Discussion Topic |
|
This thread has been locked |
Ed Hartouni
Trad climber
Livermore, CA
|
|
Dec 28, 2018 - 12:25pm PT
|
^^^I dont' agree, people who are seriously into chess also don't agree, and I think that the link to the Google paper upthread regarding debugging tools for neural networks suggests that the descriptions regarding how the network arrives at its conclusions indicates that this is quiet a bit more subtle than just the output state of the network.
Interestingly, these are similar problems with humans, my guess is that those same machine debugging tools will be useful in understanding biological neural networks.
To the extent that a chess grandmaster can explain how they executed a particular game, the story is the same with AlphaZero, it has to do with experience playing the game. An advantage that AlphaZero has is that it can play many more games than a human could. In sufficiently complex neural networks, that results in "insight."
AlphaZero would also defeat the "Chinese Room" argument by example.
|
|
healyje
Trad climber
Portland, Oregon
|
|
Dec 28, 2018 - 12:30pm PT
|
We'll have to agree to disagree.
|
|
Ed Hartouni
Trad climber
Livermore, CA
|
|
Dec 28, 2018 - 12:37pm PT
|
exploring our disagreements is what this thread is all about, by declining you make it very much less interesting
|
|
healyje
Trad climber
Portland, Oregon
|
|
Dec 28, 2018 - 12:45pm PT
|
Pretty simple, I laid out my position. I'm quite familiar with chess and technology and believe in this instance, almost by definition given the play, chess experts just aren't all that expert in this particular case. Astonishment and wonder on our part at having overlooked something significant in such a well-worn domain isn't "insight" on the machine learning program's part - ML isn't sentient and doesn't have insights.
|
|
MH2
Boulder climber
Andy Cairns
|
|
Dec 28, 2018 - 01:35pm PT
|
ML isn't sentient and doesn't have insights.
How do you, healyje, know when someone else has had insight?
|
|
healyje
Trad climber
Portland, Oregon
|
|
Dec 28, 2018 - 02:00pm PT
|
Code isn't someone and chess experts not understanding says more about chess experts than ML.
|
|
Ed Hartouni
Trad climber
Livermore, CA
|
|
Dec 28, 2018 - 02:31pm PT
|
they are expert in chess, and students of the game (meaning they've studied many many games)
not only that, but playing chess at that level also involves trying to understand what your opponent's strategy and tactics are, so the human chess experts are very aware of "chess thought."
you're response opens up a number of questions based on your assumption of what an "insight" is, and similarly what "sentience" is,
insight: "the capacity to gain an accurate and deep intuitive understanding of a person or thing"
sentience is a "sentient state," where sentient:
1: responsive to or conscious of sense impressions, sentient beings
2 : AWARE
3 : finely sensitive in perception or feeling
certainly AlphaZero has demonstrated insight, whether or not it is sentient depends on our ability to define things like "responsiveness," "consciousness," "awareness" etc which we have so far failed at doing in this thread.
|
|
WBraun
climber
|
|
Dec 28, 2018 - 04:50pm PT
|
define things like "responsiveness," "consciousness," "awareness" etc which we have so far failed at doing in this thread.
No .... only gross materialist fail because they are ultimately always clueless .....
ALWAYS .... as the gross materialists are always stuck in the illusion of the material plane being all there is.
|
|
MH2
Boulder climber
Andy Cairns
|
|
Dec 28, 2018 - 08:45pm PT
|
Code isn't someone
Is that a sound basis for deciding what code can't do?
|
|
MikeL
Social climber
Southern Arizona
|
|
Dec 29, 2018 - 08:05am PT
|
Ed: your avatar image shows you wearing glasses. You never responded to my question regarding the reason why... in particular, we say the glasses "correct" your vision, what does that mean to you?
Apologies . . . an oversight.
Answer: to see more clearly the fine print. Short of that, I see fuzziness and cannot tell what I could be reading. When that happens, there seems to be a dullness in the moment of awareness. I believe that “correcting one’s vision” is a term or phrase that the eyeglass people use; it would not be my choice of a phrase. I would simply say, “I can’t see.”
“Seeing” has shifted to a new kind of sensibility or understanding since I began devoting myself to art projects. It’s certainly different than the way I used the term when I was a paid academic. Rather than “seeing” being some kind of output of representation, “seeing” is now expression, and I see that my expressions are intrinsically creative.
(BTW, your response to my earlier post about instrumentalism above is excellent, imo.)
|
|
WBraun
climber
|
|
Dec 29, 2018 - 08:19am PT
|
Seeing means how close one's consciousness comes to absolute truth.
One can see thru the third eye which is always free from all material impediments ......
|
|
Ed Hartouni
Trad climber
Livermore, CA
|
|
Dec 29, 2018 - 09:26am PT
|
"...I would simply say, “I can’t see."'
interesting that you would say that, as your eyes "see" without your glasses, but what they "see" is perceived by you to be in contradiction with your expectation, that is, there is fine print to read, and with the limitations of your unaided vision, you cannot resolve the print.
A simple physical correction is made, by an instrument, to allow you to resolve the fine print.
Now you are using that instrument to provide you information, through its ability to allow your eye to resolve the fine print, of the world about you. The experience of resolving that world provides you the means of understanding that world.
The glasses extend the ability of your current eyes allowing you an experience of the world.
|
|
High Fructose Corn Spirit
Gym climber
|
|
Dec 29, 2018 - 10:35am PT
|
Yes, MikeL, of course. But did you bother to actually watch the video before posting your vague, ambiguous contrarian usual?
From the video, I'm seeing what seems like MINDS vying against each other (while perhaps some here are more comfortable seeing mindless zombie robots? lol) - the entire interaction rather game-esque (involving aims/desires, planning, strategy, winning and losing) ending in a kind of checkmate.
...
Here's yet another rather cool, rather game-esque interaction between two minds of different species. I'm thinking any tool-using primates keen on design - tool designs - more advanced than termite sticks might enjoy...
[Click to View YouTube Video]
https://www.youtube.com/watch?v=GHUjeIEoPcE
|
|
healyje
Trad climber
Portland, Oregon
|
|
Dec 29, 2018 - 10:48am PT
|
They are expert in chess, and students of the game (meaning they've studied many many games) not only that, but playing chess at that level also involves trying to understand what your opponent's strategy and tactics are, so the human chess experts are very aware of "chess thought."
Yes, they are chess experts and yet, they and every chess expert in the history of the game discounted the value of certain lines of approach which AlphaZero pursued and found productive. The insight isn't on AlphaZero's part, it just didn't discount or abandon any line or approach because it had the cycles necessary to evaluate their utility. The only 'insight' is to the chess experts who long ago abandoned those lines as unproductive.
Certainly AlphaZero has demonstrated insight, whether or not it is sentient depends on our ability to define things like "responsiveness," "consciousness," "awareness" etc which we have so far failed at doing in this thread.
Again, AlphaZero has not demonstrated insight, only an ability to weigh lines from any given board position. Its principal advantage is it wasn't 'taught' chess by humans who have long-established biases built into their thinking around the game. AlphaZero, free from such biases simply evaluated the utility of every approach and line free from such bias and proceded accordingly. No consciousness, sentience or insight was involved - just a lack of bias.
|
|
Ed Hartouni
Trad climber
Livermore, CA
|
|
Dec 29, 2018 - 11:32am PT
|
so what's special about human chess?
how do humans gain insight?
how do they learn chess?
|
|
yanqui
climber
Balcarce, Argentina
|
|
Dec 29, 2018 - 11:34am PT
|
^^^^^^^^^^^^^^^^^^
Plus it practiced 19.6 million games. That's a lot of practice.
|
|
Don Paul
Social climber
Washington DC
|
|
Dec 29, 2018 - 11:41am PT
|
You all are talking about game theory, not consciousness. An amazing subject, but won't help you understand what is mind.
|
|
Ed Hartouni
Trad climber
Livermore, CA
|
|
Dec 29, 2018 - 12:04pm PT
|
I think I understand game theory,
but once again, the point is how does the machine differ from a human?
saying that human capability is somehow "innate" is the same as really not understanding.
Humans learn by playing with other humans, by reviewing play, and by discussing the play (formally and informally).
They build experience, learn openings, learn end-games, recall patterns, etc...
when they have "insight" it is a product of exactly what AlphaZero has done, or Stockfish where that program has had a lot of coaching (and biases result from that).
It is legitimate to say that the "insight" wasn't AlphaZero's, but of the humans looking at what AlphaZero did and having an insight as a result. But human chess players have done the same throughout the history of the game, by looking at the record of games, and deriving from those games new chess ideas.
AlphaZero purportedly played more many more training games than the entirety of recorded human games. That is certainly an advantage of machine chess. It did it in 4 hours.
As Kasparov said, what AlphaZero "knows" about chess is the "truth."
Could humans use this to become better players?
they already use Stockfish for game preparation, not against machines but against other human players.
Presumably AlphaZero (or at least the technology it's based on) could be used in the same way.
If in the end we conclude that "mind" is only an innate attribute of humans, then we're effectively done with the discussion on this thread.
|
|
healyje
Trad climber
Portland, Oregon
|
|
Dec 29, 2018 - 12:08pm PT
|
so what's special about human chess?
Clearly nothing beyond the fact we are capable of it at all.
If in the end we conclude that "mind" is only an innate attribute of conscious organisms, then we're effectively done with the discussion on this thread.
Fixed that for you - there is not and is not going to be a sentient/conscious AI.
|
|
Ed Hartouni
Trad climber
Livermore, CA
|
|
Dec 29, 2018 - 12:18pm PT
|
there is not and is not going to be a sentient/conscious AI.
that's an assertion that you offer without any supporting evidence.
It's a bias informed from your experience, it certainly doesn't rest on any definition of either "sentient" or "conscious" (you've offered none).
If you posit that it is an attribute of biological systems, you must have some reason (you've offered none).
|
|
|
SuperTopo on the Web
|