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.

Comments

"No idea how... the brain works"

[Improv]

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.

Re:"No idea how... the brain works"

[Improv]

The WSJ article links a paper from some researchers at Google:
http://arxiv.org/pdf/1506.0586...
The WSJ article isn't particularly good either; they misunderstand what's actually going on in the research, which seems to be about conversational modeling (a "weak AI" type of research, the "understanding" being very shallow). They point out a few applications of this kind of work though, and that seems pretty solid/useful. (It doesn't approach the goals of "strong AI", those being actually modeling semantics and deeper reasoning)

Re:"No idea how... the brain works"

[Improv]

Probably not - weak AI is typified by directly encoding domain knowledge on human capabilities into state machines, not typically meant to be neuroplausible or human-like. I believe the substrate here is wrong - real organisms learn (either as individuals or through generational building/encoding/selection towards instinct) how to do these things, and that knowledge is integrated. I don't think it'd be easy or likely that weak AI research methods will produce an integrated being with all these capabilities.

I'm sticking my neck out a bit here though; I'm not sure that weak AI research would be useless. Sufficiency versus usefulness is a complicated topic.

Also, my research was in neuroscience (led by cognitive modeling), not AI. It's a neighbouring field, but take what I have to say with at least a grain of salt.

Re:"No idea how... the brain works"

[Improv]

I agree that we don't have the full picture. That's not what mbeckman was claiming though, and saying "we know very little" because we don't have a particular achievement is an unjustified conclusion.

Re:"No idea how... the brain works"

[mbeckman]

Improv,

We are in the "cargo cult" phase of neurological research. Our level of cognitive understanding is like that of the South Pacific islanders who made bamboo replicas of WWII airplanes and radios after the GIs left. The islanders said to themselves "We must be very close to reproducing these wonders, because our airplanes and radios looks so much those of the GIs. Now we just sit back and wait for the magic goods to come out of the airplanes and wise voices to come out of the radios."

If you really don't know how little we understand about the brain, NY Times science writer James Gorman can explain it to you:

http://www.nytimes.com/2014/11...

Re:"No idea how... the brain works"

[thinkwaitfast]

You seem like someone informed. You don't belong here is more like it.

Re:"No idea how... the brain works"

[paskie]

(I work in this area of research.) You are right, the paper is about just a sequence-to-sequence transformation model that learns good replies for inputs but is not actually "understanding" what is going on.

At the same time, we *are* making some headways in the "understanding" part as well, just not in this particular paper. Basically, we have ways to convert individual words to many-dimensional numerical vectors whose mathematical relations closely correspond to semantics of the words, and we are now working on building neural networks that build up such vectors even for larger pieces of text and use them for more advanced things. If anyone is interested, look up word2vec, "distributed representations" or "word embeddings" (or "compositional embeddings").

If you already know what word2vec is, take a look at http://emnlp2014.org/tutorials...

Re:"No idea how... the brain works"

[Megol]

We have no actual understanding on several important parts of the working of a brain, we don't know how memory works, we don't understand how decisions are made (or even what it means if one want to get philosophical) and we don't understand how an intelligent being get the feeling of self.
There are a lot of theories and clues of how some mechanisms work (parts of how some levels of memory works, parts how neurons and synapses work, part of where and how some functions of the brain works, and even some mechanisms of self awareness). But that doesn't mean we actually understand it as a brain.

Mental problems and physical problems in the brain aren't really treatable at the moment. What is done is the medical equivalent of carpet bombing with drugs that have little (if any) experimental proof of helping, for some cases they help - for some not. Side effects can be serious in many ways.
  One of the most efficient and oldest treatments available is that of ECT (Electro Convulsion Treatment) which again is a carpet bombing equivalent that causes a (somewhat) controlled seizure in the brain. But even that is really done without a thorough understanding of the working mechanisms - what is known is that it is often successful for a variety of mental problems, that it works quickly compared to drugs and some details like that of signaling substances being released during the seizure and that neural growth is increased in some parts of the brain. But again understanding of a few pieces of a puzzle doesn't mean we can even begin to comprehend the puzzle as a whole. How does it work? Anybody that claims to know is a fraud.

Re:"No idea how... the brain works"

[mbeckman]

I use the term "cargo cult" because it's accurate. I'm reasonably well read in neurobiology and biochemistry, and participated in a fair amount of early neural network implementation. But the burden isn't on me to "know what I'm talking about". The burden is on anyone, including as you, claiming science knows anything about how the brain works. You're making the assertion, so you must provide the proof. I'm happy to consider any examples you have of how the cognitive function of your choice operates.

Re:Teach vs Learn

[Todd Palin]

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.

Both the submitter and WSJ got it wrong

[Vokkyt]

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.

Re:Both the submitter and WSJ got it wrong

[mykepredko]

I would argue that the process we have gone through here is a demonstration true intelligence at work.

The original reporter looked at the article, didn't understand a piece of it and asked an intern specializing in technology what this was about.

The intern couldn't be bothered, saw that it was a computer responding to human input and said it was "Artificial Intelligence".

The submitter read the article and keyed on the comment about this being a machine learning, which they feel is impossible.

Most /.ers (me included) responded to the submision and railed on about the ignorance of the media and the great unwashed.

One poster actually read TFA and pointed out that it has nothing to do with the article, submission and most comments.

I don't know how the hell we expect to create software that follows a process like this.

Re:I wrote about this!

[MightyMartian]

Does your book have dinosaurs and hot android sex? I don't just read anything, you know. I have my standards.

What I'd like to see...

[grasshoppa]

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.

Re:Strong AI is Bullshit

[davester666]

Just waiting until someone at WSJ googles for funny stuff Siri says. They will be SHOCKED at how rude she can be.

Re:There are ideas. Here's one.

[fyngyrz]

by "some ideas" you mean "some theory".

Yes, of course. What else did you think I meant? It's an idea. It's not a certainty. I'm not sure what your point is. Care to elaborate?

When I say "no idea" I mean literally we have no demonstrable understanding of any one single cognitive function of the brain. Any brain

You might have meant that, but writing "no idea" didn't (and still doesn't) actually say that. The statement was made that we have no ideas. We do, in fact, have ideas.That was the assertion, and that is my answer.

Human brains? We've got nothing.

Human brains are not what are at issue here, but even so, that statement is incorrect. We have made progress at the small scale (see Numenta's work) and there are multiple ideas out there that presently have significant merit. Personally, as someone working in the field and conversant with a lot of what's going on in the technical sense, I have a fairly high level of confidence that we're much closer than the popular narrative would have us believe. Am I right? We will see. :)

WSJ article: Dear Mr. Journal:

[swschrad]

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...

Re:Teach vs Learn

[goose-incarnated]

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.