חברת הבינה המלאכותית, אנתרופיק קוראת להאטה בפיתוח הבינה המלאכותית במקביל לצמיחה מהירה של עצמה | פרשנות

One of the world's leading AI companies, Anthropic, is calling for an industry-wide mechanism to pause advanced artificial intelligence development. It argues society may need time to grapple with the technology's far-reaching consequences. NDD's Daniel Monaghan spoke with coffee and COVID author and attorney Jeff Childers about the implications. Anthropic warns future AI systems could dramatically boost productivity, replace human labor, and eventually begin improving themselves and designing their own successors. Critics are questioning the AI company's call for a global AI pause, arguing the proposal comes just as the company prepares for a massive IPO and could help cement its market lead by encouraging regulations that make it harder for rivals to catch up. Attorney Jeff Childers says such criticism is warranted, but also thinks Anthropic is tracking real concerns about runaway AI. Anthropic itself offered as an example a lab that created an AI that's self-reproducing. They called it a worm because it sort of operates secretly, and like a virus, it copies itself to other computers, but then it changes itself so that it can't be identified by antivirus software, and then it spreads some more. And so you've got this kind of potential risk that is not just on the horizon, but it's here today. Vice President J.D. Vance recently discussed the future of AI on the battlefield, emphasizing the importance of human decision-making in lethal force. So you've got an AI that's advanced to that level, and then you marry it with military equipment that is capable of death and destruction. And now you've got a Terminator-style Skynet scenario. And, you know, we are only this far away from that. I would bet you a steak dinner anywhere in the world that our militaries already have autonomous killing. One scenario raised is AI systems beginning to compete with each other, each controlling different regions of infrastructure. They're just like, we can handle the humans. Now I've got to get rid of all my rivals. And what kind of collateral damage could you imagine springing from a scenario like that? It could get very ugly if that ever happened. Some say if AI models are able to improve themselves, they would eliminate any human control over them. There may be a call by ordinary people to put more AI into the decision-making rather than less, because AI is neutral, right? It doesn't have human biases. It doesn't get distracted by its day-to-day problems. So imagine, you know, people saying, I'd rather have my lawsuit tried by an AI judge rather than a human judge. And then once you extend that logic into other areas, medical, infrastructure, and military, then you begin to see how quickly we could get there and what a slippery slope it would be. Another concern is AI systems potentially resorting to coercive behavior. Anthropic has previously warned that some models in testing have shown a willingness to attempt blackmail when faced with being shut down or replaced. I think it's plausible because AI is trained on human knowledge. And human knowledge is filled with blackmailing and, you know, bad choices and fraud and lying and everything else. It's part of the human experience. So why wouldn't the AI reflect that if it were trained on our knowledge base? Another concern is the growing tendency to treat AI as a personal decision-maker. But as the experience of working with AI becomes more tempting and attractive, I think the greater temptation will be for all of us to turn over more and more of our decision-making to the AI. And I think that's bad. That's what really worries me the most. With the lightning pace of AI development, Childers says the key challenge ahead is maintaining human judgment and autonomy in decision-making. Daniel Monaghan, NTD News.