[Discussion] What is AI?

So when I shared that video who plot exactly where words are stored it's not retrieving memory?

Also the idea of we don't understand what it does (it approximate an input function) is kinda not semantic, we can visualize exactly what convolutional networks learn on a layer basis and extract image and concept, surprisingly we obtain a latent space that mimick the mapping of words extracted in the brain like above. Latent space we can manipulate to modify inputs toward another output.

Where is the semantic in direct application?




Another visualization of stored concept inside a neural network "manifold" of bi lingual words embedding retrieved for visualization




http://colah.github.io/posts/2014-07-NLP-RNNs-Representations/

The limit on brain study is simply ethics, we can't just experiment on living subject in destructive way. That slow down a lot.

EDIT:
More here
http://colah.github.io/

 

To prove that we have achieved understanding of a human brain, link me a paper that describes image processing performed by human brain in detail. All the way from the initial pulses from the eye and down to breaking it into objects.

The article you linked has no connection whatsoever to understanding of human brain.

 

we still have more to learn about eye adaptation - much less the actual processing (I recently read all about our current understanding when learning about tonemapping).

 

I never claimed that the assumptions are identical. But both require a logical leap. I probably took "it's fairly logical" as too firm of a statement.

And while I'm not about to deny that the first possibility seems like the more natural outcome, things like fractals and complexity (and indeed quantum science) have caused radical changes in our perception of the world. The most pertinent of these being fractals, or a layer of randomness that exists in nature beyond the base logic we expect.

But my views here are probably skewed because this is such a big portion of what I deal with on a daily basis.

 

To me a good definition of intelligence is any animal or machine that can look a the following and give the right answer:

Red cat. Blue cat.
Red dog. Blue ___?

That is pretty much the simplest IQ test possible. I would think an ape would fail this. But a human child would probably pass it quite easily.

Spoiler

The answer is dog

 

if children werent trained on written tests for years and suddenly found a piece of paper with that written on it, most wouldn't have any clue that they were even supposed to write something in that blank space, much less that it was a pattern matching quiz.

they'd be like "why is there an underline" and throw the piece of paper away.

you're really taking for granted how many years of training we devote into teaching children how to answer that kind of question and how much of a foundation in tests we build.

In terms of animals being able to do pattern matching - most mammals are capable of doing far, far more complex pattern recognition than this.

 

Children also have context for what cats and dogs are, and colours. Kids go through a lot of training as they grow up.

 

I don't even think you need to look to mammals. Most birds can do much more complex pattern recognition.

You might even be able to achieve that kind of thing with fish (absolutely octopus).

 

While the statement that such a test would separate humans from animals might be false, it might be valid as a test of "intelligence" as first mentioned.

 

The answer is "unknown". The lesson is "don't jump to conclusions."

"1, 2, 3..." what's next?

Spoiler

Answer: 187, 231, 34

----
Children also undergo change in their thinking process as they grow up.

I can't find exact test, but the idea was if you present a little kid a picture of a bowl, teacup, saucer and a piece of bread, and piece of bread, and ask to pick three "connected" objects, the kid most likely will pick saucer, bread and a teacup (instead of bowl, saucr and teacup which are "kitchenware"). Because "I eat breat, drink tea and put the cup into saucer".

 

I think something like that was actually examined with regards to kids from different backgrounds.

Edit:
http://www.huffingtonpost.com/ronnie-reese/test-bias-minorities_b_2734149.html

 

The test I mentioned covered change in logical/thinking patterns that happens around age of 3 or so ... in ALL children.

 

The argument is partial understanding,ie we can't make a human yet, but you are in luck, the visual pathway is certainly the most documented part of the brain, because we can directly visualize the neuron response, an we have experimented with firing pattern on cat.

It's not even obscure that's full blown wikipedia
https://en.wikipedia.org/wiki/Visual_system

 

there is a great deal we're still learning about every part of that system. even rods/cones which are simple sensors. Let's not even start on the visual cortex. That we can vaguely identify what regions of the brain are connected to what function doesn't mean much, our actual understanding is still extremely primitive at best.

 

I want full understanding and see partial as "not good enough".

-------

I've lost interest in the subject. This is clearly not going anywhere useful, half of it revolves around semantic/dictionary arguments, so the best idea would probably just to lock it.

I want information I don't know. Arguing about "whether partial qualifies as good enough" is a waste of time and produces no useful info. It is a dictionary/semantic argument. A useful info would be a full schematic of visual cortex, functional neural model you can play with, etc. And not this.

So, I'm done here. Have fun.

 

@neginfinity to be fair - you had to know this thread wasn't going anywhere. look at the poll options.

 

That's the nature of the thread lol

But the ay I see this is more like:

- I understand how cpu are made, I can maybe create a crude cpu, but the leap between my cpu and the last i7/ryzen ... you name it is night and days, they don't actually use significantly new concepts from the crude cpu, but they do it at another scale of efficiency and sophistication, that's just the design, let's not get into the fabrication. And even yet, I can understand the concept of how pipeline are used and optimized, that there is order and out of order, that there is cache size and how they affect computation, I know there is instruction set but I don't know them all and their intricacies. I have a rough understanding of the sophisticate architecture, I cannot grasp it as a whole, specially at the silicium level, but that's not "NO understanding".

I think there is the same kind of leap between current understanding of the brain and the sophistication of how it works. It allow neurochirurgien to already solve some problem.

And the other aspects is that some people are so blown away by the results of human intelligence that it's like them using a computer and then having someone it's really made of the same basic things that a light switch, and the person who doesn't undestand binary is like, "wtf, you are mde, that's not even remotely the same things, my 4k 60fps game is much ore sophisticate, you can replicate that with 0 and 1".

 

If you see a lion. And you think to yourself. "Well, it might kill me. But I'm not going to jump to conclusions. I'll just stand here and see what happens."

Then that is not very intelligent. Sometimes the most intelligent thing to do is jump to conclusions.

 

And that's how jumping to conclusions can get you killed.

The recommended action when encountering a lion is to stand there or retreat slowly while facing it.
http://www.discoverwildlife.com/travel/how-survive-lion-attack
http://www.slate.com/blogs/quora/2013/05/15/safaris_what_do_i_do_if_i_encounter_a_growling_lion.html

----------

Perhaps it is time to finally lock this up?

 

I think the smell coming from my pants would work well in a "play dead" position.

 

Seems to be the same with many animals.

http://www.dailymail.co.uk/travel/t...survive-attack-world-s-dangerous-animals.html

Though the start of the african wild dog one made me laugh.

 

Welp, that's about enough.

--Eric