there are going to be bad actors going to be bad countries there going to be criminals there going to be hackers who try to use it and again I think being somewhat skeptical of of AI is is I think healthy for people and also just to understand that in this new world you're going to have a lot of folks trying to hack into your data and just practicing better cyber security practices and just being more aware uh that there are Bad actors out there who are going to try to steal information from you and take certain steps to protect yourself if I can just quickly follow up I mean I'm thinking about things like for example autonomous weapon systems okay that that's a very a robust area where there's a i development happening right now happen to know in both countries so you know you can imagine an escalation of an arms race you have to you know these AIS they respond very quickly right if they how should we really have a human in there is that reasonable maybe we should you know lower our guard over here to be able to deal with that threat I'm talking about that kind of stuff I so I think yeah both representative buck and I have come to the conclusion that we don't think AI should ever launch a nuclear weapon by itself and we would have that view whether or not China or Russia or other countries do that we think it would be insane if they do that but we're still of the view that the United States should never get get to that place where we just let an AI launch a weapon by itself yeah I think that's right and and and I would even go uh a few steps below a nuclear weapon because if we have an autonomous system that sends a conventional missile um into a foreign Capital foreign major city um it starts something that uh humans uh might want or might not want um and so I think that uh there's no way of anticipating uh enough of the factors with AI to to really trust AI um in in initiating conflict um in responding to a conflict that is that exists is is a different story but I think when you initiate a conflict um or respond within moments to some other country initiating a conflict uh there's issues so I I I I think our bill deals specifically with uh nuclear weapons I I would even go a step further and make sure that our military is prepared to uh deal with nonclear or Conventional Weapons and I just want to add to that it is fundamental and inherent in this technology the way it's designed because we don't really again know what it's doing right we can't trace it from point A to its conclusion so this AI algorithm could be correct 99.999% of the time but that is not good enough if it's controlling nuclear weapons right because there could be this one random scenario where it goes totally Bonkers and we would not know and that's why it's so important I think to have a human in Loop both in that area and then now that R BR mentioned what he just said we're going to maybe look at other bills to introduce as well and I think to to this point I'm also curious to bring it a step down even further not just nuclear and and other missile systems but even uh say drone systems or or robot autonomous robots that can do you know door-to-door Urban combat and things like that is there a concern that this type of Technology can actually level the playing field if you will with you know foreign foreign countries that are adversaries to the United States like Iran well it can do more than level the playing field it can level buildings and and all kinds of things and so we've got to be careful um how we use it how we Empower AI uh at some point um we hope uh there is a switch that can be turned on and turned off and if we don't have a switch that you can turn on and turn off then uh you get into the the sort of uh Sci-Fi movies that we've all been seeing um in the last few decades but I I think it's really important that um and in the the concept uh of the bill that that uh Congressman Lou and I introduced is to make sure that uh there is a human in the chain of command uh for nuclear weapons um what you're talking about uh we absolutely need for uh non-nuclear weapons but also so you know say we send the Drone someplace yes as long as we have control over the uh Target of that of that drone um and should we trust the Drone so that we don't have to maintain some sort of communication with the Drone that that's something the military is going to figure out and it's really goes beyond my capability and I think that's why Ted and I feel so strongly about having a commission you know something that's also been on my mind a lot when it comes to AI has been just this idea you you know but let's take it down even a little further just an autonomous car that's driving let's say it causes an accident kills some people it's definitely it's going to happen they make mistakes we know that already um who's really responsible in those situations that is a fascinating ethical question in the United States approximately 40,000 people die every year uh in car accidents most of that is human error autonomous vehicles are going to reduce that number so it's going to save lives now what if the autonomous Vehicle Manufacturer knows though that you know if they spend another six months on their technology it can better identify certain risks and not result in uh accidents that would you know harm passengers of their driver do they wait 6 months to deploy it or do they deploy it right now because it's already better than a human who's driving