[ad_1]
WASHINGTON – When President Biden announced sharp restrictions in October on the sale of advanced computer chips to China, he pitched them in part as a way to give American industry a chance to regain competitiveness.
But at the Pentagon and the National Security Council, there is a second agenda: arms control. If China’s military can’t get the chip, the theory is it could slow efforts to develop weapons powered by artificial intelligence. That will give the White House, and the world, time to define some rules for the use of artificial intelligence in everything from sensors, missiles and cyber weapons, and finally to keep some of the nightmares conjured up by Hollywood – autonomous killer robots and computers. which locks the human creator.
Now, the fog of fear surrounding the popular chatbot ChatGPT and other generative AI software has made the chip restriction to Beijing look like a temporary fix. When Mr. Biden left a meeting at the White House on Thursday about tech executives struggling to limit the risks of the technology, his first comment was that “what you’re doing has enormous potential and enormous danger.”
This is a reflection, the national security aide said, of a new classified overview of the potential of new technologies to solve wars, cyber conflicts and – in the most extreme cases – to make decisions about the use of nuclear weapons.
But despite Mr. Biden’s warning, a Pentagon official, speaking at a technology forum, said he thinks the idea of a six-month pause in developing the next generation of ChatGPT and similar software is a bad idea: The Chinese won’t wait, and neither will Russia. .
“If we stop, guess who won’t stop: potential adversaries overseas,” Pentagon chief information officer John Sherman said on Wednesday. “We have to keep going.”
The blunt statement underscores the tension felt in the defense community today. No one knows what this new technology can do when it comes to developing and controlling weapons, and they don’t know what kind of arms control regime, if any, will work.
The matter is unclear, but very worrying. Could ChatGPT empower bad actors who previously did not have easy access to destructive technology? Could it speed up the confrontation between the superpowers, leaving only time for diplomacy and negotiations?
“The industry is not stupid here, and you’ve seen efforts to regulate itself,” said Eric Schmidt, the former chairman of Google who served as the inaugural chairman of the Defense Innovation Council from 2016 to 2020.
“So there are some informal conversations going on right now in the industry – all unofficial – about what AI security rules look like,” said Mr. Schmidt, who has written, with former secretary of state Henry Kissinger, a series. articles and books on the potential of artificial intelligence to elevate geopolitics.
Early attempts to put fences in the system are obvious to anyone who has tried early iterations of ChatGPT. The bot won’t answer questions about how to harm people with drugs, for example, or how to blow up a dam or disable a nuclear centrifuge, all operations that the United States and other countries do without the benefit of artificial intelligence tools. .
But blacklisting will only slow down the misuse of the system; some think they can completely stop such efforts. There are always hacks to get around safety limits, as anyone who’s tried to turn off the crucial beep in a car’s seat belt warning system can attest.
Although new software has gained popularity on the issue, it is hardly new to the Pentagon. The first rules on developing autonomous weapons were published ten years ago. The Pentagon’s Joint Artificial Intelligence Center was established five years ago to explore the use of artificial intelligence in combat.
Some weapons operate on autopilot. Patriot missiles, which shoot down missiles or aircraft that enter protected airspace, have long had an “automatic” mode. It enables to fire without human intervention when overwhelmed with incoming targets faster than humans can react. But it must be supervised by a human who can cancel the attack if necessary.
The assassination of Mohsen Fakhrizadeh, Iran’s top nuclear scientist, was carried out by Israel’s Mossad using an autonomous machine gun, mounted on a pickup truck, aided by artificial intelligence – although it appeared to have a high degree of remote control. Russia says it has recently begun to build – but has not yet deployed – Poseidon nuclear submarine torpedoes. If it lives up to Russia’s hype, the weapon will be able to travel autonomously across the ocean, bypassing existing missile defenses, to deliver a nuclear weapon days after launch.
To date, there are no international treaties or agreements dealing with such autonomous weapons. In an era where arms control agreements are abandoned faster than they are negotiated, there is little prospect of such an agreement. But the kind of challenge posed by ChatGPT and others is different, and in some ways more complex.
In the military, AI-infused systems can speed up the tempo of battle decisions to a degree that creates the risk of accidental attacks, or mistaken or deliberately false decisions about incoming attacks.
“The core problem with AI in the military and national security is how do you defend against attacks that are faster than human decision-making,” Mr. Schmidt said. “And I don’t think the problem is solved. In other words, the missiles are coming so fast that there has to be an automatic response. What happens if the signal is false?”
The Cold War was full of false alarm stories – once because a training tape, used to practice nuclear response, had been inserted into the wrong system and signaled a massive Soviet attack. Paul Scharre, of the Center for a New American Security, noted in his 2018 book “Army of None” that there were “at least 13 nuclear incidents from 1962 to 2002,” which “provide confidence that such near misses are a normal, if frightening, condition of nuclear weapons.” .”
Therefore, when the tension between the superpowers is lower than it is now, some presidents are trying to negotiate to build more time to make a nuclear decision on all sides, so that no one rushes into conflict. But generative AI threatens to push countries in another direction, towards faster decision-making.
The good news is that the major powers have to be careful – because they know how the enemy will react. But so far no rules have been agreed upon.
Anja Manuel, a former State Department official and now principal at the Rice, Hadley, Gates and Manuel consulting group, wrote recently that while China and Russia are not ready to discuss arms control on AI, meetings on the topic will lead to discussions. about the use of AI that appears to be “beyond the pale.”
Of course, even the Pentagon would be worried about agreeing to so many restrictions.
“I’m fighting really hard to get the policy that if you have an element of autonomous weaponry, you need a way to kill it,” said Danny Hillis, a prominent computer scientist who pioneered parallel computers used for artificial intelligence. Mr. Hillis, who also serves on the Defense Innovation Board, said pushback came from Pentagon officials who said “if we can kill, the enemy can kill.”
So, the greater risk could come from individual actors, terrorists, ransomware groups or small countries with advanced cyber skills – like North Korea – learning how to clone smaller, less narrow versions of ChatGPT. And it may find that generative AI software is well-suited to speeding up cyber attacks and targeting disinformation.
Tom Burt, who heads the trust and safety operations at Microsoft, which is accelerating the use of new technologies to transform the search engine, said at a recent forum at George Washington University that he thinks AI systems will help defenders detect anomalous behavior faster than they do. will help the attacker. Other experts disagree. But he said he feared it could “supercharge” the spread of targeted disinformation.
All of this represents a new era of arms control.
Some experts say that since it’s impossible to stop the spread of ChatGPT and similar software, the best hope is to limit the specialized chips and other computing power needed to advance the technology. That will certainly be one of the different arms control formulas in the next few years, when the major nuclear powers, at least, seem disinterested in negotiating old weapons, let alone new ones.
[ad_2]
Source link