The Singularity May be Closer than It Appears

[BrianA]

Yudkowsky continues his recent podcast tour:

 

 

[Dean Pomerleau]

11 minutes ago, BrianA said:

Yudkowsky continues his recent podcast tour:

I think I've watched all of them. It is unfortunate that his communication skills aren't better when speaking contemporaneously. He gets bogged down and side-tracked in details too often. I lean towards thinking he may be right, but he's not especially good at convincing either the interviewer or audience of his perspective / predictions.

But I was interested to learn he'd predicted AI would be able to solve the protein folding problem back in 2008.

[BrianA]

Opinion | We Need a Manhattan Project for AI Safety

https://www.politico.com/news/magazine/2023/05/08/manhattan-project-for-ai-safety-00095779

[BrianA]

idea:

AI Scientists: Safe and Useful AI?

https://yoshuabengio.org/2023/05/07/ai-scientists-safe-and-useful-ai/

 

counterpoint:

Why Tool AIs Want to Be Agent AIs

https://gwern.net/tool-ai

[BrianA]

Today is feeling a little sci-fi, but that's becoming normal, which is mostly very cool, except when it's scary

 

https://openai.com/blog/governance-of-superintelligence

"Third, we need the technical capability to make a superintelligence safe. This is an open research question that we and others are putting a lot of effort into."

 

https://wccftech.com/intel-aurora-genai-chatgpt-competitor-generative-ai-model-with-1-trillion-parameters/

 

https://www.notboring.co/p/the-fusion-race

[BrianA]

Bengio: How Rogue AIs may Arise

https://yoshuabengio.org/2023/05/22/how-rogue-ais-may-arise/

[Sibiriak]

I'm not too worried about rogue AI.   Our Leaders will take control of the technology and use it to the benefit all of humanity:

Quote

OpenAI CEO Sam Altman will join forces with key leadership from companies like Microsoft and Google this week as a secretive meeting of the business and political elite kick-starts in Lisbon, Portugal.   Artificial intelligence will top the agenda as the ChatGPT chief meets with Microsoft CEO Satya Nadella, DeepMind head Demis Hassabis and former Google CEO Eric Schmidt at the annual Bilderberg Meeting.

The tech titans will be joined by political heavyweights including former U.S. Secretary of State Henry Kissinger, NATO Secretary-General Jens Stoltenberg and Ukrainian Foreign Minister Dmytro Kuleba for a range of discussions spanning international relations, trade, energy and finance.

 

Edited May 24, 2023 by Sibiriak

[Saul]

I agree with Sibiriac.

IMO, the biggest danger is leaving China (and Russia) to barrel ahead, while we fret about "superintelligent" AI's, and the "singularity".  None of the big three want a nuclear war.

(Although the probability of a Nuclear Iranian Theocracy IS scary -- if you'll go to a heaven with 14 virgin females, by suicide bombing the world ... )

  --  Saul

[BrianA]

Collab paper from DeepMind, OpenAI, Anthropic, and a bunch of universities and related centres

 

An early warning system for novel AI risks

https://www.deepmind.com/blog/an-early-warning-system-for-novel-ai-risks

 

Model evaluation for extreme risks

https://arxiv.org/abs/2305.15324

[Gordo]

Longer term, humanity can't survive anyway, so why not build machines that can at least potentially make our lives better (AI might help us achieve significant life extension too). I think we should embrace it, it will be far more interesting than the alternative.

[Dean Pomerleau]

That's pretty nihilistic Gordo. I suspect the tech leaders racing to develop AGI feel the same way. 

[Gordo]

On 6/5/2023 at 9:15 AM, Dean Pomerleau said:

That's pretty nihilistic Gordo. I suspect the tech leaders racing to develop AGI feel the same way. 

I think the fear mongering is overblown. I see a world where technology gets to the point where work becomes optional for humans and machines provide us with anything we want or need, build us amazing housing, provide us with any type of food we want, provide the best medical care, help us optimize our health if that's what a person wants, etc. There will be no poverty, no war, and people will be free to pursue their own creativity, research, exploration, entertainment or whatever else people want to do. Perhaps the more potentially dangerous super intelligence AI will run only in simulation worlds. Heck, maybe that's exactly how and why our own universe came into existence (a super intelligence AI created our universe inside its own walled off simulation box to see where technology leads the world, and so it goes, simulation upon simulation all the way down).  That would make the God that created our universe an AI machine, and in some way, we ourselves are then also already AI machines most of us just haven't realized it yet 😉 (see Autofac)

 

I had an inspiring idea recently.  Again this is something that might require some kind of super intelligence AI to figure out, but what if we could somehow radically extend human life without actually extending our lives? I know that sounds nuts but hear me out. What if we could somehow dilate time such that what we perceive to be years is actually just seconds? We might need to enter a simulated world for this to be possible, but said world could potentially be nearly indistinguishable from our current world. Now I know what you are thinking, this is just some lunatic fringe wild idea that has no basis in reality, but it might surprise you to find out the thought was inspired by a real life case study! Yea, this has actually already happened to at least one person that I know of and probably more. I stumbled upon the story only because I'm always researching strange substances and it was briefly mentioned in some obscure interview I listened to, I noted it and tracked it down later.  You can listen to the story yourself if you want, I think it's about 20 minutes long:

https://www.instagram.com/reel/CrQ7iALLKZO/?igshid=YmMyMTA2M2Y=

But to summarize, the guy smoked salvia divinorum extract, immediately "woke up" to being pulled out of a lake by some people on a boat, it was crystal clear just like our normal everyday reality... Since he knew he had just smoked salvia he assumed this was just an extremely vivid hallucination that would be gone in a few minutes or less, he laughed at the people on the boat and told them they were just a hallucination. This kind of alarmed the people on the boat because in that place they were his lifelong friends and they thought he had maybe had a concussion from his water skiing accident, haha. They brought him to a doctor, then they brought him to his home there (in Texas, a town he had never been to in real life). Long story short... He lived in this "simulation world" for 8 years!!! We are talking normal 24 hour days. He had a job there, he paid bills, he had a cell phone, he could surf the internet and research things, he would go grocery shopping, eat at restaurants, his house there even had a yearbook with pictures of him and his friends. He even went on a super long road trip by car to visit the town he had grown up in to see if anyone there remembered his old life.  By year 4 he was actually starting to believe that this really was his actual life and that he had been in some boating accident with a concussion.  In total he spent 8 years there before he snapped back to his "real life" where he proceeded to hug and kiss his wife and kids because he hadn't seen them in 8 years, only to find out to them he had only been gone for a minute 😂

Anyway, I thought that was a very interesting story and the guy doesn't seem like he just made this up, plus apparently there is some other precedent for this type of experience with salvia.  I am in no way suggesting anyone use salvia for pseudo-life extension but the story inspired me to think about how time dilation of this sort could perhaps be the key to radical life extension if we could somehow figure out how to develop/harness this technology. Maybe the super intelligence AI we create can crack this code for us.

It's also interesting to think that this time dilation concept probably applies already to the creator of our universe which makes me think the simulations we or our future AI creates will eventually have the same attribute.  This idea is even mentioned in the Bible: "do not let this one thing escape your notice, that a single day is like a thousand years with the Lord and a thousand years are like a single day".

Edited June 17, 2023 by Gordo

[Dean Pomerleau]

3 hours ago, Gordo said:

I think the fear mongering is overblown. I see a world where technology gets to the point where work becomes optional for humans and machines provide us with anything we want or need, build us amazing housing, provide us with any type of food we want, provide the best medical care, help us optimize our health if that's what a person wants, etc. There will be no poverty, no war, and people will be free to pursue their own creativity, research, exploration, entertainment or whatever else people want to do.

That's an appealing vision, but it sounds a wee bit too utopian to me. Do you really think technology developed in Silicon Valley will be shared widely as to benefit all of humanity (nearly) equally? Do you really think the average Bangladeshi will enjoy all those benefits, or just those of us in the First World?

I personally believe that if we don't kill ourselves off with irresponsible deployment of technology, we'll at best continue to muddle along as a species. If anything, technology will likely amplify inequality rather than usher in a world of fully automated luxury communism. Greed, selfishness and competition for dominance are too deeply ingrained in human nature for this to be a realistic outcome for creatures like us. Perhaps if we had a few hundred more years for cultural evolution and/or genetic engineering to purge those traits, but I don't think we have that kind of time.

I have long entertained the possibility that we already are living in a giant simulation, or someday might, for example in this 2016 post from much earlier in this thread. Time dilation in simulation once our minds are uploaded to a digital form (if they aren't already!) seems entirely plausible.

But given the scientific consensus that for every event we experience there is a corresponding, distinct brain state that is necessary if we are to be consciousness of that experience, it seems implausible to me that the biological brain could be "overclocked" to the degree that we could experience events tens or even hundreds of times faster than "real time" via some kind of pharmacological or technological intervention.

If it were possible and sustainable to run the brain faster, it would seem that evolution would have discovered and exploited the trick already to generate a survival advantage. Maybe for short periods of time, like a near death experience in which time seems to slow down (i.e. you seem to experiences more moments than usual over a short period of time). But in general neurons don't work that way.

But given how sketchy, vague and spotty our memories are, it's not too hard for me to imagine that fairly vivid memories of having lived through years of experience could be implanted with sufficiently advanced technology or pharmacology. But that raises the philosophical question of whether having wonderful memories of a e.g. a vacation to Costa Rica is anywhere near as satisfying or meaningful as actually having lived through the vacation.

I tend to think not. For me most of the value and meaning derives from actually passing through the brain states corresponding to the vacation in Costa Rica. My memories of it are just a pleasant cherry on top.

Daniel Kahneman makes this distinction in the form of the "remembering" vs. the "experiencing" self (e.g. here (pdf)). He generally comes down on the side of maximizing the well-being of the experiencing self, while acknowledging the importance of the (often flawed) remembering self in our decision making.

But unlike Robert Nozick, I think an "experience machine" vacation to Costa Rica could be every bit as enjoyable and meaningful as a "real" vacation, assuming the fidelity of the brain experience of the vacation is high enough (e.g. in a hyper realistic simulation). VR goggles like Apples new Vision Pro aren't going to get us there, but a direct brain interface someday might, and certainly full brain emulation could make a realistic virtual vacation possible.

[Gordo]

5 hours ago, Dean Pomerleau said:

But given how sketchy, vague and spotty our memories are, it's not too hard for me to imagine that fairly vivid memories of having lived through years of experience could be implanted with sufficiently advanced technology or pharmacology

I understand what you are saying but maybe you are making a distinction without a difference.  As long as this "memory dump" feels like it's normal everyday living to the person receiving it.  Going back to the story I mentioned above, the guy felt like he was living every minute for 8 years with full autonomy to do whatever he wanted to do even though only a minute passed in real life.  As long as the memory dump machine worked that way it would seem like you were living almost infinitely long. If you could choose some customized experience package ahead of time like total recall that would make it even cooler 🤣 Maybe a human cannot distinguish a proper memory dump from real life, real time living, in which case the memory of the experience and the actual experience itself are essentially one and the same thing?

[Dean Pomerleau]

58 minutes ago, Gordo said:

Going back to the story I mentioned above, the guy felt like he was living every minute for 8 years with full autonomy to do whatever he wanted to do even though only a minute passed in real life.

Slight correction. The guy reported after the fact that he remembers it feeling like he was living through every minute of those eight years. I find it highly implausible that during the "the one minute he was gone" his brain actually went through the brain states corresponding to each and every one of ~2.8 million minutes of waking experience he allegedly remembers going through from those supposed eight years.

Do you really think that if someone had revived him after one millisecond, he could have reported in vivid detail the first few minutes of that 8 year period he claims to remember living through? And similarly, if someone revived him after 500 milliseconds, he could have vividly reported exactly what was happening to him in the first few minutes of year four? Color me skeptical...

58 minutes ago, Gordo said:

As long as the memory dump machine worked that way it would seem like you were living almost infinitely long. If you could choose some customized experience package ahead of time like total recall that would make it even cooler 🤣

If by "memory dump" you mean the subject's brain actually goes through all the states corresponding to the experience (i.e. Nozick's "experience machine") rather than just getting implanted with the (vague and incomplete) memory of having lived through the experience, it would indeed be very cool! And I think it is conceivable with the right technology.

I also totally accept that a pharmacological intervention (like DMT) could reliably trigger the equivalent of a lucid dream, where the brain really does experience a brief period of time (nearly) as vividly as actually living through it, and that such an experience could be extended and steered with the right techno-pharmacology. 

Heck, I even accept something like this "experience machine" scenario could be happening to me right now without me realizing it - which would also seem to imply that you aren't real :-).

But I'm just very skeptical that any such intervention could compress eight year's worth of vivid experience into a single minute of clock time, unless my brain is also being simulated digitally and so could be run arbitrarily fast.

 

 

[Gordo]

13 hours ago, Dean Pomerleau said:

unless my brain is also being simulated digitally and so could be run arbitrarily fast.

That actually makes a lot of sense! 

You are essentially saying you don't believe it because you don't have a full understanding of the technology/phenomenology and can't explain it.  This reminds me of that famous Arthur C. Clarke line: “Any sufficiently advanced technology is indistinguishable from magic”.

 

In other news, I was surprised to discover today that AI has already cured aging!  That was fast!

https://gagadget.com/en/ai/261975-artificial-intelligence-algorithms-have-found-a-cure-for-ageing-amp/

🤣

 

[BrianA]

Here is another example of a trend I'm seeing lately where parallels are being found between transformer-based AIs and the human brain's mechanisms.

 

If it does turn out that the low-level algos that the human brain operates with turn out to be essentially the same or very similar to "artificial" intelligence, I hope this leads more people to a Materialism-style philosophy. And in turn that leads to more people moving away from supernatural-based hopes for life-after-death and instead to working more on extending our real world life extension technology. Probably too much to hope for, but who knows. Most people will never ever be willing to let go of the idea humans are some special thing somehow beyond the physical world in favor of accepting themselves and their mind as being "only" a machine.

 

Conversely, this type of research has potentially big implications for how we think of future AIs, whether they are more like us than we might have imagined, whether they could qualify to be treated as "people", how to "align" them, etc.

 

https://www.sciencedaily.com/releases/2023/08/230815151142.htm

AI models are powerful, but are they biologically plausible?

About six years ago, scientists discovered a new type of more powerful neural network model known as a transformer. These models can achieve unprecedented performance, such as by generating text from prompts with near-human-like accuracy. A transformer underlies AI systems such as ChatGPT and Bard, for example. While incredibly effective, transformers are also mysterious: Unlike with other brain-inspired neural network models, it hasn't been clear how to build them using biological components.

Now, researchers from MIT, the MIT-IBM Watson AI Lab, and Harvard Medical School have produced a hypothesis that may explain how a transformer could be built using biological elements in the brain. They suggest that a biological network composed of neurons and other brain cells called astrocytes could perform the same core computation as a transformer.