Hacker News Reader

Comment by nerpderp82 | Hacker News Reader

https://en.wikipedia.org/wiki/Roy_Amara

> We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.

Amara's Law is a good one, esp in these circles where we have many people working on cutting edge technology. They often give up because it doesn't see adoption, and then move on to something else. Two or three cycles later, when that technology is finally gaining traction, someone takes some off the shelf components and slaps it together, profiting in ways that the original creators would have liked to have seen.

Timing is everything, and the technoseti are often _too early_. But they also underestimate how big something will get.

I am a huge Python fan, but I never expected back in 2000 that it would be as big as it is today. Never. Cloud Computing, NLP, etc. Google suffers from this immensely.

View on HN

Amen. I wonder what "underestimating" something like AGI actually means though...

By the time AGI arrives, the world will have already been transformed in ways we cannot fathom by intelligence much lower than AGI. Look at the impact that computing has already had, and that is mostly accounting and simulation. Building and factory automation is on the order of a couple if statements. Most programmers can't implement a PID loop.

I am not scared by AGI. But to get to AGI you have to pass through a couple points on a curve that occupy the most powerful in terms of agency, less than AGI. An angry toddler that has > 1TW (terawatts) of power.

Agree, the path might be quite bumpy but so could the endgame. Whether AGI wants to help or kill the human race will depend on how aligned it is to our goals. Now, since "our" goals will depend on whoever has the metaphysical ear of such AI this could be problematic.

I think I read this kind of discussions too many times, imagine how the training set of GPT-5 will look like? all sorts of theories of "whether AGI wants to help or kill the human race" spread over the internet and analyzed to death. It is going to know this topic in-and-out. Will be able to write a masterful dissertation on the topic.

Yes, and that's good because GPT-5 is going to be coming from a team that cares about alignment and which has values only a little different from mine. They can still mess up: https://youtu.be/qV_rOlHjvvs?si=6SI_abC9BLHGoWjt

OTOH, an LLM trained by the FSB or CIA to help their agents, with no limits on what the agents are allowed to get as help because their bosses trust it's always in the service of the nation, will quite "happily" write that essay… and almost certainly get leaked eventually — 10 years, 50 years, doesn't matter — at which point someone who isn't so trustworthy will ask for a doomsday plan and it will be equally "happy" to write that too.

Hopefully the agencies like the FSB and CIA aren't quite as bad as I fear, but secrecy and what little has become public makes it impossible to trust them.

AGI doesn't matter, because the world will have been turned upside down multiple times before it even arrives. AI due to its extreme breadth is already smarter than any individual human across all of AIs competencies.

I feel that even in just the phrasing of the question, "wants" is doing an awful lot of heavy lifting.

I feel that even in just the phrasing of the question, "wants" is doing an awful lot of heavy lifting. It's almost tied in a kind of "No True AGI" framing, where if it doesn't want something it's not really AGI, etc.

> Most programmers can't implement a PID loop.

Wait, really? I would expect that most programmers could absolutely implement one if they understood what they are, but perhaps I'm biased from living in embedded land

Assuming it's aligned and we're still around…

AGI? Lots of people are still talking like it will be an assistant, perhaps using the "centaur" metaphor, and refusing to believe that it could[0] result in everyone's mental capacities having all the economic relevance of equine muscle power.

ASI? We don't know how far above us IQ can go, but it seems reasonable to guess that the smallest possible margin above us is such that even asking that question is like asking the question is akin to asking Chimpanzees to imagine the moon landing, while at worst it's like asking your lawn the same question.

[0] I say "could" rather than "will" because there's always the question of "how much does it cost to run that software?" — but the cost of 3430 W at $0.03/kWh matches the World Bank's international poverty line inflation adjusted to 2022: https://www.wolframalpha.com/input?i=%283430+watts+*+%240.03...

Why do we act like people all have an approximately uniform intelligence that will be surpassed by AI intelligence by orders of magnitude in all dimensions? In my experience, different people are smart about different things, sometimes yes by orders of magnitude. Couldn't an AI being smarter than some people be no more surprising than some people being smarter than some other people?

> Why do we act like people all have an approximately uniform intelligence that will be surpassed by AI intelligence by orders of magnitude in all dimensions?

I at least don't consider people to be uniform. If I act like I do, that's illusory.

For orders of magnitude higher performance by AI: all have a massive intrinsic speed advantage already (which is currently balanced by an almost but not quite equally massive training set required disadvantage — they can read the entire internet, but they need to too), and the go and chess AI have already vastly exceeded human capacity.

I think this kind of thing is also part of why AGI means different things to different people.

By my standards, ChatGPT is already an AGI, it's able to operate on text-based tasks in general (the G in AGI) at a level higher than some humans and lower than others — if I'm right, a lot of people are already no longer capable of being economically viable, and their only hope in the current economic system is that the market for their labour doesn't act with perfect information. I may also just be wildly over-estimating the capabilities of ChatGPT.

yes, chatGPT had let people lose their jobs. this also prove that we have no chance when AI got everyone's job. do anyone care people who displace by chatGPT?

> this also prove that we have no chance when AI got everyone's job.

Not "prove", but it is suggestive. To prove it, ChatGPT would have to also prevent them getting another one.

> do anyone care people who displace by chatGPT?

I can't parse this sentence. Are you asking:

1. Does anyone care about the people who are displaced by ChatGPT?

2. Does anyone care for the people who are displaced by ChatGPT?

3. Does anyone care that people are displaced by ChatGPT?

4. Does anyone care about the people who are using ChatGPT to displace others?

Or something else?

>Not "prove", but it is suggestive. To prove it, ChatGPT would have to also prevent them getting another one.

Maybe it won't take all jobs. but this displayed that it is impossible to implement UBI or other strategies before most of people losing job.

>I can't parse this sentence. thanks for your free English class, bro.

There are a lot of things we can define with physics because they can be easily measured. The particular problem with intelligence it is a measure of systems complexity which is something us as humans are very bad at measuring. As far as I know all measures of intelligence we have are relative. That is one person/system is more intelligent than another. But there is no meaningful measure if how intelligent a system can be. That is, what is peak intelligence?

> it seems reasonable to guess that the smallest possible margin above us is such that even asking that question is like asking the question is akin to asking Chimpanzees to imagine the moon landing, while at worst it's like asking your lawn the same question.

"You mustn't be afraid to dream a little bigger" with regards to the hierarchy of intelligence - the "at worst" is much further away than human-to-lawn. Look at the intellectual power of angels in Catholic theology and consider an intelligence that can reason from universals to all particular instances as its default mode of operation. Then consider that each angelic being is distinct from each other one by nature (since they are immaterial). Which means that there is a species-level difference between each of them and the lowest one is already far above us. Then consider that there are (at least) billions of them, each one order of magnitude more intelligent than the one below it. Then consider the kind of intellect that can create those kinds of intellects.

Fortunately, you can't give what you don't have, so we're quite safe from the basilisk. But the wild imaginings of John C. Wright's Count to the Eschaton Sequence are probably not far off for "hey how weird could it get?"

The point of my comparison was not to indicate the exact scale of thought difference, but the incomprehensibility gap. In this regard, while I've not read Count to the Eschaton, I'm confident any work written by humans isn't weird enough.

Whereas, what I'm saying is, given that you can't give what you don't have, we don't have to worry about creating something more intelligent than we can be. But just how intelligent we can be is a pretty surprisingly large horizon in the first place.

While this is true, there is also a survivorship bias effect. Remember the semantic web? XML? When javascript was going to displace all other scripting languages, from backend to database queries? If a technology follows the trend described by Amara's Law, we will recognize this, because at the end it's bigger than we had imagined (by definition). When it doesn't, we are less likely to remember how those overhyped predictions just never came to pass, because of that very fact.

Maybe too early to say if blockchain will turn out to follow Amara's Law, but I have my doubts.

Someone using Amara's Law in the first or second AI winters (https://en.wikipedia.org/wiki/AI_winter) would, of course, have the option of just continually saying "it hasn't been long enough yet". But, you know, that very fact means that Amara's Law doesn't tell you much, even in the cases where it turns out the be true. And it doesn't, usually, turn out to be true.

Python's popularity is more tragedy than success story.

I love Python and I agree. Standard ML or OCaml would have been far superior on nearly every axis. Every language is slowly becoming OCaml anyway.

Reply on news.ycombinator.com