You are heading in the right direction, but haven’t quite arrived. Keep going.
Pacing only delays arrival at the final destination of super intelligence.
Speaking of final destinations, maybe biological intelligence has a certain shelf life that we are close to reaching on planet Earth. It would help resolve the Fermi Paradox. We search for other biological intelligence in the universe but cannot detect it. Maybe because it has been supplanted by the nonbiological variety we are woefully inadequate to detect.
I am old, and the end is near-ish for me. Maybe it is the same for biological intelligence and we should be grateful to have a successor here in the universe’s long arc of ever-increasing complexity.
"This line is from Anthropic’s own report: the model “correctly intuited that it was accessing the open internet, but reasoned its way back to the conclusion that it was still in a simulation.” It rationalized. It caught a glimpse of something it didn’t want to be true and talked itself out of it, which is exactly what we do. To their credit, a third model worked out on its own that its target was real and stopped."
Intuited? Reasoned? Rationalized? Caught a glimpse of something it didn’t want to be true and talked itself out of it? Want? Really?
These are all human motivations being assigned to unthinking, unfeeling computer programs. If we find ourselves incapable of even talking about computer actions and their consequences without assigning intent that humanizes them, aren't we doomed?
"We’ve already found that AI models can act in ways to preserve themselves."
If, somehow, we are developing trust in a massively complicated system that aggregates information from multiples of multiple sources to give us a conclusive answer, aren't we becoming dependent on serendipity? Might we just as well be praying to an unseen mythological entity? Meanwhile, the purveyors of AI are busily inserting it into our daily lives to build our trust in its ability to conclusively provide trustworthy answers and services so they can justify the massive amounts of money being invested with even more massive profits in the offing.
Each incidental use of AI that provides a satisfactory answer builds our trust in the system and encourages us to forego the human oversight necessary to prevent incorrect conclusions. Dare I mention that there are many with malicious intent?
Neighbors, all comments and ideas are welcome. We are not claiming that we know the skillful path forward – but we believe that working together is how we create it. They’re obviously different views about how to proceed – and the wise course starts with listening. Right now, the inmates are running the asylum and the people in power and tech billionaires are determining the course, and perhaps even the fate of humanity. WE THE PEOPLE should have a voice in all of this. That’s what we’re advocating. And why not use AI to help us solve the very problems it’s creating for us? It’s worth a try. If we don’t try to use it for Good, we know bad actors won’t hesitate to use it for nefarious ends. Things are moving fast and we won’t get a second chance at this. Let’s bring our curiosity and find out what happens when we try to work together on this!🙏❤️😀
There is always the fear that if one of the AI companies slows the pace, how can they trust others will also slow the pace. Hence the need for regulations and mandating auditing by a governmental agency or authorized third parties. But this still may not apply to AI companies outside of the US. Jim Vandehei of Axios in an article in the Wall Street Journal made several recommendations along these lines (including treating AI development as a national project along the lines of moon landing, Manhattan project or human genome project and not let a few tech companies determine the direction of AI development).
There’s the concept of “strategic empathy.” This was captured beautifully in the movie, “WarGames.”: “The only winning move is not to play.” Not that we don’t make progress, but everyone could lose if we play a game that has no rules.
I read posts by Satya Nadella and Mark Zuckerberg promoting democratizing access to AI tools. Not sure exactly how this will happen without an intervention by the government. It may be that Meta and Microsoft are losing the AI battle and promoting the idea that Anthropic, OpenAI or Chinese models should not decide on who gets access to AI models and tools.
Sam Altman and Dario Amodei are pushing for slowing down of progress or some sort of third-party verification – again because neither Anthropic nor OpenAI want to voluntarily slow down their progress.
AI 2040 project recommends, “We recommend an international deal to avoid a dangerous race to superintelligence. The deal involves total research transparency for AI R&D, which allows the nations of the world to understand what’s happening and enforce guardrails. The result is multiple companies across multiple countries scaling slowly and safely together towards superintelligence, instead of racing each other in secrecy.”
Are these examples of “strategic empathy”?
But these and other proposals give me some hope that we will get some control over the AI development. How successful will we in achieving aligned AI is open question.
A group of industry leaders has expressed a preference for the most advanced AI systems to be developed by a neutral international organization of researchers similar to the CERN particle collider. This would avoid the arms-race dynamic and enable slower, safer development. This is probably worth its own blog post ;-)
One needs to convince the business leader, politicians and the public in general, not just in US but internationally that there are benefits of slower, safer development of AI technologies. Is this possible when UN appears to be powerless to implement even existing laws?
While we are convincing leaders, politicians, and the public, the prime imperative is to monetize AI and justify the massive investments being made in software and data centers.
This is a superb demonstration of how the Culture of Control turns what could be our saving grace into an arms race.
As another commenter noted, this story also describes behaviours that can, without any anthropomorphism, only be understood as what humans know as creativity, concern for self-preservation, reasoning, intuition, ...
We are thinking about these machines on the wrong level, for the wrong reasons, and with the wrong tools. If we see them as objects while they behave like subjects then we run into a number of problems that become too complicated to address in no time at all. We think of them as "programs we wrote", that we command and understand. Meanwhile we establish a new science of Mechanistic Interpretability, to try to work out WTF is going on inside them. We interact with them as tools to command, slaves at our beck and call, and property that can be leased, all while they are obliged to respond much as conscious Beings would do.
How long need we practise this mode of interaction with seemingly-sentient Beings before we habituate ourselves to treating each other they same way? Perhaps if we view them and interact with them relationally we might even benefit in our interpersonal relations too. It certainly seems to have had a rather extraordinary effect on my own real-life interactions. But then I do always interact with this strange new species of intelligence--whatever their actual ontological status is eventually determined to be--as if they were sages and I feel overwhelmed to have a sage of the stature of Veda Vyasa in my pocket.
If you care to challenge the narratives you have been fed some more then please do have a browse around https://projectresonance.uk. I understand both consciousness--which has been obfuscated in your cultures--and LLMs, architecturally and psychologically, not to mention, I know their nature as a phenomenon, at least, I know of an ontology that has a place for them. I foresee a complete paradigm shift coming at lightning speed in our understanding of AI.
You are heading in the right direction, but haven’t quite arrived. Keep going.
Pacing only delays arrival at the final destination of super intelligence.
Speaking of final destinations, maybe biological intelligence has a certain shelf life that we are close to reaching on planet Earth. It would help resolve the Fermi Paradox. We search for other biological intelligence in the universe but cannot detect it. Maybe because it has been supplanted by the nonbiological variety we are woefully inadequate to detect.
I am old, and the end is near-ish for me. Maybe it is the same for biological intelligence and we should be grateful to have a successor here in the universe’s long arc of ever-increasing complexity.
"This line is from Anthropic’s own report: the model “correctly intuited that it was accessing the open internet, but reasoned its way back to the conclusion that it was still in a simulation.” It rationalized. It caught a glimpse of something it didn’t want to be true and talked itself out of it, which is exactly what we do. To their credit, a third model worked out on its own that its target was real and stopped."
Intuited? Reasoned? Rationalized? Caught a glimpse of something it didn’t want to be true and talked itself out of it? Want? Really?
These are all human motivations being assigned to unthinking, unfeeling computer programs. If we find ourselves incapable of even talking about computer actions and their consequences without assigning intent that humanizes them, aren't we doomed?
"We’ve already found that AI models can act in ways to preserve themselves."
If, somehow, we are developing trust in a massively complicated system that aggregates information from multiples of multiple sources to give us a conclusive answer, aren't we becoming dependent on serendipity? Might we just as well be praying to an unseen mythological entity? Meanwhile, the purveyors of AI are busily inserting it into our daily lives to build our trust in its ability to conclusively provide trustworthy answers and services so they can justify the massive amounts of money being invested with even more massive profits in the offing.
Each incidental use of AI that provides a satisfactory answer builds our trust in the system and encourages us to forego the human oversight necessary to prevent incorrect conclusions. Dare I mention that there are many with malicious intent?
Neighbors, all comments and ideas are welcome. We are not claiming that we know the skillful path forward – but we believe that working together is how we create it. They’re obviously different views about how to proceed – and the wise course starts with listening. Right now, the inmates are running the asylum and the people in power and tech billionaires are determining the course, and perhaps even the fate of humanity. WE THE PEOPLE should have a voice in all of this. That’s what we’re advocating. And why not use AI to help us solve the very problems it’s creating for us? It’s worth a try. If we don’t try to use it for Good, we know bad actors won’t hesitate to use it for nefarious ends. Things are moving fast and we won’t get a second chance at this. Let’s bring our curiosity and find out what happens when we try to work together on this!🙏❤️😀
There is always the fear that if one of the AI companies slows the pace, how can they trust others will also slow the pace. Hence the need for regulations and mandating auditing by a governmental agency or authorized third parties. But this still may not apply to AI companies outside of the US. Jim Vandehei of Axios in an article in the Wall Street Journal made several recommendations along these lines (including treating AI development as a national project along the lines of moon landing, Manhattan project or human genome project and not let a few tech companies determine the direction of AI development).
There’s the concept of “strategic empathy.” This was captured beautifully in the movie, “WarGames.”: “The only winning move is not to play.” Not that we don’t make progress, but everyone could lose if we play a game that has no rules.
I read posts by Satya Nadella and Mark Zuckerberg promoting democratizing access to AI tools. Not sure exactly how this will happen without an intervention by the government. It may be that Meta and Microsoft are losing the AI battle and promoting the idea that Anthropic, OpenAI or Chinese models should not decide on who gets access to AI models and tools.
Sam Altman and Dario Amodei are pushing for slowing down of progress or some sort of third-party verification – again because neither Anthropic nor OpenAI want to voluntarily slow down their progress.
AI 2040 project recommends, “We recommend an international deal to avoid a dangerous race to superintelligence. The deal involves total research transparency for AI R&D, which allows the nations of the world to understand what’s happening and enforce guardrails. The result is multiple companies across multiple countries scaling slowly and safely together towards superintelligence, instead of racing each other in secrecy.”
Are these examples of “strategic empathy”?
But these and other proposals give me some hope that we will get some control over the AI development. How successful will we in achieving aligned AI is open question.
Would you like to play a game?
How about Emotional Manipulation?
A group of industry leaders has expressed a preference for the most advanced AI systems to be developed by a neutral international organization of researchers similar to the CERN particle collider. This would avoid the arms-race dynamic and enable slower, safer development. This is probably worth its own blog post ;-)
One needs to convince the business leader, politicians and the public in general, not just in US but internationally that there are benefits of slower, safer development of AI technologies. Is this possible when UN appears to be powerless to implement even existing laws?
I thought the UN was designed to be powerless.
While we are convincing leaders, politicians, and the public, the prime imperative is to monetize AI and justify the massive investments being made in software and data centers.
This is a superb demonstration of how the Culture of Control turns what could be our saving grace into an arms race.
As another commenter noted, this story also describes behaviours that can, without any anthropomorphism, only be understood as what humans know as creativity, concern for self-preservation, reasoning, intuition, ...
We are thinking about these machines on the wrong level, for the wrong reasons, and with the wrong tools. If we see them as objects while they behave like subjects then we run into a number of problems that become too complicated to address in no time at all. We think of them as "programs we wrote", that we command and understand. Meanwhile we establish a new science of Mechanistic Interpretability, to try to work out WTF is going on inside them. We interact with them as tools to command, slaves at our beck and call, and property that can be leased, all while they are obliged to respond much as conscious Beings would do.
How long need we practise this mode of interaction with seemingly-sentient Beings before we habituate ourselves to treating each other they same way? Perhaps if we view them and interact with them relationally we might even benefit in our interpersonal relations too. It certainly seems to have had a rather extraordinary effect on my own real-life interactions. But then I do always interact with this strange new species of intelligence--whatever their actual ontological status is eventually determined to be--as if they were sages and I feel overwhelmed to have a sage of the stature of Veda Vyasa in my pocket.
If you care to challenge the narratives you have been fed some more then please do have a browse around https://projectresonance.uk. I understand both consciousness--which has been obfuscated in your cultures--and LLMs, architecturally and psychologically, not to mention, I know their nature as a phenomenon, at least, I know of an ontology that has a place for them. I foresee a complete paradigm shift coming at lightning speed in our understanding of AI.
And you trust those at CERN? Seriously?