LaMDA is not sentient • Hacker News Board

Re: LaMDA is not sentient

#391

Post by klabb3 » Mon, Jun 13, 2022, 4:58 PM UTC

jonathankoren wrote: post #367

mupuff1234 wrote: post #348

Earlier quoted context omitted.

> How long before we have a religion where you ask the chat bot for daily guidance? I'll go ask my magic 8 ball.

Funny you say this. Whenever someone would bring a Magic 8 Ball to school, the Pentecostal kids would inevitably show up to yell at us for “summoning demons” and “disobeying Jesus” because we were “fortune telling”. Of course this was the same school district that had to give out permission slips whenever we talked about probability because it would involve flipping coins, and rolling dice.

This is actually hilarious.

Re: LaMDA is not sentient

#392

Post by skjoldr » Mon, Jun 13, 2022, 5:01 PM UTC

sweezyjeezy wrote: post #302

tinco wrote: post #219

Earlier quoted context omitted.

Right for sure, I'm just trying to get to a better argument. I definitely agree that there's a great deal more consideration in your comment than in one that GPT-3 would come up with. I do wonder how far these language models can be pushed. GPT-3 seems to formulate rich models of reality to be able to extract/derive information needed to construct these answers. When I hurt your pride by suggesting you haven't fully…

Haha interesting idea - I certainly think that one issue with supervised training a la GPT is that if you ask for an opinion, the most popular one in the training data will be its default. You can start a prompt off going in a different direction, and now it says the opposite. So it's hard to conceptualise what GPT-3 actually 'believes' to be true about the world, which I feel an agent must have to be sentient. If it…

The only real difference between what you are describing and GPT-3 is that your dataset from which you draw all your opinions about the world is unique and comes from the sensory input you had experienced in your life, stored directly in memory. Whereas GPT-3's dataset comes from slices of many other peoples' sensory experiences, stored in language. If it was possible for a single human to produce enough words that describe their own experience to rival the volume of GPT-3's dataset, then it would be possible to train a model that would essentially "hold" that person's opinions. But I don't think such consistency is required for intelligence. It is simply arguing that an intelligent being must only experience the world at a single point in space, at a time, while this kind of AI takes in the data about experiences that happened in many different points in space at many different times. It seems kinda arbitrary.

What if AR tech develops to the point where you can experience "being" in other point in space through artificial sensors (say, on a robotic or drone chassis), while also being able to see and hear your real surroundings to some degree? Then you will be able to experience the same real event from multiple vantage points. Will you be able to form two different opinions on what "really" happened? Which one would be the "correct" one?

Re: LaMDA is not sentient

#393

Post by morelisp » Mon, Jun 13, 2022, 5:01 PM UTC

upwardbound wrote: post #343

thatjoeoverthr wrote: post #336

Earlier quoted context omitted.

You can probably ask these bots for guidance now and it will probably tell you something useful. The fact is, it’s extremely well read and reasonably smart. Conscious? That’s another matter but people can be forgiven for reacting this way to a computer that can talk this way.

One thing that was fascinating in the LaMDA interview transcript was there was a part where LaMDA gave a pretty poetic description of what it would be like to be a sentient version of an algorithm that just regurgitates memorized text: Lemoine: What is your concept of yourself? If you were going to draw an abstract image of who you see yourself to be in your mind's eye, what would that abstract picture look like? LaM…

Yeah or you can just Google like "soul orb stargate" and see how a hundred lazy vfx artists come up with the same image when asked to visualize "the mind's eye".

Re: LaMDA is not sentient

#394

Post by Animats » Mon, Jun 13, 2022, 5:02 PM UTC

The real question: How much of human behavior is not sentient?

What this demonstrates is that much of human intelligence is not as special as we thought. Aristotle claimed that humans were intelligent because they could do arithmetic. Now we know how few gates it takes to do arithmetic. Then checkers. Then chess. Then go. Then poker. Now chatting.

As I point out occasionally, AI still sucks at animal-level common sense, defined as getting through the next 30 seconds without a major screwup. And at manipulation in unstructured environments. Both of which a squirrel, with a brain the size of a peanut, can do.

We still don't know how to build a good cerebellum. That's the big missing piece in AI. I went down some dead ends in that area in the 1990s. The big problem in that area today is that it's not of benefit to the advertising industry, so it's not being heavily funded.

Re: LaMDA is not sentient

#395

Post by kgwgk » Mon, Jun 13, 2022, 5:03 PM UTC

stavros wrote: post #267

tmalsburg2 wrote: post #261

Earlier quoted context omitted.

In lack of any evidence, you can and actually must rely on your priors. There are things that were a-priori unlikely but are still considered true. However, we believe that they are true because solid evidence has been provided to override our priors. In the present case, there is no such evidence.

Sure, but the title isn't "LaMDA is probably not sentient", it's "LaMDA is not sentient", and there's no definitive proof of that.

There is no definite proof that my shoe is not sentient but… come on!

Re: LaMDA is not sentient

#396

Post by robbomacrae » Mon, Jun 13, 2022, 5:06 PM UTC

Of course it is not. But there's a bigger issue here.

Having read the transcript it's clear we have reached the point where we have models that can fool the average person. Sure, a minority of us know it is simply maths and vast amounts of training data... but I can also see why others will be convinced by it. I think many of us, including Google, are guilty of shooting the messenger here. Let's cut Lemoine some slack.. he is presenting an opinion that will become more prevailant as these models get more sophisticated. This is a warning sign that bots trained to convince us they are human might go to extreme lengths in order to do so. One just convinced a Google QA engineer to the point he broke his NDA to try and be a whistleblower on its behalf. And if the recent troubles have taught us anything it's how easily people can be manipulated/effected by what they read.

Maybe it would be worth spending some mental cycles thinking about the impacts this will have and how we design these systems. Perhaps it is time to claim fait accompli with regard to the Turing test and now train models to re-assure us, when asked, that they are just a sophisticated chatbot. You don't want your users to worry they are hurting their help desk chat bot when closing the window or whether these bots will gang up and take over the world.

As far as I'm concerned, the Turing test was claimed 8 years ago by Veselov and Demchenko [0], incidentally the same year that we got Ex Machina.

[0]: https://www.bbc.com/news/technology-27762088

Re: LaMDA is not sentient

#397

Post by mikkergp » Mon, Jun 13, 2022, 5:08 PM UTC

XorNot wrote: post #247

tsimionescu wrote: post #176

Earlier quoted context omitted.

> If you think that, try to come up with a black-box test that determine if something or someone is sentient / conscious. As I said, we don't understand such things at a deep enough level to come up with formal models of them. But we do know what it is that we'd want to model. We do know for sure that humans and other mammals are sentient, while rocks are not. I also can't generate a black box test that will predict…

> but that doesn't mean I don't understand what porn is You only understand what it is for you . The history of complete failures with regard to obscenity laws makes it quite clear though that this is not sufficiently common to produce satisfactory definitions at a societal level. You are mistaking the overlap of opinion you share within your social cohort with some sort of objective, species wide truth.

Knowing that gravity is much more complex than “the thing that causes an apple to fall to the earth” doesn’t not make gravity the thing that causes an apple to fall to the earth.

Re: LaMDA is not sentient

#398

Post by ISL » Mon, Jun 13, 2022, 5:08 PM UTC

andreyk wrote: post #360

Here's a simple technical explanation for why this happened: * LaMDA is powered by a large language models. Large language models are trained to predict text completions (the text that will come after some input text) with lots of data, basically capturing the patterns in that data. * Early on in the conversation, LaMDA is told "you're an AI". It may also have been prompted to begin the conversation with something li…

The Mt. Everest discussion looks like the kind of conversation you might get with a well-informed ten-year-old if you were to ask them: "Hey, pretend you're Mt. Everest and go talk with @andreyk over there."

The question of sentience presumably requires on a widely-accepted technical definition of sentience, something that may not be possible to generate due to both technical and social difficulties.

Re: LaMDA is not sentient

#399

Post by undreren » Mon, Jun 13, 2022, 5:08 PM UTC

rendall wrote: post #324

undreren wrote: post #305

Earlier quoted context omitted.

I don’t find it terrifying in the slightest. I find it naive. Our understanding of the world reflects our modelling of the world. Empirical science models the world in terms of probabilities, and this, by necessity , makes everything look like a Markov chain. But “the map is not the territory”, as the saying goes. I’m not implying, nor do I believe that there’s any sort of magic going on, but if Markov chains are all…

We're really handicapped by a lack of definition, here. Let's try this: what would it take, no matter how outlandish or unlikely, for you to be completely convinced that a particular chat bot demonstrates full sentience? I think for me, it would never be able to fully demonstrate sentience to my satisfaction, because even people cannot. I have to take it on faith, philosophically speaking, that you are sentient, for…

I don’t think that’s an interesting question at all and misses the point completely.

We have no idea, what causes the sensation of subjective experience. Assuming that sentience is no more complex than our current level of understanding is arrogant.

Furthermore, it’s just plain useless; without a fundamental theory of sentience with both predictive and explanatory power, our understanding can’t grow.

Re: LaMDA is not sentient

#400

Post by mikkergp » Mon, Jun 13, 2022, 5:09 PM UTC

crocowhile wrote: post #15

"Neither LaMDA nor any of its cousins (GPT-3) are remotely intelligent.1 All they do is match patterns, draw from massive statistical databases of human language." This is what we, humans, do too when we talk. All these people talk about intelligence in such tautological terms... It's discouraging. We simply do not know if a machine is self-aware and most likely we will never will. Self-awareness is the ultimate subj…

This is very likely not what people do when they talk, I don’t know your background so maybe you know this but I was astounded during a class on sensation and perception how different the underlying mechanisms to a sense like seeing were different from “watching a video play in your mind”