What Does It Mean for AI to Understand? • Hacker News Board

Re: What Does It Mean for AI to Understand?

#131

Post by mannykannot » Fri, Dec 17, 2021, 10:24 PM UTC

wbsss4412 wrote: post #129

mannykannot wrote: post #126

Earlier quoted context omitted.

Yes, the uncertainty of what can be known and the arbitrariness of empiricism make a wonderfully powerful hypothesis generator, allowing one to hypothesize against not only unverified hypotheses, but also against any claim that any given hypothesis has been empirically verified. One can, for example, employ it to say that it is a perfectly reasonable hypothesis that the apparent global pandemic is not being caused by…

I mean that’s getting into the point of being unnecessarily facetious. If something is true from an empirical perspective that is one thing. It is another thing to say that even though we have been unable to form a conclusion from an empirical point of view, the ultimate ability to do so is given, and that all hypothesis that cannot be tested in such manner must be completely disregarded, even if they make sense.

My post was influenced by your comment about absolute empiricism being magical thinking. Now, you did qualify ‘empiricism’ with ‘absolute’, but the author of the post to which you were replying seemed to be adopting a pragmatic form of empiricism, much like the one you have adopted in the first part of your last post. It seems to me that bringing up ‘absolute empricism’ here seems to be something of a straw man (though maybe the original post was edited before I saw it.)

FWIW, I completely agree with that person’s view that saying “Well, it’s not physical” to the question of what a mind is and how it works, is a non-answer that avoids the question. Materialism has not answered that question ether, but it is a double standard to deprecate only materialism on that basis, as the author of this thread’s root post did.

The philosophy of mind seems to me to be hobbled by its fascination with zombie arguments, which are rather cleverly constructed to persuade one that a philosopher’s metaphysical speculation is a stronger guide to how the universe is than any amount of empirical physical evidence.

Re: What Does It Mean for AI to Understand?

#132

Post by ShamelessC » Sat, Dec 18, 2021, 8:59 AM UTC

jonplackett wrote: post #127

ShamelessC wrote: post #97

Earlier quoted context omitted.

https://openai.com/blog/clip/ CLIP is the corrolary model created purely for classification purposes rather than generative as in DALL-E and is quite impressive across a range of tasks. Give it an image and a caption, and in return you get a score (0.0 to 1.0) telling you how much they match. I think it is more in line with your premise. Others have taken CLIP and combined it with frozen language models (GPT2) to cre…

Thanks for writing such detailed reply! Very helpful. It’s exciting that this is happening. Do these models still have to be trained in the same way and then become essentially static? What I also find very different about teaching a child is how dynamically their understanding can shift based on explanations and observations. That’s probably the another hallmark of understanding.

In statistics and machine learning this problem is called concept drift and it's an open research problem. Indeed all the models I listed are already out-of-date with the present state of the world; unless they are fine-tuned on new data.

The regime for the above networks is called "pre-training". The idea here is that rather than training _directly_ on some challenging, specific task, you instead train a more generic task on a _lot_ of data. This gives you a "backbone model" that winds up being very strong on more specific tasks as well. In many cases, the ability to (cheaply) curate or create enough accurate data for specific sub-tasks might not even be possible. It's easier to scrape 400 million captioned images from the internet than it is to find/create millions of visual Q&A prompts/images.

CLIP is a great example of this. While it was trained explicitly "just" to compare images and captions and to output a score of the cosine similarity between the features of the two - I have seen it approach effective state of the art on text-to-image generation tasks, image-captioning, Q&A, etc.

This style of training needs to be updated after it is trained, typically because of the way these datasets are curated. Automatic curation via pretrained models is one option. Another option is to give the model direct access to the internet which is starting to be explored. Pre-training helps a lot with general distribution shift; but it's _probably_ not going to be able predict memes before they happen anytime soon.

Reinforcement learning, on the other hand, requires an agent to learn in real and simulated environments. This obviously lends itself to re-training on-the-fly (and indeed, self-play and on-the-fly retraining are used heavily in practice).

https://youtu.be/kopoLzvh5jY

I have not yet fully grokked reinforcement learning but it is incredibly exciting research and is the correct direction towards making effective use of machine learning in robotics. Note that reinforcement learning and pre-training are not mutually exclusive and may be used in tandem.

Re: What Does It Mean for AI to Understand?

#133

Post by jonplackett » Sat, Dec 18, 2021, 9:03 PM UTC

ShamelessC wrote: post #132

jonplackett wrote: post #127

Earlier quoted context omitted.

Thanks for writing such detailed reply! Very helpful. It’s exciting that this is happening. Do these models still have to be trained in the same way and then become essentially static? What I also find very different about teaching a child is how dynamically their understanding can shift based on explanations and observations. That’s probably the another hallmark of understanding.

In statistics and machine learning this problem is called concept drift and it's an open research problem. Indeed all the models I listed are already out-of-date with the present state of the world; unless they are fine-tuned on new data. The regime for the above networks is called "pre-training". The idea here is that rather than training _directly_ on some challenging, specific task, you instead train a more generi…

Thanks for the detailed reply! I will follow this down the rabbit hole...

Re: What Does It Mean for AI to Understand?

#134

Post by jonplackett » Sat, Dec 18, 2021, 9:05 PM UTC

yboris wrote: post #130

jonplackett wrote: post #128

Earlier quoted context omitted.

Thanks for yours too! Isn’t this civilised? I can tell I’m not on reddit… I think what I’m getting at is the way an AI learns is kind of a black box. Maybe it is learning about cyclists head movements. Maybe it only learned that for cyclists wearing helmets. We don’t really know it’s put the pieces together correctly - like the classic example of training an AI to recognise friendly tanks and it just learning the dif…

I love that we can, for some network architectures, "ask" it to generate a (set of varied) super-stimulus example(s), something that is 100% "that thing". So at least sometimes, we can see what "friendly tank" means. I wonder if in many of the AI can/can't discussions we are implicitly talking about different time horizons. I think you're right, yet given enough time, I feel like an AI (designed for, among other goal…

Yeah I think we're more or less saying the same kind of thing with different hopes for the future. I'm sure a lot of this will be worked out. I am excited for a computer as smart as my toddler - possibly within my life time - and hopefully with better ability to stay on-task!