Peter Thiel – Artificial General Intelligence Isn't Happening | Hacker News Reader

Peter Thiel – Artificial General Intelligence Isn't Happening

mindmatters.ai

44 pointsmindcrime45 comments

> Those things, he thinks, are uncertain. We should worry about what’s happening now in everyday time, to which, in his view, too few are paying heed: The growth of total AI-based surveillance and the disappearance of privacy.

That’s real rich coming from the founder of Palantir

Don't hate the player hate the game I guess. If Palantir didn't build the tools someone else like Clearview AI would.

Anyone can now buy motion activated cameras with enough storage for months of video, some even battery powered for <$100 each. A couple of those pointed at your residence is enough to remove a ton of privacy. Now imagine some company/blockchain rewards ppl for uploading those motion caps.

The argument is a lazy abdication of responsibility. Also I'm like 90% sure that Hoan Ton-That, Clearview's founder gave the same exact argument for why he built it

This is one place where government should step in and we should add regulations if enough people agree. It's ridiculous to expect every company to choose not to pursue this as part of some unspoken morality.

> If Palantir didn't build the tools someone else like Clearview AI would.

Thiel is an investor in Clearview.

Don't worry, Thiel invested sizeably in Clearview too. I lost it hearing how he denounced AI as "basically Communist" while actively building and funding it. Did the money he earned damage his brain into forgetting how he made it?

If you believe in the power of a weapon/tech and others are ignoring it, isn't the smart move to own & control it?

Yeah, I think this past decade people really conflated the deep learning breakthroughs with imminent actual AI. In reality what was invented here was advanced pattern recognition, not anything close to actual reasoning.

I don't predict anything good in the economy once people realize this. A lot of "AI" projects out there seem to me like they will eventually require some reasoning that is impossible to implement.

Perhaps this new AI winter will cause a crash, and the powers will use their newfound pattern recognition tools to make sure things don't get out of hand.

My perception is that AGI may be quite possible eventually, but pattern recognition is so powerful that it may make the question moot.

Like, a machine that can move like a pack animal (e.g. Boston Dynamics) took, what, well over a hundred years longer than automobiles and trains?

I have seen some effort to create reasoning with deep learning. Wonder if it's any good. AFAIK the idea is to bet what is causing changes modify it weight and see if you get better results. This way network learns it's dependencies explicitly and can kinda "reason". Anyone in the field could confirm?

Even if AI funding drops to zero that wouldn't cause the real economy to crash. The dollar amounts are tiny in relative terms.

There are still tons of places where even modest deep learning could have a huge impact. The focus in industry seems to be moving to deployment.

Here are four things I've been certain of for my entire life and become more certain of with every passing day:

1. We will never go to Mars (and it is an insult to my intelligence to try to argue otherwise).

2. Computers are never going to gain consciousness.

3. The Multiverse is the stupidist load of horse*h&t to have ever entered anyone's mind. Only unserious thinkers take the idea of the Multiverse seriously.

4. There is no materialist explanation for the existence of life, and the existence of life is the only interesting question in existence since it forms the foundation for the existence of every other question.

4 is an interesting point. My curious brain often wonders over it. Frustrating as it is, you're probably right. Can you even imagine what would happen if somehow we came to really, foolproof know the answer? I imagine an absolute calamity as religion falters(assuming it wasn't a religious answer, that is), people lack will, perhaps crime and suicides rise?

As we've seen these last couple of years, people presented with foolproof evidence will just continue believing whatever they believe.

We have foolproof evidence that life could not possibly have arisen through random chance but apparently nothing can dislodge the narrative that it must have.

I can only speak from the Christian perspective. The only people to have their beliefs challenged would be fundamentalists, which represent a minority of Christians. The Bible says God is responsible for creating life, it doesn't give detail on how He did it. Christians have played important roles in understanding the "how" of things like the Big Bang and genetics.

Thanks for your perspective. It's an interesting question/problem. Even if I'm to accept a big bang or anything else, it just leads to more questions. Like who created the ingredients that could cause said bang, etc. I know it's often explained away by science, but it seems one could say 'well, what created those components' for infinity. So any suitable answer would have to start at the very very beginning, and in a way a layperson could understand. No assumptions, no handwavy 'it just did'. Same if it were God/a god. How did that come to be?

I think OP is right in that it will never realistically happen.

I would change 4. to “there is no materialistic explanation for existence” since we have a pretty accurate (as far as we can tell) explanation for how we got to life once stars started exploding and forming the necessary elements for life. Maybe I’m quibbling though.

We do not have a "pretty accurate" explanation for how we got to life once stars started exploding. In fact, we have no idea whatsoever and the more we learn the more infinitely intractable the problem appears. We're moving further away from an answer as our understanding grows, not closer.

I should have been clearer - the hypothesis of abiogenesis provides an accurate explanation for the origins of life from non-life and is accepted by the vast majority if non-religious biologists as the most likely explanation for how life started… and, for what it’s worth, there is absolutely no evidence for any other explanation.

The Miller-Urey experiment demonstrated that organic material can result from inorganic material in 1952; the abiogenesis hypothesis has only grown stronger since then, not weaker. https://en.m.wikipedia.org/wiki/Miller–Urey_experiment

Amen. Funny how Peter Thiel parlays something that was obvious to many, if not most, people outside the SV echo chamber into some sort of new insight…

SHOULD we go to Mars? Where do you stand on that?

This entire list is wildly wrong.

1) Humans will go to Mars if for literally no other reason that “to do it first”

2) Is there something special about meat computers called brains or your definition of “consciousness” that precludes this?

3) The multiverse is a mathematical solution to quantum mechanics, what’s your serious thinker alternative?

4) Depends on what question you’re asking, but instead you’re blithely vague. Questions of “why” are not in the realm of empiricism, but questions of “how” might be.

Multiverse was a typo. They meant "metaverse".

1) "Go to" as in "visit" like we did for the Moon or "go to" as in stay there?

3) No, the multiverse in an "interpretation" of QM, it isn't a solution, it brings no additional facts. So per Occam razor..

>2) Is there something special about meat computers called brains or your definition of “consciousness” that precludes this?

This statement contains so many false assumptions.

For starters, there is much more to consciousness than just the brain... when has there ever been a brain in isolation that was conscious?

And we also don't know how the brain (and the rest of the organism/environment) create consciousness.

I think we might "go to Mars" in the same way that we went to the moon. But yeah, there's nothing good out there for humans.

This. Unless there is some economic reason to go to Mars, it will remain a place of scientific curiosity. Just like Antarctica, we will not be colonising Mars.

I think robots are a better use of space exploration money. The cost of one human mission could send swarms of robots, with way more science getting done.

Why won't we go to Mars (ever)? You may be sceptical of it happening in the next 100 years but very confident in it happening in the next 500 years, it is a very strong claim to argue that it will never happen in the broader context of increasing global wealth and massive improvements in space/energy technologies.

We may place humans on Mars for periods of time, perhaps even lengthy periods but the idea of an alternate earth is in fact ludicrous. We seem to have a hard time just managing the planet we're on.

Why is it ludicrous - do you have a specific issue that you object too? How is living on Mars very different from e.g. the international space station or conditions in a nuclear sub in terms of human habitats? The other bits seem relatively easier than the problem of getting people to the planet. In general I think arguments of the form "it can't be done soon because of resource constraints" are very weak - there are plenty of examples of people making such arguments and being proven wrong a couple of decades later.

We've already been to Mars

I guess they mean humans on mars?

1) dunno, I hope we will

2) if we can throughly understand our own brains we can perhaps emulate them in various other media?

3) you mean many world's hypothesis? I guess it's not very relevant for most people

4) I think there are materialist explanations for life, they are getting more convincing I think. As in we could eventually model scenarios in which chemical processes lead to self perpetuating processes that resemble primitive life and can evolve into life

> Computers are never going to gain consciousness

The problem with "consciousness" is that is really hard to define. Until we can measure for it, I agree that computers will never be it.

Personally I think we will have machines that are very smart, that can perform many of the mental tasks that a human can, but will be almost alien in the way they think. But that's ok. They should remain tools that allow us to solve hard problems.

The problem as I see, at least in the medium term, is that pattern matching, no matter how advanced is still just a simulation of intelligence, not 'real' intelligence. What is missing is the ability to reason, when the facts (data) are inconclusive, missing or wrong. Also any AGI will need a very complete abductive reasoning system, I just don't think we have the tools and understanding yet to build AGI, and certainly not if the next generation of AI practitioners continue to think Python is the way to build an 'intelligence'

The problem as I see, at least in the medium term, is that pattern matching, no matter how advanced is still just a simulation of intelligence, not 'real' intelligence.

I don't know that we can take that as a given. The "it's just pattern matching" argument is pretty old, but I'm not sure it's ever been shown conclusively that all of human intelligence can't be reduced to some form of pattern matching.

Now, in terms of "implement something on a digital computer", so far our approaches to "reasoning" (using symbols and formal logic) seem very, VERY different from our approaches to "pattern matching", and maybe in that domain such a reduction isn't feasible. Or maybe it is. I don't think anybody really knows.

My own suspicion (and that's all that it is) is that in "digital computer" domain, we should use formal logic where it makes sense, pattern matching ANN's where it makes sense, and hybridize the systems. But that approach brings in some hard challenges of its own.

And FWIW, experts like Geoffrey Hinton have argued against the need for any such hybridization and believe that neural networks (of some form... maybe one we haven't invented yet, who knows?) should be enough for human-level intelligence.

I just don't think we have the tools and understanding yet to build AGI

I would agree with that. We don't yet have the required understanding (whether or not we have the needed tools is one I'm more 50/50 on), but I do think we'll get there eventually. It's just hard to predict if "eventually" means "tomorrow" or "37,000 years from now".

It doesn’t appear that Thiel is saying AGI won’t happen. Rather he observes that people are talking about it much less now than they were (presumably because Strong AI is still far off); and furthermore that it’s not that important to him as a startup investor because of the massive scale of data and compute resources required. Implicitly, I take him as acknowledging that the multi-trillion dollar surveillance AI companies have the scale, the reach, and the time to make gradual progress towards AGI.

These are reasonable notions, even weak AI is sufficient to potentially upend the world economic and social order. Frankly, if true AGI were to materialize within the next decade, we would be ill-equipped to cope with the repercussions.

I do, however, believe that many if not most AI pessimists are disconnected from the progress being made on a wide range of critical fronts.

Why do people listen to this clown? He got lucky with the Meta investment and hacking Roth IRAs. But his ideas are so retro.

I don't necessarily agree with Peter Thiel but overall his record of accurate predictions is higher than most other pundits and business leaders. He's plugged into a lot of high level conversations that most of us don't have access to.

Definitely agree with the parts about Elon Musk and Larry Page not talking about AI anymore.

My perception of OpenAI is that it's running on the fumes of Elon Musk circa 2015. It was launched in that year or so with a lot of hype, much of it due to Elon's name being attached. Then he lost interest, and it's been getting less and less ambitious as time goes on. That's the opposite of what you'd expect if they were on the path to AGI.

My (somewhat first hand) perception of self-driving cars is that the industry is running on the fumes of Larry Page circa 2006-2010. Larry attended DARPA self-driving challenges, and spun up a self-driving project at Google. Through incredible PR they managed to convince the entire auto industry to jump on board, investing tons of money for acquisitions like Cruise.

All self-driving projects have also been getting less ambitious over time. I noted a few years ago that in 2016 Chris Urmson already told people to adjust their expectations -- that it would take 25 years to roll out. Still in 2018 people were repeating claims that Page and Urmson made back in ~2010, and some are still doing that today.

I thought self driving cars were already on the road in various parts of the country, doing test rides and all that.

Relative to the expectations and predictions made in 2006, 2010, and 2015, it's a disappointment. Relative to what executives expected when they spent billions of dollars on it, following the lead of Google, it's a disappointment.

The people who were talking about it back then like Larry Page, Sebastian Thrun, and Chris Urmson no longer really talk about it.

If you only started paying attention to self driving cars in 2018 or so, it might be very exciting!

Reply on news.ycombinator.com