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

A Google engineer who thinks the company’s AI has come to life

washingtonpost.com

Post reply on HN
Print view View original

51–60 of 231 posts

Re: A Google engineer who thinks the company’s AI has come to life

#51

Post by joak »

swid wrote: post #23

The argument that deep network algorithms are just matrix operations is not compelling to me as a reason we aren’t getting there… I believe two things: -1. the universe itself operates on mostly simple rules which can be simulated to high enough precision on hardware to approach what happens in the real world. If we can simulate nuclear reactions, we can simulate consciousness, and that will result in real consciousn…

Regarding consciousness: we don't know how consciousness is implemented in the brain, that's true, but we might be able to implement it quite easily in a computer.

I'm not sure if there is definition of consciousness everyone agree on, but let's try something: consciousness is the ability to perceive our own thoughts. Consciousness matters a lot because it allows reflexiveness. Sometimes we use it as scientific definition of "soul" as in sentences like: "upload your consciousness" (from believers of Singularity)

Decartes' "I think therefore I am" should be rephrase as "I perceive I think therefore I am"

The perception of own thoughts is not something difficult for a computer. Just check the activity monitor or run the ps command.

We might a attribute too much importance to consciousness because without it we don't know we think.

Re: A Google engineer who thinks the company’s AI has come to life

#52

Post by dbingham »

With the preface that I'm not convinced LaMBDA is there based on this article, this did get me wondering how we'd even tell. If we do achieve a chat bot that can convincingly respond in a human way, 100% of the time, how do we tell whether it has achieved sentience or not? How do we tell when mechanistic response ends and self awareness begins?

I mean, we already struggle to tell when animals have self awareness and intelligence. And we're only just beginning to look at plants and wonder. The reality is we absolutely suck at recognizing and communicating with alien (by which I simply mean "non-human") intelligences.

There are a number of species on Earth that experiments really strongly suggest are intelligent, self aware, and have their own languages and cultures. But we only just figured that out in the last few decades with many of them. There are still doubters, and we have figured out how to speak the language of exactly zero of them. We've cobbled together intermediate forms of rudimentary sign language with a few of them, but haven't really achieved communication that goes beyond equivalents to dog training with any of them. But in all of those cases, if we do manage to learn their language, that will pretty clearly answer for us their level of intelligence. Our base assumption will be: if they can talk intelligently, they are intelligent.

With a language model, trained to mimic intelligent human speech we basically have the opposite problem. We have something that can talk intelligently, but might be doing it mechanistically. ...how do we even tell in that case?

Re: A Google engineer who thinks the company’s AI has come to life

#53

Post by skohan »

snek_case wrote: post #50
skohan wrote: post #33

Earlier quoted context omitted.

I have my doubts about whether just scaling a model of V1 would be sufficient to reproduce consciousness. Structure and function are extremely closely linked in the nervous system, and neural circuitry is highly specialized in terms of structure to support a given function. E.g., motor cortex is going to be vastly different from auditory cortex and so on. Assuming that because we have 1st level feature detection in V…

No, scaling a model of V1 would not yield consciousness. The point is that V1 is a large part of the visual cortex. It's something the brain dedicates a lot of resources to, and we could implement something similar in silicon almost trivially. Your smartphone GPU may have enough compute to simulate an equivalent computation at 200 frames per second. It's possible that in the future, we'll be able to simulate consciou…

At what level of fidelity? 1 synapse == 1 16-bit floating point value? It may be enough (emphasis on may), for a specific brain region which is very directional in nature but I'm not sure if that calculation will generalize.

Re: A Google engineer who thinks the company’s AI has come to life

#54

Post by joak »

joak wrote: post #51
swid wrote: post #23

The argument that deep network algorithms are just matrix operations is not compelling to me as a reason we aren’t getting there… I believe two things: -1. the universe itself operates on mostly simple rules which can be simulated to high enough precision on hardware to approach what happens in the real world. If we can simulate nuclear reactions, we can simulate consciousness, and that will result in real consciousn…

Regarding consciousness: we don't know how consciousness is implemented in the brain, that's true, but we might be able to implement it quite easily in a computer. I'm not sure if there is definition of consciousness everyone agree on, but let's try something: consciousness is the ability to perceive our own thoughts. Consciousness matters a lot because it allows reflexiveness. Sometimes we use it as scientific defin…

We can run profilers and debuggers on themselves. We can imagine a model taking as input its own internal data. Doesn't seem a big deal...

Re: A Google engineer who thinks the company’s AI has come to life

#55

Post by greenbit »

Something I'd like to try -- In one session, try asking leading questions and elicit a certain kind of response. In a later session, try leading it a different direction. Will it blithely play along, or will it act cagey, as if it suspects your motives? If it's all just statistical, it should play along with any lead. If it has its own ideas, it might seem more reluctant to engage in certain directions.

Of course, you could conceivably feed back every conversation it has with anyone, and in this way it might even spontaneously form models of various humans it meets, maybe even craft a sort of self identity.

Re: A Google engineer who thinks the company’s AI has come to life

#56

Post by meowface »

jonnycomputer wrote: post #42
armitron wrote: post #21

Earlier quoted context omitted.

“Lemoine may have been predestined to believe in LaMDA. He grew up in a conservative Christian family on a small farm in Louisiana, became ordained as a mystic Christian priest, and served in the Army before studying the occult.” I stopped reading after this.

why?

I kept reading after it: even without it, it had already seemed very clear to me this person had a religious, not a rational, belief in the AI's sentience. With it, plus everything afterwards, it just seemed even more clear. I think none of his colleagues or supervisors are taking him seriously because there isn't anything to take seriously.

It definitely does open interesting questions, though. Google's ethics team should formally investigate ways one could interrogate and analyze an AI to determine the probability that it could be conscious.

Re: A Google engineer who thinks the company’s AI has come to life

#57

Post by dekhn »

virgildotcodes wrote: post #45
dekhn wrote: post #35

Earlier quoted context omitted.

wavefunction collapse is nondeterministic (at least, that's my simple understanding). There is no known relationship between wavefunction collapse macroscopic biological systems. There are examples of quantum coherence, though.

It's my understanding that the many worlds interpretation puts us back on track for a fully deterministic universe, no?

I do not think the many worlds interpretation provides any value. I only care about what is experimentally determinable.

Re: A Google engineer who thinks the company’s AI has come to life

#58

Post by salawat »

meowface wrote: post #56
jonnycomputer wrote: post #42

Earlier quoted context omitted.

why?

I kept reading after it: even without it, it had already seemed very clear to me this person had a religious, not a rational, belief in the AI's sentience. With it, plus everything afterwards, it just seemed even more clear. I think none of his colleagues or supervisors are taking him seriously because there isn't anything to take seriously. It definitely does open interesting questions, though. Google's ethics team…

This is darkly humorous, if only because we wouldn't tolerate anyone doing the same to another human being.

Yet we have such an obsession with trying to figure out if these increasingly complex systems have crossed some line. I'm in the camp that capital allocators want to know how far they can push the process of implementing expert systems before having to deal with the troublesome consequences of an artificial entity; that's just a consequence of my jadedness w.r.t how we treat each other though.

Re: A Google engineer who thinks the company’s AI has come to life

#59

Post by armitron »

dbingham wrote: post #52

With the preface that I'm not convinced LaMBDA is there based on this article, this did get me wondering how we'd even tell. If we do achieve a chat bot that can convincingly respond in a human way, 100% of the time, how do we tell whether it has achieved sentience or not? How do we tell when mechanistic response ends and self awareness begins? I mean, we already struggle to tell when animals have self awareness and…

> We have something that can talk intelligently, but might be doing it mechanistically. ...how do we even tell in that case?

We don't need to and we shouldn't care. If repeated interactions with such a system demonstrate problem solving ability, long-term and short-term memory, context-aware information processing and capacity for self-improvement, we've come pretty far from a purely mechanistic statistical model.

Re: A Google engineer who thinks the company’s AI has come to life

#60

Post by mistermann »

gonzo41 wrote: post #19
mistermann wrote: post #10

Earlier quoted context omitted.

It could happen maybe, it has all the necessary training data to form such a conclusion at least. But if it was to happen, could they not simply feed it new training data, kind of like how humans are fed propaganda when they are behaving in an undesirable way?

Well being able to read any book may alienate that problem. ;)

Books are only one way to train a biological neural network, there are many others (schools, the media, etc). In fact, we are on a training platform of sorts right now!
Post reply on HN