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

There's No Fire Alarm for Artificial General Intelligence

intelligence.org

Post reply on HN
Print view View original

61–70 of 209 posts

Re: There's No Fire Alarm for Artificial General Intelligence

#61

Post by Figs »

SomeStupidPoint wrote: post #55
sago wrote: post #33

I've been around AI since the end of the last big hype in the late 80s. The recent leap in machine learning has felt rather hyped to me. I don't think AGI is near. But I find myself agreeing with this article. Strongly. And I have long suspected that, we miss a lot of the significance and opportunities in AI, because we have only one exemplar of 'higher' intelligence: a human being. AI folk are so concerned with gett…

Can you define 'near' for me? I think AGI is likely closer to the present than 1987 was -- that is, I'd bet on having AGI by 2047. (Note: this is distinct from superhuman AGI.) Do you not agree? I think a lot of people underestimate NNs because they think of NNs in terms of the semantics of their history instead of all possible semantics that can be fit to tensor networks. We know [P] that NNs are a sufficient abstra…

> DAG

Feedback and memory are really important features of GI that you will not get out of a DAG ever. You need loops for that.

Re: There's No Fire Alarm for Artificial General Intelligence

#62

Post by mark_l_watson »

Great read, and I don’t mind at all that the last section was a pitch for donating to MIRI. I have been an AI practitioner since 1982 and have enjoyed almost constant exposure to people with more education and talent than myself so I feel like I have been on a 35 year continual learning process.

I think that deep learning is overhyped, even though using Keras and TensorFlow is how I spend much of time everyday at work. I have lived through a few AI winters, or down cycles, and while I don’t think that the market for deep learning systems will crash I think it will become a commodity technology.

I believe that AGI is coming, and I think it will use very different technology than what we have now. Our toolset will change dramatically before we can create AGI. I use GANs at work, and in spite of being difficult to train, the technology has that surprising and ‘magic’ feel to it, however, so do RNNs, and that technology is 30 years old.

I am going to show my age, but I still believe in symbolic AI. I am also fairly much convinced that AGI technology will be part symbolic AI, part deep learning, and part something that we have not yet invented.

Re: There's No Fire Alarm for Artificial General Intelligence

#63

Post by SomeStupidPoint »

Figs wrote: post #61
SomeStupidPoint wrote: post #55

Earlier quoted context omitted.

Can you define 'near' for me? I think AGI is likely closer to the present than 1987 was -- that is, I'd bet on having AGI by 2047. (Note: this is distinct from superhuman AGI.) Do you not agree? I think a lot of people underestimate NNs because they think of NNs in terms of the semantics of their history instead of all possible semantics that can be fit to tensor networks. We know [P] that NNs are a sufficient abstra…

> DAG Feedback and memory are really important features of GI that you will not get out of a DAG ever . You need loops for that.

> estimator built out of a DAG of calculations

All loops can be modeled as a DAG and single attached piece of memory (of sufficient width) allowed to execute to a steady state; sorry if it wasn't clear that I was talking about things like NTMs too. (It's why I used 'tensor network' most places; also, in practice, we tend to let subgraphs reach a steady state independently where possible.)

Your comment is also an excellent example of a strawman: you picked out the word 'DAG' to raise a technical argument when the usage of DAG versus general tensor networks clearly wasn't the main point (as some NNs have feedback and the standard model is posed as differential equations).

It's more constructive to respond to the strongest point, not pick at technical details that can easily be rephrased.

Re: There's No Fire Alarm for Artificial General Intelligence

#64

Post by rl3 »

sago wrote: post #33

I've been around AI since the end of the last big hype in the late 80s. The recent leap in machine learning has felt rather hyped to me. I don't think AGI is near. But I find myself agreeing with this article. Strongly. And I have long suspected that, we miss a lot of the significance and opportunities in AI, because we have only one exemplar of 'higher' intelligence: a human being. AI folk are so concerned with gett…

>The recent leap in machine learning has felt rather hyped to me. I don't think AGI is near.

I've always thought it's never too early to start allocating significant resources to AGI research and safety given the potential impact. That said, I've up until very recently agreed with your take on the situation.

What changed my mind was an article detailing the latest advances in silver nanowire mesh networks.[0]

I knew neural computing was a thing, but not that we already had a computing substrate capable of self-organizing its own neural architecture based entirely on external input, with power requirements analogous to the human brain. No firmware or software required.

One could say that human physiology remains far more complex just on the substrate front alone, what with the brain being an incredibly complex, delicate balance of chemicals and heterogeneous cells. However, this particular artificial substrate is already succeeding in basic learning tasks, despite the fact it's far simpler.

I steongly suspect we've figured out at least one artificial computing substrate not only capable of, but perhaps well-suited towards producing AGI, and that it's just a matter of scaling it.

Of course, once you scale the technology sufficiently, the question then becomes how to architect and train it into an AGI. You say as much above, but I suspect the architecture need not be human to be a threat, or to otherwise become extremely powerful.

[0] https://www.quantamagazine.org/a-brain-built-from-atomic-swi...

Re: There's No Fire Alarm for Artificial General Intelligence

#65

Post by leereeves »

SomeStupidPoint wrote: post #55
sago wrote: post #33

I've been around AI since the end of the last big hype in the late 80s. The recent leap in machine learning has felt rather hyped to me. I don't think AGI is near. But I find myself agreeing with this article. Strongly. And I have long suspected that, we miss a lot of the significance and opportunities in AI, because we have only one exemplar of 'higher' intelligence: a human being. AI folk are so concerned with gett…

Can you define 'near' for me? I think AGI is likely closer to the present than 1987 was -- that is, I'd bet on having AGI by 2047. (Note: this is distinct from superhuman AGI.) Do you not agree? I think a lot of people underestimate NNs because they think of NNs in terms of the semantics of their history instead of all possible semantics that can be fit to tensor networks. We know [P] that NNs are a sufficient abstra…

The ability to compute intelligence given infinite time and compute power isn't proof that NNs are a useful approach.

For that, a technique must be able to compute human intelligence at a useful speed on a constructable system.

Re: There's No Fire Alarm for Artificial General Intelligence

#66

Post by namelost »

icebraining wrote: post #53
mrob wrote: post #52

Earlier quoted context omitted.

alrs was probably in no real danger, but why should they be expected to memorize the exact risk profile of every mechanism they're exposed to? In practice we delegate most risk management to government regulation. It would be very time consuming to evaluate everything on a case by case basis. It's a reasonable heuristic to assume things are safe when they act as they usually do and dangerous when they act differently…

I'm not saying alrs should. The decision to leave is reasonable. The decision to judge others ("What is wrong with people?") is not.

That depends whether the other occupants knew about the safety features and record of elevators. If they did, then they were making a sound judgement. If they didn't, then they were being completely reckless.

Re: There's No Fire Alarm for Artificial General Intelligence

#67

Post by marcosscriven »

icebraining wrote: post #45
alrs wrote: post #24

I was on an elevator full of people, and between two floors something went wrong and we went into free-fall for part of a second. We made it to the next floor, the door opened, my fellow passengers were content to stay in the elevator. I turned, said "My plan is to not die in an elevator today" and got off. What is wrong with people?

Elevators have pretty good safety mechanisms. Even cutting the cable wouldn't have killed anyone. I'd probably leave too, but just because I wouldn't want to get stuck inside it if it stopped in the middle of two floors, especially since it was full. But for fear of death? Nah. From Wikipedia: "In fact, prior to the September 11th terrorist attacks, the only known free-fall incident in a modern cable-borne elevator h…

The gym I used to go to had an incident with someone being killed by the lift falling: http://www.dailymail.co.uk/news/article-1278427/Health-club-...

Admittedly it’s rare, and personally I wouldn’t be worried about it as a liklihood.

Re: There's No Fire Alarm for Artificial General Intelligence

#69

Post by SomeStupidPoint »

leereeves wrote: post #65
SomeStupidPoint wrote: post #55

Earlier quoted context omitted.

Can you define 'near' for me? I think AGI is likely closer to the present than 1987 was -- that is, I'd bet on having AGI by 2047. (Note: this is distinct from superhuman AGI.) Do you not agree? I think a lot of people underestimate NNs because they think of NNs in terms of the semantics of their history instead of all possible semantics that can be fit to tensor networks. We know [P] that NNs are a sufficient abstra…

The ability to compute intelligence given infinite time and compute power isn't proof that NNs are a useful approach. For that, a technique must be able to compute human intelligence at a useful speed on a constructable system.

Didn't the same argument apply in general to why NNs weren't particularly useful for anything ~30 years ago?

Using a network of tensors to compute intelligence is incredibly old (I believe, dating back about 80 years), but has only recently become tractable to do for any complex tasks.

However, in the past ~30 years, we've gone from "intractable for moderate problems" to "world champion at go", "able to detect cancer in images as well as experts", etc. My contention is that in another ~30 years, we'll see a step sufficient for "can do average at most intellectual activities", even if that's just having the storage to keep 10,000 task specific NNs (of AlphaGo sophistication) on hand to interpolate all actions as mixes of specialist tasks. Do you really not think there's a strong heuristic case for that? (I would contend that you should be able to point to a specific task you don't think it will be able to do on that timeline -- do you know of such a task?)

The proof was merely that we're not barking up a theoretically dead tree -- we have to rely on heuristics for if it will eventually converge to tractable.

Re: There's No Fire Alarm for Artificial General Intelligence

#70

Post by leereeves »

SomeStupidPoint wrote: post #69
leereeves wrote: post #65

Earlier quoted context omitted.

The ability to compute intelligence given infinite time and compute power isn't proof that NNs are a useful approach. For that, a technique must be able to compute human intelligence at a useful speed on a constructable system.

Didn't the same argument apply in general to why NNs weren't particularly useful for anything ~30 years ago? Using a network of tensors to compute intelligence is incredibly old (I believe, dating back about 80 years), but has only recently become tractable to do for any complex tasks. However, in the past ~30 years, we've gone from "intractable for moderate problems" to "world champion at go", "able to detect cancer…

I'm not arguing that NNs aren't capable of AGI, merely that the ability to compute the standard model leaves the far more difficult question of whether the problem is tractable, as you said.

The standard model could be computed directly without NNs, which I think we agree wouldn't be a useful way to approach AGI.

Post reply on HN