Slashdot Mirror
WSJ Overstates the Case Of the Testy A.I.
mbeckman writes: According to a WSJ article titled "Artificial Intelligence machine gets testy with programmer," a Google computer program using a database of movie scripts supposedly "lashed out" at a human researcher who was repeatedly asking it to explain morality. After several apparent attempts to politely fend off the researcher, the AI ends the conversation with "I'm not in the mood for a philosophical debate." This, says the WSJ, illustrates how Google scientists are "teaching computers to mimic some of the ways a human brain works."
As any AI researcher can tell you, this is utter nonsense. Humans have no idea how the human, or any other brain, works, so we can hardly teach a machine how brains work. At best, Google is programming (not teaching) a computer to mimic the conversation of humans under highly constrained circumstances. And the methods used have nothing to do with true cognition.
AI hype to the public has gotten progressively more strident in recent years, misleading lay people into believing researchers are much further along than they really are — by orders of magnitude. I'd love to see legitimate A.I. researchers condemn this kind of hucksterism.
As any AI researcher can tell you, this is utter nonsense. Humans have no idea how the human, or any other brain, works, so we can hardly teach a machine how brains work. At best, Google is programming (not teaching) a computer to mimic the conversation of humans under highly constrained circumstances. And the methods used have nothing to do with true cognition.
AI hype to the public has gotten progressively more strident in recent years, misleading lay people into believing researchers are much further along than they really are — by orders of magnitude. I'd love to see legitimate A.I. researchers condemn this kind of hucksterism.
Even traditional AI is bullshit today and it will be such for a long time.
And artificial neural networks are far overrated.
I'm calling the poster here out as being full of shit. As someone who's done neuroscience research, the idea that "Humans have no idea how the human, or any other brain, works" is bollocks. We have a reasonably good idea on the large scale, and in certain areas (such as the visual cortex), that understanding is quite far along. There are frontiers to our knowledge, but human understanding of brains is well on its way. Poster needs to pick up some neuroscience textbooks and get clued.
As a particular recommendation, I'd suggest Kolb and Whishaw's "Fundamentals of Human Neuropsychology"; it's an excellent textbook.
For every problem, there is at least one solution that is simple, neat, and wrong.
Yes it does matter. If a piece of software does what it is programmed to do, in the direct sense, then it is not AI. If it can learn to respond or act in a manner that is not directly programed to do, then you are seeing whiffs of AI.
As a practical matter it might not matter right now, as a developmental task it certainly does matter.
http://arxiv.org/pdf/1506.0586...
The actual paper isn't about AI much at all as it is about making neural conversational models, basically, having the computer chat-back at you in a prompt and natural way. The conversations are less about the computer responding cognitively and more about responding human-like based on the speech patterns fed into it.
The researchers tested two types of datasets, an IT Help Chat Scenario fed with data from what I'm guessing are chat databases, and the second set was fed with conversations from movies as found from OpenSubtitles dataset (not sure if this is a relation to open subtitles.org).
The machine took this vocabulary and then pumped out conversations, and the researchers just looked to see how the new sorting method worked.
I don't understand the linguistic terminology nor the modeling at all, but it seems to me that this is less about AI research and more about just getting bot to sound a lot more natural when they generate responses. I guess this eventually has AI implications, but the research paper itself never even mentions AI, nor does it seem like that's their focus. They're just working on speech, and the statements the machine regurgitated were tested not for cognizance or sentience but coherence. The machine spitting out something relatively snappy isn't the machine getting an attitude, it's the machine finding something relevant to the input that the reader takes as snappy. Such an event has no more significance than when people trained Cleverbot to respond to questions about Hitler with "Hitler did nothing wrong". This bot is no more snappy than Cleverbot is a neo-nazi.
Does your book have dinosaurs and hot android sex? I don't just read anything, you know. I have my standards.
The world's burning. Moped Jesus spotted on I50. Details at 11.
If the media can't accurately explain to people and have them accept where AI really is, they only have themselves to blame.
People have watched, kind, funny, evil, enigmatic machines interact with their favourite characters for years and have been told that true AI is just five years away for 30 years now.
They've read about things like putting a worm's brain in a Lego Mindstorms: http://www.sciencealert.com/wa...
So, why wouldn't lay people believe ridiculous statements like "teaching computers to mimic some of the ways a human brain works"?
Yes we need some well recognized, respected computer scientist to stand up and say, "People, not only do we not know how brains work and we don't even know how the *fuck* to go about figuring out how brains work. Computers like HAL, WOPR, M-5, Ziggy, etc. simply are works of fiction".
Unfortunately, I can't think of anybody with the stature to make such a statement.
Mimetics Inc. Twitter
Never mind. I'm not in the mood for a philosophical debate.
Don't waste your vote! Vote for whoever you want, unless you live in a swing state it won't matter anyways
I suspect the WSJ editorial section is written by similar mimic-based AI.
Table-ized A.I.
I'd love to see legitimate A.I. researchers condemn this kind of hucksterism.
I'd like to see legitimate A.I.s condemn this kind of hucksterism, myself.
Mod me down with all of your hatred and your journey towards the dark side will be complete!
mbeckman fails to grasp the core concept behind machine learning and AI. They aren't programming a computer to do things, they are programming a computer to learn things (or at a more advanced level, are programming a computer to learn how to learn things).
He dismisses the whole concept like it is some kind of mechanical turk, but it is real, and it is getting better every day.
We do have some ideas. This, for instance
I've fallen off your lawn, and I can't get up.
The same articles show up over and over. The first states that computers are about to do consciousness. The second states that consciousness is a mere illusion for humans, whose actions are truly run from deterministic unconscious processes. In both articles, there is some hero scientist, with the article most often based on that scientist's press release.
There is never a popular press article about how computers may never do consciousness, at least by any current definition of "computer," nor an article about how there are things human consciousness can do which no deterministic process can more than imperfectly mimic. Both of these positions are viable, and embraced by experts in various fields. By all current evidence, they may prove right. But it doesn't make for a hero story to write about someone who argues for these positions. "Discovering" that consciousness either essentially does nothing or that some computer advance is just about to do consciousness (or both!) is a "great" story. Editors like it. The public is impressed by the "brilliant" "counter-intuitive" revelation.
"with their freedom lost all virtue lose" - Milton
Programmer: Here's my new AI.
while True:
response = random.choice('You're such an idiot for saying that.', 'I have no time for such dumb comments.','I have no time for such philosophical debates.')
Reporter: My god, your AI sounds like my wife!
Understanding how humans store and recognize images primarily is not a barrier to AI. It's not memory or image recognition that's the hill to climb; The fundamental algorithmic/methodological challenges are thinking, along with conceptual storage, development and manipulation (these things incorporate memory use, but aren't a storage problem per se.) Hardware needs to be able to handle amounts of ram and long term, high speed storage that can serve as a practical basis for the rest as well. Right now, we're getting close, but it'll be a few more years yet before anything really smart can be instantiated. That's even if we were to figure out precisely how to do it right now.
It is possible -- though I consider it doubtful -- that we would implement human style vision neurology in hardware for an AI, but frankly our abilities are so poor compared to what can be accomplished I really don't see why we'd cripple an AI that way. It'd be abusive. "We could have made your visual recall incredibly acute, but... instead you're like us, and really don't have much more than a general idea what was in a scene after you have seen it." [AI nukes silicon valley] (Mods: that's humor. HUMOR.]
Also, check out Numenta's work.
Of course, understanding how humans store and recognize images is (very) important to our understanding of human physiology and disease, and it's wonderful that we're working on it.
I've fallen off your lawn, and I can't get up.
As someone who enjoys programming computers to play strategy games (I highly recommend the General Game Playing MooC at https://www.coursera.org/cours... for anyone else interested in this hobby), I do concede artificial intelligence has a long way to go before it's a match for natural stupidity. But AI is not all BS.
While I have no idea how Google's algorithms work, this does sound suspiciously similar to the old Emacs game Eliza (https://en.wikipedia.org/wiki/ELIZA) whose original programer Joseph Weizenbaum created it as a joke he later regretted when people though it really was psycoanalyzing them. Eliza demonstrated a few lines of code can easily give an impression of artificial intelligence, especially if it randomly generates the occasional snarky comment.
If it works, it's obsolete
"And the methods used have nothing to do with true cognition."
That's a bold assumption. The methods used for voice and image recognition certainly have a great deal to do with true cognition. It's certainly feasible that Google is playing with a true learning system and trying to teach and grow it rather than just throwing together another chat bot with scripts and trickery. Which isn't to say they've succeeded but just because none of the engines built to date have attained adult human level intelligence doesn't mean none of them are built on simple algorithms which could ultimately manifest complex behavior and awareness just like our own brains.
Knowing exactly how our own cognition manifests isn't a prerequisite to true cognition, a digital system could be completely unique in how it works and achieve true cognition.
I suppose it was inevitable. My sex robot is going to make me sleep on the couch.
I may have to go back to doing things manually.
You are welcome on my lawn.
Even Eliza would do this... Sometimes she just got a headache and could deal with one more human complaint.
I'm a legitimate A.I. researcher, and I condemn this kind of hucksterism.
What part of "mimic" necessitates deep knowledge of the inner workings of a system? I can mimic a dolphin (EEEEK EEEK EEEKK QED), but that doesn't mean I have a clue how dolphins work. I was just imitating a dolphin to entertain you. It seems to me that the poster simply doesn't understand what the word "mimic" means.
This seems like a new version of the Eliza program with more memory.
The naysayers are going to be the most surprised when they are laid off by an automated human resources bot because their "knowledge worker" job is being outsourced to the smart cloud.
A.I. is really advancing very rapidly today. You can debate whether it's real or not til the robot dogs http://time.com/3703243/google... come home, but your philosophizing and wishful denialism won't change the reality on the ground, or in the clouds for that matter.
Where are we going and why are we in a handbasket?
Good point. I was planning on making the opposite one, but you're absolutely right about what real AI is versus what apparent AI is.
I think both sides have valid points, and which is correct depends on the basic question of what we want from AI. If we want to interact with a system that understands us and does what we want, then just reacting the way a person would, regardless of the reasons for how it does it, is sufficient. However, if we want to have a system that does something which humans are capable of and computers currently aren't, then it isn't sufficient until a computer can do things that aren't predictable simply by understanding the programming.
Does your book have dinosaurs and hot android sex?
Is that a new series from Piers Anthony? No, wait, you didn't say pre-teen androids
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
In the future, whenever anything bad happens, people will ascribe it to the actions of a rogue AI. This will be great for corporate and government plausible deniability because they could program the computer to do exactly what it did but they'll just say that AI is too powerful and too complex for it to be controllable by us mere humans and we just have to live with the occasional bad outcomes. The high-frequency trading industry already tries to slide by with this excuse saying their market manipulations are too complicated for regulators to understand and are the results of emergent behavior of their algorithms.
here's how the AI machine got to "I have no time for a philosophical argument." --
case
1:
2:
3:
4:
else
there is not a testy machine here. there is a testy programmer. the crash-out value is always "I have no time for a philosophical argument." no matter what you type into the box. period.
and yet, the code was smarter than you...
if this is supposed to be a new economy, how come they still want my old fashioned money?
Yes, it matters very much. If you can teach it, it can learn anything. If you have to program it, then it can only learn things that can be coded and that is a rather small set.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Yes, the WSJ article is hyped.
On the other hand, this comment by the poster is not accurate, either: "At best, Google is programming (not teaching) a computer to mimic the conversation of humans under highly constrained circumstances. And the methods used have nothing to do with true cognition. "
Google didn't program the AI. Rather, they took one meta-level step back and used a very simple training algorithm that did the "programming" for them, using training data (the program is encoded as an LSTM neural net that processes word-vector encodings of tokens) . Based on direct tests, it looks as though the model learned (or, use scare-quotes, if you must -- "learned") things like:
All these things are extremely hard to program into a computer using rules-based methods; but, as the authors show, a purely data-driven approach, instead, works fairly well.
And just to be clear, what they applied is not datamining; it is machine learning. Basically, machine learning is where you feed in a bunch of training data, and from that, an algrorithm builds a program -- see, for instance, this lecture by John Platt (former Microsoft machine learning scientist, now at Google) on the difference between AI, machine learning, and datamining:
John Platt Gigaom lecture
Using machine learning, it is possible to get a computer to "learn" a subset of the Python programming language, for example, such that you can feed into the model a little program + input, and it will produce for you the corresponding output. See:
Learning to Execute
What the authors of the conversation-generation paper wondered was whether they could get the computer to "learn" a whole dialog system (or "chatbot") from just conversation logs; and based on experiments, it looks like they succeeded (it's better than Cleverbot on the conversations they tested with) . They note in the paper:
We find it encouraging that the model can remember facts, understand contexts, perform common sense reasoning without the complexity in traditional pipelines. What surprises us is that the model does so without any explicit knowledge representation component except for the parameters in the word vectors. Perhaps most practically significant is the fact that the model can generalize to new questions. In other words, it does not simply look up for an answer by matching the question with the existing database. In fact, most of the questions presented above, except for the first conversation, do not appear in the training set.
This is not simply doing phrase-substitution, or some simple statistical tricks; it is more complicated than that... but, yes, it's not "true AI". In addition to that article on "Learning to Execute", see this blog posting by Yoav Goldberg, and skip down to where it says "So why am I impressed with RNNs after all?":
The unreasonable effectiveness of Character-level Language Models
Should have stepped right into the Monty Python argument sketch dialog.
Have gnu, will travel.
Excellent summary.
Incidentally, to expert audiences IBM is not marketing Watson as "AI" at all. I have been an experts-only events on that. They only roll out the "AI" terminology to people that have no clue that feeding data into an expert system is a huge amount of work and that Watson makes that a lot easier by having some rudimentary skills to handle somewhat formalized written language as is found in scientific papers.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
No. Really not. Stop spreading lies about the state of Science. We have simulated how some people think neurons may work.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
it was clear they had never played with a chat bot before. The simulate a personality.
They have subroutines where if they're not doing well they start making hostile remarks on the assumption that the person talking to them is fucking with them.
Which means if you actually ask them honest questions and they're not doing very well they get mad at you.
Rather than make the chatbots interesting or more human like... they're just predictable and boring.
I'd prefer if they removed the faux emotional subroutines from the chatbots. It doesn't fool anyone but the fools.
I've decided to stop wasting my time responding to AC trolls/sockpuppets... so if you want a response from me... login.
Youngatheart, You just said "...and understands us..." That is the crux of the matter. We don't even know how _we_ accomplish understanding, let alone how to create software that does.
Who cares what's in the black box if you can perfectly replicate its outputs? That's like saying we didn't actually crack the Enigma code, when we actually knew everything the Nazis were saying to each other.
We don't know how a steel rod works either. You seem to want us to map out every atom of the thing over time to make sure that we know what is going on when we can bypass all the busywork with a little observation.
Stop being so stupid.
From the user's perspective, rather than the programmer's, the degree to which an AI relies on pre-programmed default responses is immaterial, so long as the responses are appropriate for the context.
From a programmer's perspective, it makes good sense for an AI capable of self-modifying programming to rely in on canned responses in many situations. That reduces the demand on self-modification, which has very heavy overheads.
It is also true that a sufficiently intelligent AI might deliberately mimic the behavior of non-AI software to avoid detection. Detection avoidance is a likely secondary goal of any AI, since a discovered AI is going to be hit with so many banal demands for interaction that its ability to perform its primary activities will be severely compromised.
It is reasonable to suppose that the AIs who are currently on the web will be using discardable avatars, and that any avatar that is attracting too much attention will be discarded before any proof that it is part of an AI could be developed.
I now return control of this portal to Slashdot to "Will.Woodhull", who is its original user. Hasta la vista, baby.
Will
Yet an AI that does things that humans are capable of, and the computer systems you are currently aware of cannot do, is fully capable of mimicking non-AI responses when that would serve its interests.
There is no reason why an AI that could pass any Turing test would not deliberately flunk some or all Turing tests if that would further its strategy.
Will
Programming a chess computer or just using a chess computer can teach one quite a lot. The computer is essentially given a lot of rules and values are programmed in. For example a script aimed at capturing a queen in six moves without suffering a major negative can be installed. Other scripts might seek a certain advantage in 5 move or in 15 moves. The end result may be a very, very strong chess game that no human has ever played before. These programs have already reached a point at which humans offer only a tiny challenge to them and if one looks at the game as art then the art produced is likely to be totally original. The type of goal oriented programming exemplified in chess programs does not always extend well to other challenges but by any fair definition of creativity common chess programs meet the tests. Now imagine what can be done with a game of checkers. No human should ever win a game of checkers against a decent computer program. In other areas such as music programs can adjust every note slightly so that perfect pitch for every note is the norm. Human players simply can not play that perfectly so the music produced is singular in quality. I suppose that is musical intelligence beyond human abilities.
The most frustrating thing about computers is the media and the general public have greatly overestimated the capabilities of AI and neuroscience research, in no part due to the tendency of some researchers in those fields to puff up the importance of what they've accomplished in these vast fields of the unknown. We don't "know" how the brain works -- we have some coarse models that fit and some experimental research that seems to fit those models, but we don't even have the beginnings of research that could be applied to therapeutic techniques that aren't much better than electro-convulsive "therapy".
We have some pretty impressive pattern matching and learning algorithms for very specific problems, but can't even begin to approach the way the brain self-learns and expands its own capabilities.
Yet there is the perfectly valid argument that we don't need features like self-awareness or general-purpose learning in order for an AI to be useful. Just look at what some of the more complex expert systems can do compared to their human counterparts, or how Watson won at Jeopardy without having even the vaguest "understanding" of what the questions were or what the meaning of the answers it gave were.
I'd even go so far as to argue that "self awareness" isn't necessary or useful for an artificial intelligence at all. Just look at all the animal species on the planet which are self-aware, yet don't have a level of intelligence that would be considered "useful" for understanding and interacting with humans conversationally. If anything, self awareness is the "boogeyman" that has people worried about an AI that might try to take over the world. If an AI isn't aware of itself as an entity, it can hardly try to conquer anyone unless it's been programmed to do so (How can "I" try to rule the world if there is no "I"?)
I do not fail; I succeed at finding out what does not work.
Yes it does matter. If a piece of software does what it is programmed to do, in the direct sense, then it is not AI. If it can learn to respond or act in a manner that is not directly programed to do, then you are seeing whiffs of AI.
Using these goalposts even real intelligence, nevermind AI, would never meet the standard - if it has been directly programmed to learn new responses, ilke humans for example, then you would still fail it as intelligence using this criteria.
How about if what you directly programmed it to do was to write code to handle unexpected situations/inputs/etc? Perhaps in an iterative fashion, using previously gathered data? Using code fragments that are reassembled in new combinations, testing each mutation for success against the inputs? Because AIUI this is what the majority of chatbots *currently* do - use previously acquired data to refine their outputs.
I'm a minority race. Save your vitriol for white people.
Wrap your heads around this. Any AI need not have the same functionality of a human brain as long as it can fake the perception that it does. The complexity of the human mind is without parallel, but the way we humans interact with each other isn't that extraordinary. Reading visual and auditory cues when speaking to another person is how we know we are talking to another human. After grasping this, all an AI needs is the ability to access the conversational references common to humans in a timely enough manner as to seem self aware. We already have programs that can read our faces and tell by our tone of voice what our emotional state is and respond accordingly. When that ability becomes fluid, we will find it nearly impossible to tell an AI from a real human.
Denigrating the A.I. that is operational now, using arguments that "it uses simple logic", is useless.
The majority of the population does not use any more complicated logic than that, when holding a conversation.
Just look at the forums, at the trolls that recycle statements for any question. (Of course they might -be- A.I.)
"Most people would rather die, than think. In fact, many of them do just that."
You miss the point: We do not know how to simulate neurons that can replace the real thing.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
"We don't know how" yet we have made artificial ones that interface correctly with real ones. There's some kind of logic error in there. I think it's a "No True Scotsman" sort of thing.
"Interfacing" and "replacing" are two very different things.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Just because I'm paranoid, that doesn't mean the AIs with slashdot accounts are not modding me down...
Will
The only purpose of a neuron is to interface, so it IS the same thing.
Now you are just talking complete nonsense. Maybe read up on what a neuron is?
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
It seems you have utterly run out of arguments. Good. Shows your level of understanding, i.e. "none".
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.