Ask HN: Why not regulate AI like nuclear power?
3 comments
The software and hardware is advancing quickly.
Without an actual geopolitical solution, militaries (and their industry partners) are racing to develop the hyperspeed general purpose strategic capability faster than their rivals.
This will lead to widely deployed, effective, and inexpensive AI reasoning within just a few years. Outputting at many multiples of human speed. Deployed first at scale in connection to hypersonic missiles, drones, satellites, and advanced malware targeting infrastructure. But shortly thereafter (assuming it does not immediately result in disaster), becoming a ubiquitous locally deployed consumer technology.
Without an actual geopolitical solution, militaries (and their industry partners) are racing to develop the hyperspeed general purpose strategic capability faster than their rivals.
This will lead to widely deployed, effective, and inexpensive AI reasoning within just a few years. Outputting at many multiples of human speed. Deployed first at scale in connection to hypersonic missiles, drones, satellites, and advanced malware targeting infrastructure. But shortly thereafter (assuming it does not immediately result in disaster), becoming a ubiquitous locally deployed consumer technology.
Agree progress is rapid, but I was under the impression model training for LLMs was still the preserve of mega-corps, with the support of eg NVDA providing the hardware.
Military may be doing the same behind closed doors of course, but still a large endeavour.
Are you suggesting it is too late for regulation?
Military may be doing the same behind closed doors of course, but still a large endeavour.
Are you suggesting it is too late for regulation?
Who is going to regulate the US and Chinese militaries? The CEO of Palantir just published an open letter calling for a Manhattan Project for superintelligent AI warfare.
https://archive.is/KSOv4
What needs to be regulated is the performance of the hardware designs. I have not seen anyone except for myself specifically say this. And no one listens to me. So it seems that society may not figure that out until it's too late.
What needs to be regulated is the performance of the hardware designs. I have not seen anyone except for myself specifically say this. And no one listens to me. So it seems that society may not figure that out until it's too late.
Western militaries have justice systems, including courts and prisons. I presume they also follow best practice in lots of domains, despite some legal exemptions. They also clearly have the aim to preserve society in a safe state.
Even if militaries are outside the remit of these new AI orgs, they could still be useful in the civilian world.
Even if militaries are outside the remit of these new AI orgs, they could still be useful in the civilian world.
This is the plot to neuromancer
I’ve tried and failed to read it twice - I just find the dialogue and characters so cringeworthy. Nice world building tho.
I was the same way. My struggle with the book was the way that Gibson presents new lore. He doesn't say "this is an ono-sendai deck; it allows the user to enter cyberspace. it connects to your head here, here, and here". That's how Neal Stephenson writes. Gibson just presents you with an in-universe object or character or company, and talks about it as if you already know what he's talking about, and it's up to you, the reader, to figure it out.
He even does this with narrative. Two male characters will be having a conversation, and all of a sudden the next line will say something like
> "[dialog]", she said
Without introducing who SHE is. You have to wait for the next page to figure out, based on her description.
It drove me insane for years, but now I like it and I've gotten used to it. I like count zero and its characters more than neuromancer. I'm reading virtual light right now but am taking a break to read John Romero's Doom Guy. I'll get to Mona Lisa Overdrive eventually
He even does this with narrative. Two male characters will be having a conversation, and all of a sudden the next line will say something like
> "[dialog]", she said
Without introducing who SHE is. You have to wait for the next page to figure out, based on her description.
It drove me insane for years, but now I like it and I've gotten used to it. I like count zero and its characters more than neuromancer. I'm reading virtual light right now but am taking a break to read John Romero's Doom Guy. I'll get to Mona Lisa Overdrive eventually
Regulating AI and nuclear power differently is a result of their fundamentally different nature, potential risks, and societal impact. Here are some key reasons why AI and nuclear power are treated differently in terms of regulation:
1. Nature of Technology: AI and nuclear power are vastly different technologies. Nuclear power involves the use of nuclear reactions to produce energy, and it poses inherent risks related to radiation, nuclear waste, and potential accidents. AI, on the other hand, is a technology based on algorithms and data processing, primarily aimed at automating tasks and decision-making. While AI can have significant societal implications, its core functionality does not carry the same physical risks as nuclear power.
2. Historical Context: Nuclear power has a long history dating back to the mid-20th century, and its regulation was established in response to the catastrophic events like the Chernobyl and Fukushima disasters. These incidents demonstrated the need for strict safety measures and international cooperation in the nuclear industry. In contrast, AI has seen significant advancements in a relatively short time, and its risks are still being understood and evaluated. As AI technology continues to evolve, regulations are being developed and updated to address emerging challenges.
3. Risks and Impact: Nuclear power has the potential for large-scale catastrophic accidents with immediate and long-term consequences, such as radioactive contamination and health risks. The regulation of nuclear power is aimed at minimizing these risks, ensuring proper waste disposal, and safeguarding against nuclear proliferation. While AI also carries risks, they are often related to data privacy, algorithmic bias, and ethical implications. The negative consequences of AI are generally less immediate and widespread than those associated with nuclear power.
4. International Cooperation: Nuclear power is subject to extensive international cooperation and regulation through organizations like the International Atomic Energy Agency (IAEA). The international nature of nuclear energy and its potential global impact necessitates a unified approach to safety and security. AI, while also a global phenomenon, does not involve the same level of physical infrastructure or potential for catastrophic incidents, making it less reliant on strict international regulatory frameworks.
5. Development Stage: AI is still in a relatively early stage of development compared to nuclear power, which has been in commercial operation for decades. As AI continues to advance, regulations will likely evolve to address new challenges and concerns.
While AI and nuclear power are different in many aspects, it is crucial to acknowledge the importance of appropriate regulations for AI technology. As AI becomes more pervasive in various industries and aspects of daily life, it will be essential to strike a balance between fostering innovation and ensuring ethical use, data protection, and societal well-being. As such, governments and international bodies are actively engaged in developing regulatory frameworks and guidelines for responsible AI development and deployment.
1. Nature of Technology: AI and nuclear power are vastly different technologies. Nuclear power involves the use of nuclear reactions to produce energy, and it poses inherent risks related to radiation, nuclear waste, and potential accidents. AI, on the other hand, is a technology based on algorithms and data processing, primarily aimed at automating tasks and decision-making. While AI can have significant societal implications, its core functionality does not carry the same physical risks as nuclear power.
2. Historical Context: Nuclear power has a long history dating back to the mid-20th century, and its regulation was established in response to the catastrophic events like the Chernobyl and Fukushima disasters. These incidents demonstrated the need for strict safety measures and international cooperation in the nuclear industry. In contrast, AI has seen significant advancements in a relatively short time, and its risks are still being understood and evaluated. As AI technology continues to evolve, regulations are being developed and updated to address emerging challenges.
3. Risks and Impact: Nuclear power has the potential for large-scale catastrophic accidents with immediate and long-term consequences, such as radioactive contamination and health risks. The regulation of nuclear power is aimed at minimizing these risks, ensuring proper waste disposal, and safeguarding against nuclear proliferation. While AI also carries risks, they are often related to data privacy, algorithmic bias, and ethical implications. The negative consequences of AI are generally less immediate and widespread than those associated with nuclear power.
4. International Cooperation: Nuclear power is subject to extensive international cooperation and regulation through organizations like the International Atomic Energy Agency (IAEA). The international nature of nuclear energy and its potential global impact necessitates a unified approach to safety and security. AI, while also a global phenomenon, does not involve the same level of physical infrastructure or potential for catastrophic incidents, making it less reliant on strict international regulatory frameworks.
5. Development Stage: AI is still in a relatively early stage of development compared to nuclear power, which has been in commercial operation for decades. As AI continues to advance, regulations will likely evolve to address new challenges and concerns.
While AI and nuclear power are different in many aspects, it is crucial to acknowledge the importance of appropriate regulations for AI technology. As AI becomes more pervasive in various industries and aspects of daily life, it will be essential to strike a balance between fostering innovation and ensuring ethical use, data protection, and societal well-being. As such, governments and international bodies are actively engaged in developing regulatory frameworks and guidelines for responsible AI development and deployment.
Agree that AI is at a different stage of development and has a different risk profile, but my idea would be to learn from those 100+ years of nuclear power management (successes and failures) and bootstrap a similar set of organisations for AI.
I don’t see anything in your reasons that would fundamentally preclude such a setup. I guess that’s really my question - is there a gotcha for AI development that requires a completely new regulatory solution?
I don’t see anything in your reasons that would fundamentally preclude such a setup. I guess that’s really my question - is there a gotcha for AI development that requires a completely new regulatory solution?
Coordinate internationally via a ‘International AI Agency’ and restrict access to powerful GPUs