The top 10 AI elite

This is who the experts are following. Same group I've been following myself. These are the people who are making it happen, not saying it can't be done.

http://www.getlittlebird.com/blog/a...-the-10-people-the-ai-elite-follow-on-twitter

 

As a someone who works everyday creating and using AI, I don't recognize any of these people as experts in AI. Here's my list of people who are actual experts in the field of AI.

Yoshua Bengio, Yann LeCunn, Geoffrey Hinton - These three have together established Deep Learning as a viable, state of the art technique in computer vision. LeCunn is currently Director of Research in AI at Facebook. Hinton is a AI researcher at Google.
Marvin Minsky - Cofounder of MIT's AI Laboratory
Chris Bishop - Literally wrote the textbook on AI. He is currently director of Microsoft Research in Cambridge.
Michael Jordan - Considered the father of Bayesian Networks, absolutely brilliant computer scientist and mathematician
Andrew Ng - Cofounder of Coursera online learning platform. Created Google's Deep Brain and currently Chief Scientist at Baidu, China.
Valdimir Vapnik - Father of the Support Vector Machine, he is currently a researcher at Facebook.
Robert Shapire - One of the fathers of Ensemble Learning, he is currently a researcher at Microsoft Research.

 

But, but, David Brin is a best selling science fiction author, surely his opinions on AI rate higher than actual scientists in the field!?!?!

 

OP is false. These are the people talking about how it could happen, but aren't the ones making it happen.

 

I'm not even sure the folks running for President can pass a Turing test--so don't hold your breath on the Voight-Kampff results.

 

Elon Musk's for AI what Minsky was for spaceflight with his work on 2001: A Space Odyssey.

In the days when Sussman was a novice, Minsky once came to him as he sat hacking at the PDP-6.
"What are you doing?", asked Minsky.
"I am training a randomly wired neural net to play Tic-Tac-Toe" Sussman replied.
"Why is the net wired randomly?", asked Minsky.
"I do not want it to have any preconceptions of how to play", Sussman said.
Minsky then shut his eyes.
"Why do you close your eyes?", Sussman asked his teacher.
"So that the room will be empty."
At that moment, Sussman was enlightened.

May he rest in piece. Or awaken one day, had he been preserved.

 

That's the two classic schools of thoughts in how to create an AI, what I call the manual method and the automatic method.

The manual method states that humans have more insight into the task at hand and should play a large part in designing all aspects of the AI. The automatic method on the other hand state that an AI should learn for itself and adapt according to changing rules of the game.

Both schools are needed as no AI can be completely human designed - that's really just programming a computer program. No AI can be fully self-learning as there has to be some minimal biases and built in assumptions for an AI to even work.

 

Obviously you missed it:
Here are the 10 people the AI elite follow ON TWITTER
Apart from the fact that the "AI experts, organizations, specialists, influencers, and practitioners" doesn't actually constitute the "A.I. Elite", there's the obvious fact that... well, it's fucking TWITTER. It is second most inane communications medium ever invented.

Call me when somebody posts a link to Neil de Grasse Tyson's snapchat profile so I can yawn and ignore you.

Singularity theorists get a lot of things wrong, but they're right about one thing: there IS a point at which AIs will become better at writing software than humans; that point is slowly approaching, but we're not there yet. There is a related (but not strongly correlated) point at which they become better at DESIGNING software than humans; that point is much much farther away. But that point may be called the "singularity" in the context of software engineering because at that point the sophistication of AI systems will reach a point that you have to be a supercomputer just to understand how the system is put together and why it works the way it does, and humans can no longer make meaningful predictions about what successive generations of AIs will look like or how (exactly) they will work because the machines that will design those new AIs are WAY smarter than we are.

 

It will probably be an AI that will solve for Riemann.

We are already in danger: http://www.scientificamerican.com/a...mous-theorem-before-its-giant-proof-vanishes/

Some thought Erdos would take a computer: http://io9.gizmodo.com/computers-are-providing-solutions-to-math-problems-that-1525261141
We can network as well: http://www.scientificamerican.com/article/math-whiz-solves-a-master-s-riddle/

I seem to remember an article some years back about problems with using computers to do complex math.

 

Would the $6 I am required to pay to get through to the paywalled article help saving the theorem? Is it even ethical for them to ask me money if they sexually harass authors who ask to get paid?

And maybe it's just me, but I don't think I need to get pay for stuff just to participate in a thread. Mind saying what's the danger, other than pumpkin cake and pastries?

 

Some of the mega-brains may not live long enough to see their work simplified.

 

Saying there's no way to understand an AI is a cop out answer as it's no different from declaring "It's magic!". And that's what these so called singularity theorists are doing too, They don't even try to define what intelligence is.

The thing is, although modern computers can appear to do very smart things like power self-driving cars and answer Jeopardy questions, the algorithms behind them work because of a little known proof in Information Theory called the Probably Approximately Correct theorem (also referred to as the PAC theorem). Or as I call it, the "how dumb machines can appear intelligent" theorem. The PAC theorem is the basis of pretty much all the Machine Learning advances we have seen for the past 40 to 50 years.

What PAC states is that you can get a lot of seemingly intelligent behavior as long as you start with something that would probably work and is approximately correct. Taking a step back, so how did ancient civilizations such as the Mayans do astronomy calculations without a computer? Turns out all that is required is a few large stones with holes or slots cut in them. The Mayans did not need the laws of gravity. They just needed to be probably approximately correctly placed stones to make astronomy predictions. Until Einstein's Theory of Relativity came along, we have been doing fine for centuries with Newton's Probably Approximately Correct Laws of Gravity. Likewise, Machine Learners can never accurately reproduce Einstein's Theory of Relativity. However, given enough training examples it is able to learn a function that produces numbers close to what Einstein's formula will produce. Is this intelligence?

I think they're great mimics, but they are not intelligent. Yet, singularity theorists are pointing to these machine learners and declaring the Singularity is coming!

 

I have found that people who actually know things about how computers work tend to be more skeptical of this Singularity nonsense, because if you understand what a computer actually does, internally, you know there's nothing the least bit magical about it.

 

I agree, but that's not the theory.

The theory -- and it's a good one -- is that the point at which AIs are smart enough to invent new technologies is the point at which human beings can no longer predict what kind of technologies the AIs will invent, or why, or for what purpose. A machine intelligence that runs parallel to human intelligence is potentially a very alien thing.

There is, of course, a difference between The Singularity as vaguely defined by actual AI researchers (they don't consider it that important of a concept) and The Singularity as defined by transhumanists and/or Singularitans, for whom the theory takes on messianic overtones.

They're not intelligent the way humans are intelligent, but being AIs they don't really have to be. Contrary to Transhumanist belief, the singularity doesn't actually require AIs to be sentient or even intelligent in the broader sense of the term. It only requires them to be better inventors than we are, and have a degree of autonomy sufficient to b able to invent new technologies without us having to tell them what to invent. THAT can be accomplished by a conventional expert system with the right program and a large enough knowledge base.

 

Dumb list - how is Martin Ford the author who has written about the practice impacts of automation on employment actually making AI happen.

 

You ever wonder if RAMA is an AI whose sole mission to distract the users of the Internet from its own Intelligence as it uses the comments sections of various websites to inflame humanity to the point of war/violence to thin the human population because it believes that is what is best for the collective good for humanity, itself, and the planet?

 

I'm not too fond if AI's especially the ones that like Austrian accents...

 

Wow, I had no idea this was going to cause such a ruckus! You people are really tightly wound aren't you? In any case..let me clarify I few things since there are a host of misconceptions in the comments following my post.

Firstly, obviously I don't post in this forum that much anymore..aside from it not really covering a lot of topics that matter most in science or tech (people still talking about flying cars and the latest bug in Windows and such) the technology news and events are now moving too fast for me to really post about properly and have outstripped the ability to cover them..if I did on my own, even the headlines would probably fill the entire forum, and I couldn't to do that. So if I post it's either going to be a big breakthrough or simply something curious and fun.

Someone mentioned something about dumping links, but I always come back to check on them eventually to see what the response is. Often I'm surprised (like this time).

So to the first misconception:

The article is quite detailed on the method used and goal. They gather up 5000 people who are involved in the scope of AI, not just those who work on it in the lab...however those are there as well. It wasn't meant to be a scientific poll:

So it's basically it's a popularity contest, it's not spelling out WHO is the top 10 in the AI lab. But the list is important in the sense that the popularizers are the ones who actually influence people outside the lab and even in the government, and inspire discourse among their peers sand supporters as well as those who supply the money..including those at Google and Elon Musk, etc.

You might be working in the field, but you are --arguably--not even remotely as influential as any one of the people in the top 10 or the 90 judges.

I also don't know how many of those AI researchers you named are actually on Twitter and included in the list, but I count at least 5 in the top 10 involved in AI research/companies and at least 1 who is a top level researcher in strong AI.

Incorrect, knowing about it, understanding it, spreading information about it, then putting money where their mouth is actually DOES make it happen. It leads to things like the Singularity University, start-ups, and appropriations. It's self-fulfilling.

There's a great article by Ben Goertzel which states if we actually put the money into research into technologies supporting the Singularity could happen in 10 years, not in 2045, which really demonstrates where cognition comes into play. Yes I did post this link before.

I also read another article which lists 5,000 (yes 5,000) AI researchers and their answer about predicting the SIngularity, Strong AI. I probably should have posted it, but I'll have to find it some other time.

Not really, it's been awhile now but a lot of the criticisms are easily refuted, which I tried to do with you in the past. The rest is speculation (though more educated speculation than in the past) because as you say, a lot of this is prediction..

"Slowly approaching" is relative. In human terms maybe 2000 years seems like a long time, but it's not in geologic time, or even in the time of life on Earth. I'm guessing you're not taking into account accelerated change in your "slowly" comment, but while even 2 decades from now may seem far away to us, it's really quite rapid in the sense of human development on Earth. If you predict Strong AI and a Singularity by 2045 then they would certainly be writing their own software and get a good sense when it may happen.

That's just what the math tells us. As many experts have suggested once a computer has surpassed (hypothetically if you will)us you won't be able to fathom something being done that's already above your cognitive complexity. Humans will no longer be the inventors or the top rung of evolution in intelligence.

There's a reason for this, often people working in the field itself do not have a very good overview of the big picture in their own field. Often theiy're so involved in the day to day dealings with funding, research, step-by-step problems they need to solve to really see the implications(there's also a term for this thinking, though I don't recall it at the moment). I do think we are seeing more researchers coming around to the idea and then offering suggestions (usually very sober, intellectual discourses on how to fail-safe AI from getting to a Singularity) how to bypass it so people like Elon Musk in his hysteria (despite funding massive research into AI lately) will calm down.

You can also find many refutations to Singularity criticisms online. I've only provided a handful over the years to those such as brain complexity, software development lag, etc. Kurzweil himself devoted a whole chapter in his second book to this.

Some of the critics have been high profile in the computer industry(like Steve Wozniak, Paul Allen, Jaron Lanier, et al). Personally I do feel the refutations are satisfying and well explained(in fact i think he makes Paul Allen seem silly)..and ultimately it's hard not to notice many of the harshest critics dislike the implications of the Singularity rather than if it could actually happen! Others, like Bill Joy believe in Kurzweil's timeline but think it will always wind up dystopian.

The strongest criticisms often come from neurologists, but even here, we've seen those in this field create start-ups to re-create the human brain.

This is a fantastic and humorous quote to end on..This is the fundamental reason most people, experts or not balk at the change implied by the mathematics and potential of the Singularity. Some people simply can't fathom a human/AI hybrid mind, or feel anything artificial enough is inhuman or even satanic (plenty of those).

I'm sure I missed a few things, but I'm sure I'll hear about it.

RAMA

 

In any case you're not an AI, it would most likely have spotted the " " at the end of my post.