Deep Learning: mathematics and neuroscience (2016) [pdf] | Hacker News Reader

Deep Learning: mathematics and neuroscience (2016) [pdf]

cbmm.mit.edu

145 pointsdangiankit19 comments

Lots of interesting points but I don't know if I agree with the conclusion: " Beyond the case of Deep Learning, I believe that a concentrated effort on the basic Science of Intelligence and not only on the Engineering of Intelligence is high priority for our society "

In my view, it seems like the so-called "engineering of intelligence" is continuing to prove extremely fruitful and racing ahead of the "science" of intelligence. It may well be that models are growing in complexity beyond what we can explain/understand theoretically.

Also, there are already a ton of extremely useful existing tools for intelligence, and to me it seems like what's missing is not better algorithms but more engineers to weave intelligence into more parts of life.

Agree with the part or weaving more intelligence into parts of our life.

Although with the current wild west of collecting everything about everyone to sell useless shit might be where all intelligence weaving might happen

Policy recommendation: whenever you read "deep learning" think something like "Universal Regression". As in, really good curve-fitting.

> Build robots that are ethical

This poses the question, are humans ethical and good for the planet? Fueled by out genetic drives we cause more deforestation and earth movement than natural forces. We farm animals in the millions and exploit them for our benefits in tiny spaces causing a lot of psychological damage.

Before we go on to make smarter more powerful humans/robots should we ask ourselves whether that is actually better for the planet?

I'm having trouble getting past the first sentence: "The problems of Intelligence are, together, the greatest problem in science and technology today."

Please don't post unsubstantive comments here. Nitpicking the wording of an article's first sentence is practically a caricature of the Internet Dismissal.

I also think bergie's comment is a fair critique. If the very first sentence of an article is that absurd/unfounded/hyperbolic/etc, it's certainly fair to call it out.

He is providing minimum viable evidence that article is not particularly well written. I found his comment helpful.

A standard internet boast about not getting past sentence n, where the snark gets bigger as n gets smaller, is nothing worth posting here.

The article's author may not be a native English speaker. If so, the dismissal was not only unsubstantive but parochial. We have a lot of international content on HN and users need to read it charitably (and read other things charitably too).

Hmm, ok that's a good point.

That is a very fair critique. I don't know how that sentence passed the peer review without someone spilling coffee on their keyboard.

This isn't a peer reviewed paper

He's pointing out a sentence that wouldn't pass muster in a primary school book report. Yours is the only "unsubstantive comment" I see here.

It then goes on to describe really basic problems billions of humans are naturally really good at, things we have no actual problem with in our real life, as opposed to say things that we are not good at, and therefore are real problems we have.

I'm not saying we shouldn't make AI to rival humans at those things, but they simply aren't our greatest problems.

It's true, in domains such as vision, humans excel. But the implementation of computer vision is an open door for much more powerful robotic applications to emerge, in industry, transportation, medicine, administrative work and caretaking.

Making vision cheap and effective for computers will help humans not need to do such backbreaking work as picking fruit (which was protected from automation because it depended on vision and dexterous manipulation). In agriculture, computer vision could lead to plant-level care, reducing the need for chemicals.

"The problems of Intelligence" - the many problems that one might associate with discussions about intelligence. They are indeed numerous.

"are, together, the greatest problem in science and technology." - You probably get this: together, those problems are the "golden ring" (or brass ring - see the carousel metaphor at https://en.wikipedia.org/wiki/Brass_ring) of science.

my paraphrasing would be:

"Intelligence is the greatest problem in science and technology today."

You cannot start an article like that unless it is your essay and you're a first year student.

You're wasting reader's time on meaningless, unproved statements. Perhaps he should demonstrate why that is the case. Or say "X considers Intelligence ..." and then quote him.

Well, he was born in Italy and English is likely not his first language. And his is a name I've heard all my career so, yes, I think he can start an article pretty much any way he pleases and I'll gladly read it.!8-))

But also I agree with Dr. Poggio's first statement in the paper (i.e.,"The problems of Intelligence are, together, the greatest problem in science and technology today."). I have a proof that this is so, however I haven't the time to post it this small textbox and be assured of no typos. Tempus fugit.

But if you know of other problem(s) that you believe are greater then you should indeed reveal them to us.

Perhaps he is writing a 'betoog' (I do not know how to translate this), a piece in which the point is not to inform the user but to convince them of a certain point of view.

Reply on news.ycombinator.com