Discussion · Senate
Mark R. Warner Democrat
“Artificial intelligence is advancing quickly and is already used in our military. Flawed AI almost caused us to mistakenly bomb a Chinese ship. We need basic safety rules, required testing, and a government panel to check AI tools before they are used.”
“Madam President, I will wrap it up here. I have been in Virginia talking about this for a while. We have a lot of data centers, which is of some concern. But, boy oh boy, there was a seismic change starting as the companies themselves started to say: Holy heck. And just think about that. Mythos came out in March. National security and others said: This is going to change warfare. The OpenAI Hugging Face incident happened in July. That was 3,000 agents working together. In August, OpenAI got 10,000 agents to work together. It was on a different problem; they solved a math problem. But the idea is that there is kind of an exponential increase in sophistication, as these agents work together to cover their tracks, that, as Senator Schatz said, potentially could move beyond human control. If we just say ``Well, that is still hypothetical,’’ let me give you two last comments. Just today, we found that actually a model had already been released from OpenAI that attacked the Australian healthcare system. If it attacked our system, I think we would be pretty upset. Even more compelling--some of us have written a letter to both the folks on the intel side and on the defense side--there was a ship from China sailing towards the Middle East, and we got very close to bombing that ship because the AI tool and the intel said there might be weapons on that. Thank goodness that at some point, somebody did a little more checking, and we didn’t do that, but we were close to a geopolitical event that could have been catastrophic. These tools are already built into our systems, and clearly, you know, the way our military executed on Maduro--you may not agree with the greatness of it, but it was brilliant. But we don’t have safety protocols. So we are not asking for the world. Let’s do the testing early. Let’s make sure that you have a safety standard in all of these companies. Let’s make sure we do the sandbox right, report on the incident. We could discuss, you know, who is on that panel. It needs to be mandatory. It needs to have government. It needs to have national security. And Lord knows we need to have researchers, scientists. All of these individuals are asking for this. Who are we to turn that down? I am going to go back to where we started. This will be the most significant technology innovation in our lifetimes. It will make wireless and the internet and the cloud small in comparison. There could be enormous positives coming out of this. There are also going to be disruptions. But if we don’t do this in a safe manner, then shame on us. Again, I come back. You wouldn’t climb on an airplane that hadn’t been checked. We are just saying let’s do those basic safety checks. I yield the floor. I suggest the absence of a quorum.”
Madam President, I will wrap it up here. I have been in Virginia talking about this for a while. We have a lot of data centers, which is of some concern. But, boy oh boy, there was a seismic change starting as the companies themselves started to say: Holy heck. And just think about that. Mythos came out in March. National security and others said: This is going to change warfare. The OpenAI Hugging Face incident happened in July. That was 3,000 agents working together. In August, OpenAI got 10,000 agents to work together. It was on a different problem; they solved a math problem. But the idea is that there is kind of an exponential increase in sophistication, as these agents work together to cover their tracks, that, as Senator Schatz said, potentially could move beyond human control. If we just say ``Well, that is still hypothetical,’’ let me give you two last comments. Just today, we found that actually a model had already been released from OpenAI that attacked the Australian healthcare system. If it attacked our system, I think we would be pretty upset. Even more compelling--some of us have written a letter to both the folks on the intel side and on the defense side--there was a ship from China sailing towards the Middle East, and we got very close to bombing that ship because the AI tool and the intel said there might be weapons on that. Thank goodness that at some point, somebody did a little more checking, and we didn’t do that, but we were close to a geopolitical event that could have been catastrophic. These tools are already built into our systems, and clearly, you know, the way our military executed on Maduro--you may not agree with the greatness of it, but it was brilliant. But we don’t have safety protocols. So we are not asking for the world. Let’s do the testing early. Let’s make sure that you have a safety standard in all of these companies. Let’s make sure we do the sandbox right, report on the incident. We could discuss, you know, who is on that panel. It needs to be mandatory. It needs to have government. It needs to have national security. And Lord knows we need to have researchers, scientists. All of these individuals are asking for this. Who are we to turn that down? I am going to go back to where we started. This will be the most significant technology innovation in our lifetimes. It will make wireless and the internet and the cloud small in comparison. There could be enormous positives coming out of this. There are also going to be disruptions. But if we don’t do this in a safe manner, then shame on us. Again, I come back. You wouldn’t climb on an airplane that hadn’t been checked. We are just saying let’s do those basic safety checks. I yield the floor. I suggest the absence of a quorum.