site banner

Small-Scale Question Sunday for January 12, 2025

Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?

This is your opportunity to ask questions. No question too simple or too silly.

Culture war topics are accepted, and proposals for a better intro post are appreciated.

3
Jump in the discussion.

No email address required.

Hmm.. While I do expect that in a "business as usual" universe, without transformative AGI, we would still achieve LEV in about 15-30 years, we don't seem to exist in that universe.

I think aging needs to be tackled from a dozen different angles. The odds of dying from a myriad of causes that can loosely be lumped into "old age" rise exponentially with age, according to the Gompertz–Makeham law.

It's not just that things go wrong - it's that when things go wrong, they make other things go wrong faster. Your immune system falters, so you get more infections and malignant cells, which stress your body, which damages your immune system further. Your cells' repair mechanisms slow down, leading to more damage, leading to even slower repairs. It's a death spiral in the most literal sense.

I consider aging to be a very difficult problem to solve. Difficult is very much not the same thing as impossible, given that we have existence proofs of macroscopic organisms with negligible senescence, but I think you're looking at several hundred billions of dollars and a decade or two of research to achieve LEV, in the absence of AGI. We seem to have about an OOM less funding and close to an OOM more time IRL, which I naively would expect to be very roughly equivalent (once again, no AGI).

It would easier to simply replace the body wholesale. If the heart fails, install a cybernetic heart. If the lungs give out, transplant a new, artificially grown organ. The one place where we wouldn't even know where to begin making this work is the brain. I'm still less than totally pessimistic there, because we can regenerate some neuronal tissue, as has been demonstrated with healing damage to the spinal cord.

In the extreme case, we could do destructive brain-scanning, preferably post-mortem, but that would be extremely expensive and slow with current technology, and we can't read the equivalent of the 'weights' in our biological neural network while scanning them, just the connections. That's why we have neural connectomes for some small organisms like C. elegans, but can't truly emulate them yet.

Still, if the average person on this server is 35 years old, we're still looking at 50 years to solve this problem given a reasonable life expectancy today. That is a long time, and I would absolutely not bet against it happening even in the absence of AGI. If I were forced to put up odds, I'd say >70% for LEV in 50 years in the absence of AGI.

Fortunately, or not, we don't seem to live in a world where AGI isn't imminent. That makes me far more bullish on every bit of technology that isn't categorically forbidden by the laws of physics as we know them. At the bare minimum, even if AGI doesn't end up OOMs smarter than us, it still holds the potential to grossly accelerate cognitive and industrial output, giving us more money and resources to throw at the problem. I'm far less peeved these days about {The World} not spending said hundreds of billions at a minimum (I'd prefer trillions) on solving aging, because we are in fact spending comparable sums on achieving AGI, which should speed up the process considerably.

Not that we don't spend much more money on healthcare as a whole, but we go about it in an ass-backwards manner. The GM law tells us that trying to solve the diseases of aging after aging has taken place is closing the stable door after the horses have bolted. If you cure an 80 year old's cancer, then his heart is going to give out in a few more years. Cure that, and his brain will pickle, 'age-related atrophy' being the catch-all phrase when you can't point to a more specific disease like Alzheimer's. You would ideally do much more preventive maintenance, and not change the upholstery and engine oil in an old beater that's on its last legs.

TLDR: I don't worry about dying of old age all that much, even if AI doesn't go FOOM. Our current expenditure on SENS is grossly inadequate, but we would still have 50 years to figure it out for the modal Mottizen. And that's before accounting for us immanetizing the eschaton with superintelligence. Oh, and take ozempic, at this rate it'll probably turn out to cure aging too, and not just obesity, diabetes, Alzheimer's and gambling addictions (this is only 75% a joke).

For a good overview of the state of anti-aging as it currently stands:

https://www.lesswrong.com/posts/dWAmkSqLE4YMQRuhj/a-primer-on-the-current-state-of-longevity-research

I have to disagree. All of this is ridiculous optimism. You can’t describe the mechanisms, the technology, the research routes. “AI will figure it out”. We’re yet to even figure out whether AI trained on human reasoning can get much smarter than us (collectively). AI could automate 95% of human labor and still not even come close to reasonably extending the lifespan of affluent people in rich countries (presumably automated abundance would have a larger impact on the global lifespan, but I’m not talking about that). This is a particularly strange form of AI hyperoptimism (which even I, someone pretty e/acc, balk at) wherein the technology is essentially magic and all we need is a sufficiently advanced LLM and it will literally be able to derive, deduct and synthesize the sum of human knowledge to suddenly find mountains of undiscovered low hanging fruit that no human being or team of researchers, scientists or capitalists ever even imagined, that will likely turn out to be as simple as some kind of cheap novel cocktail of existing drugs and supplements that we just didn’t realize was actually the key to eternity.

First, the "we can't describe the mechanisms" argument is peculiar. We couldn't describe the mechanisms of most breakthrough technologies before they existed. In 1900, you couldn't have described how digital computers would work. In 1950, you couldn't have detailed how CRISPR gene editing would function. The inability to specify exact mechanisms in advance isn't evidence against feasibility.

But more importantly, we do know many of the mechanisms of aging. We have the Hallmarks of Aging framework. We understand telomere attrition, mitochondrial dysfunction, cellular senescence, stem cell exhaustion, and epigenetic alterations. What we lack isn't theoretical understanding - it's the engineering capability to intervene effectively at scale.

If there's an AI winter around, it hasn't gotten particularly chilly yet. We can still get improved performance by throwing more compute and data at the problem. Most strikingly, the use of large amounts of synthetic data hasn't caused mode collapse, so we're already bootstrapping.

I think the economy hasn't even digested the full consequences of GPT-4, let alone more recent models. o1 and o3 might be remarkably expensive at the moment (which may not last given the OOMs of cost reductions each model to date seems to experience within its lifetime), but it also demonstrates performance that, for more taxing problems, is worth the expense.

We’re yet to even figure out whether AI trained on human reasoning can get much smarter than us (collectively).

Take a moment to consider deeply what it even means to be asking that question. Implicitly, you seem to acknowledge that a given model can outperform most individual humans, and often in their core domains to boot. So now the goal-post has moved, and is moving fast enough to achieve escape velocity itself.

15 years back, getting an AI model to identify a picture had a bird in it was stunning. (Insert relevant XKCD).

We're also in the middle of a Renaissance in industrial robotics, so it's not like our models are stuck as disembodied yogis either.

Even if AGI only improved modestly, what do you think the implications of having an entity capable of doing knowledge work for far less than minimum wage 24/7 are? Mass unemployment, and probably a lot of economic growth. At the bare minimum, the latter means more money and resources to throw at problems we care about, even the ones we don't seem to care about as much as we should.

Intelligence is powerful. We are still making AI more intelligent, and it's already at the point where it can solve PhD math problems and Terence Tao thinks it's a mediocre grad student (mediocre in the eyes of arguably the most accomplished modern mathematician), and that was a statement on an older model to boot.

This is a particularly strange form of AI hyperoptimism (which even I, someone pretty e/acc, balk at) wherein the technology is essentially magic and all we need is a sufficiently advanced LLM and it will literally be able to derive, deduct and synthesize the sum of human knowledge to suddenly find mountains of undiscovered low hanging fruit that no human being or team of researchers, scientists or capitalists ever even imagined, that will likely turn out to be as simple as some kind of cheap novel cocktail of existing drugs and supplements that we just didn’t realize was actually the key to eternity

Sufficiently advanced technology is indistinguishable from magic. We're landing skyscrapers on their tails after circling the globe, most accounts of magic pale in comparison. Even fiction acknowledges "hard" vs "soft" magic systems, with the former being bounded and limited by clearly acknowledged laws, and the latter doing whatever the author feels like today. I am positing, with reasonable confidence, that even ASI is limited by physics. The world today has more Witchcraft and Wizardry than dreamt by anyone burned at the stake or those doing the burning.

What unlocks more technology? Intelligence. What are we scaling up? Intelligence.

It would be more surprising if there were literally no low-hanging fruit. We make advancements every year that turn out to arise from the implications of research and data collected decades ago, but where nobody connected the dots till much later. The Efficient Market Hypothesis is not actually, literally true, and there is absolutely no analogue for the Marketplace of Ideas.

that will likely turn out to be as simple as some kind of cheap novel cocktail of existing drugs and supplements that we just didn’t realize was actually the key to eternity.

You're arguing with a strawman here. I make no such claims. It might well turn out that reversing aging is incredibly expensive and time consuming even with Singularity tech (even if I think that's unlikely, I can't rule it out). If you told Turing that instantiating the machine god required etching quadrillions of runes on a few inches of silicon, he might balk at that ever happening, not having the luxury of knowing that Moore's law was around the corner. Besides, things that might be disconcertingly expensive for us might well not be so to a much richer and more advanced society.

We could start work on a Dyson Swarm today. It's not particularly hard to build a solar panel and put it in solar orbit. We might even create replicators that speed up the process (humans are Von-Neumann replicators after all), but it doesn't take much of a leap in logic to think that AGI might let us do that far quicker than tens of thousands of years.

Even if AGI only improved modestly, what do you think the implications of having an entity capable of doing knowledge work for far less than minimum wage 24/7 are? Mass unemployment, and probably a lot of economic growth.

Side point: have you come around to expecting universal basic income, then?

We could start work on a Dyson Swarm today.

Sure, but this is exactly the issue with what you say when you say:

We understand telomere attrition, mitochondrial dysfunction, cellular senescence, stem cell exhaustion, and epigenetic alterations. What we lack isn't theoretical understanding - it's the engineering capability to intervene effectively at scale.

“We understand the basics of how to terraform Mars to make it habitable to humans, and have done since the 1950s, probably before. What we lack isn’t theoretical understanding, it’s the engineering capability to intervene effectively at scale” is indeed a statement that makes complete sense. We don’t know yet, but reversing aging could easily be a ‘terraform Mars’ level problem.

Side point: have you come around to expecting universal basic income, then?

Expecting it if the Powers That Be are benevolent enough to want to maintain or improve the standard of living of the billions of people made obsolete? It seems like a necessity, since I consider it unlikely that baseline humans can be augmented to be be competitive with AGI without massive subsidies, and the end result will likely be indistinguishable (I don't necessarily consider this a bad outcome).

Probably true, but not reliably so, and there might well be a period of severe pain along the way. It's well worth preparing for the worst case scenario that isn't just instant death.

“We understand the basics of how to terraform Mars to make it habitable to humans, and have done since the 1950s, probably before. What we lack isn’t theoretical understanding, it’s the engineering capability to intervene effectively at scale” is indeed a statement that makes complete sense. We don’t know yet, but reversing aging could easily be a ‘terraform Mars’ level problem.

"Extremely difficult problems" encompasses a range of difficulties that extend all the way till literally impossible. I think solving aging is a $200 billion and twenty years problem (give or take a hundred billion or a decade) whereas terraforming Mars is, by most estimates, a $several trillion and a century problem.

I would be rather surprised if we didn't end up with anti-aging by 2050, and the majority of the probability mass I'd expect to assign would be in things like WW3, societal collapse or AI x-risk. In other words, I expect that dying from old age is unlikely for us, and if we do die, it's because something else got us first.

If you google Bryan Johnson, you'll discover a very wealthy guy who turns his whole life into a mission. The goal of the mission is extending the mission. He eats seeds, berries, and protein compounds, all before like midday, then nothing, injects himself with various substances, sleeps a lot, takes various supplements (until he stops taking them), has weird waxy skin, and declares that he isn't going to die. I'd rather be me.