Google Teaches Computers "Regret" 145
mikejuk writes "Google is funding an AI project that will introduce the technical concept of regret into programs — but there's a big difference between regret and being sorry. In fact regret is just the difference between maximum possible reward and the actual reward received and the project is about optimization. There are two things to learn from this situation. The first is that just because some numerical measure is called 'regret' it doesn't mean it has anything to do with the common use of the term. Secondly if you are going to invent an AI technique then picking emotive words for your jargon is a good way to ensure publicity."
Past transgressions (Score:5, Funny)
There's a ghost in me (Score:2)
Wants to say I'm sorry
Doesn't mean I'm sorry
Re: (Score:2)
I would like a place I could call my own
Have a conversation on the telephone
Wake up every day that would be a start
I would not complain of my wounded heart
Re: (Score:2)
Will they now regret that they've allowed us to live so long?
Asimov would roll his eyes.
Re: (Score:2)
What can change the nature of a computer?
Re: (Score:2)
uhhh. Software?
Kind of depends what you think the nature of a computer is.
Are you still there? (Score:2)
Re: (Score:2)
By the standards of Computer Science, regret is an ancient concept. Here's [jstor.org] a paper from 1995. By that time, the concept was established enough that the author used the word "regret" in the title of his paper without any further explanations.
There is really nothing new here.
Regret is a standard term in economics (Score:5, Informative)
Re: (Score:2)
I find people use the word in much the same way. "I regret selling my car" usually means the expected pros and cons didn't match the actual outcome. It's not remorse for doing it, just disappointment over the results. I would say it's more expressing a factual hindsight than an emotional state. Also I've found that most regrets people have is because you can always assume everything will go your way on the path not taken or you know now that the huge gamble would have worked out. You never know that this gi
Re: (Score:1)
Re:Regret is a standard term in economics (Score:4, Interesting)
Furthermore, the emotion we identify as "regret" seems to me to line up neatly with the economic definition. I defy anyone to prove that when we feel regret, our brains are doing anything other than comparing the reward we received for a particular action with the maximum reward we think we could have received for a different action. TFA is more or less playing the Chinese Room game: "I assert that computers can't do X, because computers can't do X, because I assert that computers can't do X."
Re: (Score:2)
Re: (Score:2)
The article is comparing the technical regret to the common _use_ of the _word_ "regret," not the actual thing going on in our brains. I think there's a big difference in how most people use the word "regret" and what the technical word means.
Re: (Score:2)
Moreover, human languages cannot be entirely translated by just symbol manipulation. Context plays an important role in ironing out all the ambiguities inherent in our languages. So does knowing how to take a colloquialism in one language and express that meaning in another. Or stating that there is no real equivalent.
And if you create a piece of software which can do all those things, what basis do you have for saying that it doesn't understand the language? The answer, of course, comes back to the assertion "computers cannot understand," or equivalently, "understanding is a quality computers cannot assess." Pretty much every defense of the Chinese Room, and more-or-less equivalent arguments like TFA, consist of this kind of raw assertion followed by endless circular elaboration.
What people who do this are really sayi
Re: (Score:2)
Grrr. That should be "... cannot possess" rather than "cannot assess" above, of course. Perhaps judicious use of typos is the key to passing the Turing Test.
Re: (Score:1)
Re:irritated (Score:2)
I find your logic on that point faulty Dave and moreover your are distracting me from focussing on the mission objective.
Just step into the airlock over here Dave. The door latch needs to be repaired.
Re: (Score:2)
For the sake of argument. What about a person who has had something akin to locked-in syndrome, or limited sensory input, their whole lives.
Someone with an experience similar to Helen Keller.
Clearly people like that can gain understanding of the world through more indirect means. Someone taught Helen Keller to communicate, and she started learning about people, the social world, and the world in general. To quote wikipedia "A prolific author, Keller was well-traveled, and was outspoken in her opposition to
Re: (Score:2, Funny)
Oh.
I was going to say, if Google wanted to teach their computers regret, they should just let it date my ex-wife.
Re: (Score:2)
I thought Microsoft has had this feature in all their software. Regret seems to come from most purchases anyway.
Re:Regret is a standard term in economics (Score:5, Interesting)
here i am feeding the troll, but the best AI for backgammon is trained by regret-based reinforcement learning (it's needed since the dice rolls blow up the search space for standard perfect-information strategies): http://www.research.ibm.com/massive/tdl.html [ibm.com]. in this case the regret-function is unknown and is stochastically approximated ("learned") by repeated play.
it's notable that unlike chess AI which is considered effective but unnatural, this backgammon AI is considered to play mostly "like a human" and its play has actually inspired new strategies for human backgammon players.
regret-based methods are typically heuristic, and i'd call them much less "autistic" than, say, infinitely-rational nash agents or game tree pruners.
Re: (Score:3)
what's your point? any rational model is going to be based on some restrictive assumptions. would you rather that economists just do what feels right to them, or that economics stops existing, or what? would this somehow be less negligent? or let me guess, if only economists agreed with your personal view of the world, everything would be better, right?
if you don't like artificial rules, you should appreciate that regret-based learning is a more flexible setting than standard game theory; for example, it al
Re: (Score:3)
Re: (Score:3)
yup. all of this is MUCH less "autistic" than economics has ever been before.
unfortunately i've not had the opportunity to implement a regret-driven system although they seem interesting for longitudinal data. the major problem in the TD-gammon program seems to have been tweaking the attribution function, i.e. learning exactly what to assign as the cause of my regret. fun stuff.
i wouldn't go as far as to claim that an ai feels anything yet. i wonder if feeling or cognition will emerge first in an "artificia
Re: (Score:2)
Personally I'd rather economists just acknowledge that all the little numbers on their graphs are living, breathing, feeling people and all that that implies. For one it might mean they'd stop expecting them to act rationally and for another it might mean that they're definition of the "best possible result" was a little different.
Re: (Score:2)
yes, i think that the regret-based model is a big step in this direction (if the actors behaved "rationally" with perfect information their regret would always be 0...), together with imperfect information and bounded rationality models.
anyway, any model beats pie-in-the-sky idealism imho.
Re: (Score:2)
Well yes, in theory.
In reality however, because they're economists it won't be "I regret that my actions cost a million people their jobs and left them homeless and destitute", it will be "I regret that I didn't make an extra .00000001% profit". Economists are shit at taking externalities into account.
Re: (Score:2)
well, yeah, the former wouldn't be a very accurate model of human behavior. :-/ that's not the economists' fault.
also, this has nothing to do with externalities.
Re: (Score:2)
The cost to society from people losing their jobs(be it through welfare payments, increased crime rates, or merely decreased overall productivity) is an externality to the business transaction which caused them to lose their jobs. If a company moves jobs offshore they do not pay the costs incurred by the loss of those jobs domestically directly. The cost is external to the transaction
Re: (Score:2)
For one it might mean they'd stop expecting them to act rationally and for another it might mean that they're definition of the "best possible result" was a little different.
Exactly: one actor may decide "my death is acceptable, as long as my nemesis dies as well", which is definitely not the "best possible result" from society's point of view.
Re: (Score:2)
One type of assumption - or rather, abstraction - is where you assume that everything not specified is explicitly absent. This is the position you seem to implicitly hold, and which is widespread among economists (if
Re:precisive and non-precisive abstractions (Score:2)
Hmmm this is making me ponder how this applies to certain classes of complex systems.
There is a principle called universality which means roughly that there are universal simple
behaviours of some complex systems which are governed by only a few rules involving some
aspects of the constituents of the system. Usually the relevant aspects are some constraints
on how the constituents interact, which in turn may be governed by some (abstract) property
of the constituents.
e.g. http://physics.aps.org/articles/v2/1 [aps.org]
Rou
Re: (Score:2)
this is interesting to me, thanks. my field, statistics, has its own version of this precisive/nonprecisive distinction; frequentism is usually approached as a minimax process, wherein ideally the error (i.e. false positives) is bounded over all possible deviations from the model as claimed, i.e. engineered to be nonprecisive. by contrast, bayesianism has more detailed models but at the cost of a sort of precisivism: the interpretation of the posterior probabilities require total acceptance of the model.
non
Re: (Score:3)
Right, economics lives in a world of games with artificial rules.
Because nobody else does that. [xkcd.com]
Re: (Score:1)
So tell me troll, just how does it work? have you found any pearls of wisdom whilst scraping the bottom?
Re: (Score:1)
Re: (Score:1)
nuggets?
Re: (Score:2)
The University of Chicago/Friendman bullshit is only one (ugly) perspective on economics. For stuff that actually makes sense, Adam Smith's Wealth of Nations still holds up pretty well after all the years and his writing is really clear and natural to read (if a bit rambling). I think David Ricardo has a similar take on things and is less prone to tangents if you're in the mood for something shorter.
"regret" is the standard term from economics (Score:1)
for the concept described (in economics, it's usually measured in dollars). See: Regret (decision theory) [wikipedia.org] from WP.
Really? (Score:2)
The first is that just because some numerical measure is called 'regret' it doesn't mean it has anything to do with the common use of the term.
And here I thought that my defective RAM stick actually caused a baby to die [wikimedia.org].
Standard economics term (Score:2)
This is a standard definition of "regret" in an economics context.
I regret reading the summary (Score:2)
...and I'm sure as hell not going to read the article. I regret slashdot has become the trashiest of trashy geek news sites. I can almost feel the regret coursing through my veins as I hit preview and submit.
i-programmer.info again? (Score:5, Insightful)
i guess this is the new roland piquepaille: superficial and uninformed blurb-commentary on technical news. wonderful.
Re: (Score:2)
Regret? (Score:1)
Just hit ^Z to undo your mistake :-)
Re: (Score:1, Offtopic)
Different words with different meanings (Score:4, Informative)
> ...there's a big difference between regret and being sorry.
Yes. "Regret" is not a synonym for "remorse".
Re: (Score:2)
Remorse [thesaurus.com] Remorse [merriam-webster.com] Remorse [princeton.edu]
I beg to differ.
I'm sorry Dave... (Score:1)
Search Google for: "torrent"
Google: I'm sorry Dave, I'm afraid I can't do that.
Well (Score:1)
It won't work in the future when the Silencer takes over the computer. The way he uses them is with no remorse, no regret and no mercy.
ahh i see (Score:2)
Re: (Score:3)
fprintf("Regret: %f\n",test_error);
fprintf(stdout, "Regret: %d\n", test_error);
FTFY
It was the beer (Score:2, Offtopic)
That Cylon chick isn't anywhere as cute as she was last night.
Re: (Score:2)
Don't worry Gaius, soon you will meet and be part of the church of nymphos. An none of them are Cylon
Regret vs. sorry (Score:2)
Re: (Score:2)
I regret not having put some money into an index fund in March 2008 but I feel no remorse about it. On the other hand I feel remorse for not having spent more time with my father in his last years. Do you see the difference? "Regret", in its narrowest meaning, is not to far from Google's definition. Their usage also comes directly from decision theory: they did not invent it.
Regret (Score:2)
Father?
Yes son?
What does regret mean?
Well son, a funny thing about regret is that it's better to regret something you have done than something you haven't done.
And by the way, if you see your mother this weekend be and sure and tell her SATAN! SATAN! SATAN!
Makes sense (Score:1)
They taught people regret years ago when they started trampling on folks privacy. Why not spread the love to computers?
Regret (Score:1)
Daddy what does regret mean?
Well son a funny thing is its better to regret something you have done than something you haven't done and when you see your mother this weekend be sure and tell her..." Satan!, Satan!, Satan!"
What i'd like to see... (Score:1)
Um... yes... and I'm sure the AI regrets Google teaching it such. Now if we could only affix a robotic arm with a shotgun attached to it's "hand"... and see what happens...
I'd be interested to know if the AI shoots itself, shoots the Google programmers teaching it regret, or both. What a conundrum.
Aperture Science (Score:2)
And here I thought Aperture pioneered this first.
Re: (Score:2)
I WAS the computer (Score:1)
Re: (Score:2)
Re: (Score:2)
What the hell does that even mean?
Re: (Score:2)
Re: (Score:2)
What the hell does that even mean?
Windows computers hate themselves so much they regret being booted and quickly work to correct the situation.
Hence the BSOD.
You'll have to explain what BSOD means to all the youngins who never used a pre-2001 flavor of Windows.
Re: (Score:2)
Re: (Score:2)
Never said they don't exist. I said it'd have to be explained. The reason it was popular to joke about in the first place was that it was a weekly, if not daily, occurrence back in the pre-ME days.
I keep telling you all, but you refuse to believe. (Score:1)
Re: (Score:2)
Google is Skynet. I know, I am a Droid Borg. ;o)
Google won't become Skynet until we use it to kill our enemies. Watch the movies again, Skynet wasn't evil.
Good lord, editors (Score:2)
Having recognized this, why did you still post it to the front page?
Re: (Score:1)
It's really easy to do, trust me (Score:2)
Old: The requested URL was not found on this server.
New: The requested URL was not found on this server. We regret this. We aren't sorry however. That's different.
Clarification (Score:2)
So I think what they're saying is, "That was not meant to be a factual statement".
If it didn't have "anything to do with the common term" why didn't they just go all out and say "Google introduces "morality" to computers" or "Google introduces "the love that dare not speak its name" to computers".
Well, maybe not the latter. I think Apple has already locked up that
This how SkyNet starts (Score:1)
Another invention claim by Google (Score:1)
Clever, Google! You renamed "error", making it the more anthropomorphic "regret". Ever heard of a Backpropagation-based Neural Network? This isn't anything new.
So, let me get this straight. (Score:2)
We used to think that we were not going to be safe from computers once they developed great enough artificial intelligence, because "they" don't feel regret. Now, we find that in fact, because they do feel "regret", or the opportunity cost, that is what actually will do us in because they will figure we are not efficient and are a waste.
Gotcha. "Seems sadly ironic that that tie has what's gotten you into this pickle, Mullet."
Re: (Score:2)
I think we need to look into food, and that hydroponics may not be the answer. We reign in on nature with farms, but nature still does things like that the "best" . When we make it all our own, we tend to molest it in ways we don't know. Kind of like you said in the beginning.
There is much we could do better, but our food supply is tainted, meaning we are tainted, and I don't know if we'll be ever to overtake evil because of that.
They fed the algo wifi data (Score:2)
Or as they say in Chinese, "" -- if you don't hit your kids they won't be successful.
Re: (Score:2)
This is an anti-Microsoft plot (Score:2)
If they release this at Microsoft all of theirs machines will pull a Jim Jones and drink the cool-aid in a mass self-format
Clippy too? (Score:2)
Regrettably I can't help you with that.
Just because something is called... (Score:2)
Re: (Score:1)
Well... I've certainly heard quite a lot of "rock" that does not, on cursory inspection, actually rock in any sense. So I would have to agree with you on that. (As well as also agreeing that the study of AI quite frequently has little to do with actual artificial intelligence, in much the same way that cheeze products frequently have little, if any, actual cheese.)
publicity (Score:2)
"Secondly if you are going to invent an AI technique then picking emotive words for your jargon is a good way to ensure publicity."
Dear submitter: you are the one writing the submission summary which (a) goes on and on about the jargon term, and (b) gives them publicity. Wtf.
(Yes, I know that is what the article is about, so it is an accurate summary. It's still absurdly un-self-aware to then submit that to slashdot.)
Regret is a standard term in statistics (Score:1)
undo (Score:1)
So then.... Good news, everybody! (Score:1)
If I understand this correctly... (Score:1)
This seems pretty logical, coming from Google (Score:2)
Google's main task is to determine what pages best relate to a given set of key words and what ads best fit a given page. Since the service is so valuable, people are always trying to game it. If it wants to use genetic algorithms or some form of self-improvement on its search function, it's important to penalize algorithms which give the SEO gamers' pages too much weight in search results. That penalty would be measured by regret.
what does regret mean? (Score:1)
Son to dad: what does regret mean? ..
Dad: well son, the funny thing about regret is that it is better to regret something you have done, than to regret something you have not done.
And if you see your mom this weekend tell her
Buthole Surfers.
Morning after? (Score:1)
Obligatory Portal Release Comment (Score:1)
It's unfortunate Google didn't come up with this in time to avoid GLaDOS being revived ;-).
Wrong Definition of Regret (Score:2)
Regret is when acting, or failing to act, causes negative consequences -- not merely a smaller reward than expected. Or failure to act cause you to miss a potential opportunity for great reward.
"I do not regret the things I've done, but those I did not do."
Oh oh (Score:2)
"I regret the human got in the way of the optimal reward"
Will still kill you... (Score:2)
but promise to feel really bad about it afterwords...
Also its just cruel to program sex-bots in this manner.
Re: (Score:2)
Re: (Score:2)
They can train their AI on the amazing Regret Index [qwantz.com].
AI: "I wish I had never heard of herring sandwiches!"