Please look at the class

Artificial intelligence

from

MIT

on

YouTube

by

Winston

Dear Ziad Sobih ,

https://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-034-artificial-intelligence-fall-2010/lecture-videos/lecture-1-introduction-and-scope/

"And frequently you can substitute computing for intelligence.

So no one would say that Deep Blue does anything like what a human chess master does.

But nevertheless, Deep Blue, by processing data like a bulldozer processes gravel, was able to beat the world champion."

Konstantin

I believe there’s just a bit of a misunderstanding here. It’s not impossible to explain how neural networks, even multi-layered ones, or complex genetic metaheuristics, work. If that were impossible, then they wouldn’t exist in the first place. If I create something I cannot explain the workings of, then I have created nothing.

On the other hand, explaining how these algorithms work is not the same as tracking their results. You can definitely give a 101 explanation for the results of k-NN when applied to a textbook dataset, by using dynamic plots for instance (tons of that on YouTube). But as soon as algorithms become more refined, and datasets multidimensional, detailed explanations start to get unwieldy.

But then again, the very reason why these algorithms were created is we want them, instead of us, to do the math.

That’s the epistemological part. The ethical part is a tad more intricate. But that’s not what your question was about.

Dear Leandro Tripodi ,

All human learning is based on explanations of knowledge. This way knowledge transfers from teacher to student. How AI black-box can transfer knowledge to human person? Why we should believe AI, if it can not explain its decisions like humans do?

"But then again, the very reason why these algorithms were created is we want them, instead of us, to do the math. "

If we decide that AI is just one more math branch, why it is discussed as intelligence like human so enthusiastically?

Konstantin

Hi,

You are talking about the need of an AI to justify to humans its reasoning and conclusion-making.

In other words - humans should accept the outcome of an AI's functioning only if an AI is accountable to them. The creation must be answerable to its creator.

It sounds a bit like the God - human children, or the parents - children relationship.

Is your question related to following/ confirming AI's quality of work by humans, or human desire to control the intelligence of an independently operating computing machine?

Dear Valeria Tananska ,

"Is your question related to following/ confirming AI's quality of work by humans, or human desire to control the intelligence of an independently operating computing machine? "

Every machine humans created, even autonomous like coffee suppliers, usually could be verified and tuned by humans. If it is not verifiable and tunable - it is dangerous and unpredictable. And it also could be use for deception, because some people propose to believe in higher intelligence of black-box, not using humans own intelligence to set decisions.

AI as intelligent being does not exist, so we can not control its intelligence.

Best regards,

Konstantin

Konstantin hello,

I understand.

Regardless of someone's doubts as to how much an AI can be relied upon or (ab)used for, there is a limit to what one can expect from it in terms of compliance and accountability.

Let us consider a parent - child relationship.

Just because two people created, up-brought and educated a child, as parents, they cannot reasonably expect that when the child reaches maturity, he/she will:

- be constantly explaining him/herself to his/her parents, or

- will not disappoint the parental body when he/she does not explain.

One does not endow a creature or a machine with the capacity to make independent decisions, only to exact explanations on how he/she/ it came to a conclusion, with an eye at curbing any nefarious tendencies early on.

After all, reasoning is not always about data analysis, but also about in-coded guiding principles/ laws, intuition, protection of self and others, developing new goals.

I guess, it also depends on what humans expect an AI to do - work for humans, with humans, or in an independent capacity.

If you consider the first two options, I can understand the conceptual doubts.

:)

Dear Konstantin,

1) “All human learning is based on explanations of knowledge. This way knowledge transfers from teacher to student. How AI black-box can transfer knowledge to human person?“

That’s not really what AI is about. I believe there are common misconceptions about ML/AI that serious researchers have to come to grips with before they risk making incorrect statements for the public. There is a lot of ‘marketing’-inspired terminology involved in ML/AI solutions that we must be conscientiously wise to. When in doubt, ask yourself, “What is it that this algorithm does”?

2) “Why we should believe AI, if it can not explain its decisions like humans do?“

That’s a far nicer question. IMO the reason why we should take algorithm outcomes seriously is we know what that algorithm does to our data and the results provide perspectives that differ from our common-sense beliefs. Then we become able to question these very beliefs, because we know what the algorithm does (though we’re not interested in keeping track of its operation) and we know it’s relevant in the case.

Now, again, this concerns only the epistemological part.

3) “If we decide that AI is just one more math branch, why it is discussed as intelligence like human so enthusiastically?”

That’s another interesting question. I believe part of the answer lies in the question itself: over-enthusiasm. Part of it lies in ‘marketing’-inspired terminology that often gives people wrong ideas. Part of it lies in common-sense misconceptions arising from just a little too much science fiction.

At this point I’m suspecting you might be interested in my contractual theory of knowledge as developed in my Aforismi 2019 (don’t worry, it’s in English). If you think so too, then please follow the link below. All the best.

Preprint Aforismi 2019