Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Licensing is neither feasible nor effective for addressing AI risks

aisnakeoil.substack.com

Post reply on HN
Print view View original

111–120 of 161 posts

Re: Licensing is neither feasible nor effective for addressing AI risks

#111

Post by rl3 »

SanderNL wrote: post #110
PeterisP wrote: post #91

Earlier quoted context omitted.

The way I see it, it's clear that human-level intelligence can be achieved with hardware that's toaster-sized and consumes 100 watts, as demonstrated by our brains. Obviously there is some minimum requirements and limitations, but they aren't huge, there are no physical or info-theoretical limits that superhuman intelligence must require a megawatt-sized compute cluster and all the data on the internet (which obvious…

Using this logic, flexible, self-healing, marathon running, juggling and childbearing robots that run on the occasional pizza are just around the corner, because, nature. It might us a thousand years to get anywhere close? I don’t see the good arguments for all of this happening soon.

It'd be interesting if we could calculate the amount of power consumed in aggregate by evolutionary processes over millions of years.

Unfortunately we could probably optimize it, a lot.

Re: Licensing is neither feasible nor effective for addressing AI risks

#112

Post by discardedrefuse »

jeremyjh wrote: post #105
discardedrefuse wrote: post #103

Interesting, but ultimately irrelevant, discussion. AI licensing is just evil megacorp's attempt to strangle the nascent AI industry, because they are quickly losing dominance, and maybe even relevance. Every logical argument you throw falls on deaf ears. Their reasons for pushing licensing are not based on good faith. It was only 6 moths ago that Microsoft invested $10,000,000,000 into OpenAI. At the time, their pro…

They didn’t pay 10B for the model, but for the market potential of the company creating it, and their human capital. I think calling this a bad bet is … premature to say the least.

Obviously they paid for the whole enchilada. My point is that last year only the richest megacorps could afford to run these models. The industry was self-selecting in their favor. I think no one predicted just how fast we would blow through those hardware barriers. And now they're scrambling to throw up new, regulatory barriers to maintain market share.

OpenAI's behavior is nothing more than rent-seeking. And I guess I'm just kinda tired of seeing all this discussion on the topic of licensing like its anything other.

Re: Licensing is neither feasible nor effective for addressing AI risks

#113

Post by wizzwizz4 »

circuit10 wrote: post #77
wizzwizz4 wrote: post #74

Earlier quoted context omitted.

> but it’s really difficult to say “this group of neurons does this specific thing and that’s why this output was produced”, That's because that's not how brains work. > though OpenAI did make some progress in getting GPT-4 to explain what each neuron in GPT-2 is correlated to The work contained novel-to-me, somewhat impressive accomplishments, but this presentation of it was pure hype. They could have done the same…

When I said about a group of neurons I was talking about LLMs, but some of the same ideas probably apply. Yes, it’s probably not as simple as that, and that’s why we can’t understand them. I think they just used GPT-4 to help automate it on a large scale, which could be important to help understand the whole system especially for larger models

> I think they just used GPT-4 to help automate it on a large scale,

No, they used it as a crude description language for Solomonoff–Kolmogorov–Chaitin complexity analysis. They could have used a proper description language, and got more penetrable results – and it would've raised questions about the choice of description language, and perhaps have led to further research on the nature of conceptual embeddings. Instead, they used GPT-4 to make the description language "English" (but not really – since GPT-4 doesn't interpret it the same way as humans do), and it's unclear how much that has affected the results.

Here's the paper, if you want to read it again: https://openaipublic.blob.core.windows.net/neuron-explainer/... Some excellent ideas, but implemented ridiculously. It's a puff piece for GPT-4, the Universal Hammer. They claim that "language models" can do this explaining, but the paper only really shows that the authors can do explaining (which they're pretty good at, mind: it's still an entertaining read).

Re: Licensing is neither feasible nor effective for addressing AI risks

#114

Post by verall »

usaar333 wrote: post #82
hn_throwaway_99 wrote: post #64

Earlier quoted context omitted.

Definitely agree. I would summarize it a bit differently, but when people talk about AI dangers they are usually talking about 1 of 4 different things: 1. AI eventually takes control and destroys humans (i.e. the Skynet concern). 2. AI further ingrains already existing societal biases (sexism, racism, etc.) to the detriment of things like fair employment, fair judicial proceedings, etc. 3. AI makes large swaths of hu…

#1 has high focus more due to impact than high probability. #2 doesn't seem talked about much at this point and seems to be pivoting more to #3. #2 never had much of a compelling argument given auditability. #3 gets mainly ignored due to Luddite assumptions driving it. I'm dubious myself over the short term - humans will have absolute advantage in many fields for a long time (especially with robotics lagging and bein…

> #2 never had much of a compelling argument given auditability.

There are already AI systems which generate "scores" based on likelyhood to commit another crime (used in parole cases) and likelyhood to be a "good tenant" (used by landlords).

Don't underestimate the difficulty to fix something when the people in charge of that thing do not want it fixed.

Re: Licensing is neither feasible nor effective for addressing AI risks

#115

Post by williamtrask »

PeterisP wrote: post #91
dwallin wrote: post #38

Earlier quoted context omitted.

The theories all inevitably rely on assumptions that are essentially the equivalence of spherical cows in a frictionless universe. All evidence is that costs for intelligence likely scale superlinearly. Each increase in intelligence capability requires substantially more resources (Computing power, training data, electricity, hardware, time, etc). Being smart doesn’t just directly result in these becoming available w…

The way I see it, it's clear that human-level intelligence can be achieved with hardware that's toaster-sized and consumes 100 watts, as demonstrated by our brains. Obviously there is some minimum requirements and limitations, but they aren't huge, there are no physical or info-theoretical limits that superhuman intelligence must require a megawatt-sized compute cluster and all the data on the internet (which obvious…

> there are no physical or info-theoretical limits that superhuman intelligence must require a megawatt-sized compute cluster and all the data on the internet (which obviously no human could ever see).

Much of your "intelligence" is a function of natural selection. This is billions of years X bajillions of creatures in parallel, each processing tons of data at a crazy fast sampling rate in an insanely large/expensive environment (the real world). Humanity's algorithm is evolution moreso than the brain. Humans learn for a little while, start unlearning, and then die — which is an important inner for loop in the overall learning process.

Taken together, there is some evidence to suggest that superhuman intelligence must require megawatt-sized compute cluster and all the data on the internet (and a lot... LOT more)

Re: Licensing is neither feasible nor effective for addressing AI risks

#116

Post by theptip »

dwallin wrote: post #38
circuit10 wrote: post #23

Earlier quoted context omitted.

> There is no evidence that a runaway artificial intelligence is even plausible Really? There is a lot of theory behind why this is likely to happen and if you want a real example there is a similar existing scenario we can look at, which is how humans have gone through an exponential runaway explosion in capabilities in the last few hundred years because of being more intelligent than other species and being able to…

The theories all inevitably rely on assumptions that are essentially the equivalence of spherical cows in a frictionless universe. All evidence is that costs for intelligence likely scale superlinearly. Each increase in intelligence capability requires substantially more resources (Computing power, training data, electricity, hardware, time, etc). Being smart doesn’t just directly result in these becoming available w…

You seem to be arguing against a fast takeoff, which I happen to agree is unlikely, but nothing you say here disproves the possibility of a slower takeoff over multiple years.

> It also means that a super intelligent AI has disincentives to build better AIs.

I think this argument is extremely weak. It makes two obviously fallacious assumptions:

First, we simply have no idea how these new minds will opine on theory of mind questions like the Ship of Theseus. There are humans who would think that booting up a “Me++” mind and turning themselves off would not mean they are dying. So obviously some potential AI minds wouldn’t care either. Whether specific future minds care is a question of facts but you cannot somehow logically disapprove either possible state.

Second, you are assuming that there is no “online upgrade” whereby an AGI takes a small part of itself offline without ceasing its thread of consciousness. Again, logic cannot disprove this possibility ahead of time.

Re: Licensing is neither feasible nor effective for addressing AI risks

#117

Post by AnthonyMouse »

dwallin wrote: post #38
circuit10 wrote: post #23

Earlier quoted context omitted.

> There is no evidence that a runaway artificial intelligence is even plausible Really? There is a lot of theory behind why this is likely to happen and if you want a real example there is a similar existing scenario we can look at, which is how humans have gone through an exponential runaway explosion in capabilities in the last few hundred years because of being more intelligent than other species and being able to…

The theories all inevitably rely on assumptions that are essentially the equivalence of spherical cows in a frictionless universe. All evidence is that costs for intelligence likely scale superlinearly. Each increase in intelligence capability requires substantially more resources (Computing power, training data, electricity, hardware, time, etc). Being smart doesn’t just directly result in these becoming available w…

> Each increase in intelligence capability requires substantially more resources (Computing power, training data, electricity, hardware, time, etc). Being smart doesn’t just directly result in these becoming available with no limit. Any significant attempts to increase the availability of these to a level that mattered would almost certainly draw attention.

We know that "intelligence" can devise software optimizations and higher efficiency computing hardware, because humans do it.

Now suppose we had machines that could do it. Not any better, just the same. But for $10,000 in computing resources per year instead of $200,000 in salary and benefits. Then we would expect 20 years worth of progress in one year, wouldn't we? Spend the same money and get 20x more advancement.

Or we could say 20 months worth of advancement in one month.

With the current human efforts we've been getting about double the computing power every 18 months, and the most recent ones come in terms of performance per watt, so then that would double in less than a month.

For the first month.

After which we'd have computers with twice the performance per watt, so it would double in less than two weeks.

You're quickly going to hit real bottlenecks. Maybe shortly after this happens we can devise hardware which is twice as fast as the hardware we had one second ago every second, but we can't manufacture it that fast.

With a true exponential curve you would have a singularity. Put that aside. What happens if we "only" get a thousands years worth of advancement in one year?

Re: Licensing is neither feasible nor effective for addressing AI risks

#118

Post by hackinthebochs »

Al0neStar wrote: post #72
circuit10 wrote: post #35

Earlier quoted context omitted.

The fact that something has been covered in sci-fi doesn’t mean that it can’t happen. https://en.m.wikipedia.org/wiki/Appeal_to_the_stone “Speaker A: Infectious diseases are caused by tiny organisms that are not visible to unaided eyesight. Speaker B: Your statement is false. Speaker A: Why do you think that it is false? Speaker B: It sounds like nonsense. Speaker B denies Speaker A's claim without providing evidence…

The burden of proof is on the person making the initial claim and i highly doubt that the reasoning behind AI ruin is as extensive and compelling as the germ theory of disease. We assume human-level general AI is possible because we exist in nature but a super-human self-optimizing AI god is nowhere to be found.

Burden of proof on neutral matters depends on whoever makes the claim. But AI and the potential for doom isn't a neutral matter. The question is what should be one's default belief until proven otherwise. It is a matter of determining what decisions bring the most expected utility. The default assumption should be downstream from the utility analysis. But when it comes to AGI, extreme caution is the only sensible default position.

The history of humanity is replete with examples of the slightly more technologically advanced group decimating their competition. The default position should be that uneven advantage is extremely dangerous to those disadvantaged. This idea that an intelligence significantly greater than our own is benign just doesn't pass the smell test.

Re: Licensing is neither feasible nor effective for addressing AI risks

#119

Post by pixl97 »

SanderNL wrote: post #110
PeterisP wrote: post #91

Earlier quoted context omitted.

The way I see it, it's clear that human-level intelligence can be achieved with hardware that's toaster-sized and consumes 100 watts, as demonstrated by our brains. Obviously there is some minimum requirements and limitations, but they aren't huge, there are no physical or info-theoretical limits that superhuman intelligence must require a megawatt-sized compute cluster and all the data on the internet (which obvious…

Using this logic, flexible, self-healing, marathon running, juggling and childbearing robots that run on the occasional pizza are just around the corner, because, nature. It might us a thousand years to get anywhere close? I don’t see the good arguments for all of this happening soon.

So with your closing comment you're arguing it is possible, now you're just talking about time frames.

All I have to say is that in 1900 many thought flights by heavier than aur craft were tens of thousands of years away. 3 years later it was achieved.

Re: Licensing is neither feasible nor effective for addressing AI risks

#120

Post by pixl97 »

williamtrask wrote: post #115
PeterisP wrote: post #91

Earlier quoted context omitted.

The way I see it, it's clear that human-level intelligence can be achieved with hardware that's toaster-sized and consumes 100 watts, as demonstrated by our brains. Obviously there is some minimum requirements and limitations, but they aren't huge, there are no physical or info-theoretical limits that superhuman intelligence must require a megawatt-sized compute cluster and all the data on the internet (which obvious…

> there are no physical or info-theoretical limits that superhuman intelligence must require a megawatt-sized compute cluster and all the data on the internet (which obviously no human could ever see). Much of your "intelligence" is a function of natural selection. This is billions of years X bajillions of creatures in parallel, each processing tons of data at a crazy fast sampling rate in an insanely large/expensive…

LLMs we're messing with are text data only, we're barely starting to eat video data in multi modal LLMs. This world doesn't lack data.
Post reply on HN