Slashdot Mirror
MIT AI Acts Childish on Purpose
garibald gave us the link to an article in the electronic Telegraph about researchers at MIT who have built an interactive AI robot called "Kismet" that is as cute as any
George Lucas character, and is supposed to function on the emotional level of a
human two-year-old. The cuteness is not gratuitous. As the article makes clear,
there is a sound, scientific reason for it. (For pictures, and more technical depth than the Telegraph story, you may want to go directly to Kismet's Web Page.)
Wow.
Sometimes I think computer people Just Don't Get It.
Then I see this.
Now, alot of us geeks might whine and moan. "Oh, it's so cute. Die Jar Jar Die!" "The future of technology lies with...Terminator Furby." Whatever.
Outside of the cold, sterile world of gigabit routers and tab completion, a convincingly emotive device has been developed--artificially.
The importance of this is not to be underestimated. Large chunks of government policy are designed to protect animals with emotive properties. As a classic comedy routine went, "Sometimes I think the animals just all got in a line. 'What are you?' 'I'm a seal.' 'You're cute, honk that horn, we'll make sure nobody beats you with a club. Now what are you?' 'Cow. Moo.' 'GET ON THE TRUCK! You're a baseball glove.'"
Pet Rocks were quite the subversive satire on this maternal instinct.
Tamagotchi began the trend, and Furbies proved the consumer attraction, but I think it's the progeny of Kismet that will whip up quite a media frenzy. Who wouldn't think that the media has been waiting to report that a machine built by man doesn't want to die? It's a pent up desire; one that will be released at the first credible moment.
And along comes Kismet.
The key to Kismet really is that it spoofs emotion. Think about it for a second. The MIT guys "sniffed" humans with cameras, copied the protocol stream down to the transitions from one emotive signal to another, and (most importantly) parsed enough of the incoming emotistream to generate a seemingly interactive experience.
Shades of Eliza? Obviously. Eliza spoofed "Rogerian Psychology", where a person does nothing but ask the minimum amount of questions to keep you talking. Eliza took some aspect of human psychology and looped it on itself to create meaning to the user at the least possible computational cost.
Give Eliza and Kismet a love child equipped text to emotive speech and speech to text, and toss in a degree of anger, rage, and fear if the robot "believes" 1) it is to be deactivated(put a video sensor near the off switch) or 2) it is ignored for excessive periods of time, and a non-zero part of the population will believe it alive and as worth protecting as a cute baby seal.
Is it just me, or is it scary how much Hackerthink(spoof, parse, etc.) fits so many different situations? If you can talk to a single person in biotech for more than twenty mintues about their job and not realize they're utter hackers, you aren't paying attention. The same applies to psychologists. How many psych papers read like security bulletin? ("God refuses to patch. F1zRR has released Prozac 1.0 to compensate.")
If Kismet ever goes mainstream, the psychologists are going to have a field day. The technological revolution eliminated the need for menial workers. Kismet, scarily enough, could make shallow friendships far more awkward. "We don't do much more than talk like Kismet."
All this will go on until the media decides to flip public opinion around on its ear(thus making everybody tune in) saying "Are we nuts? This Kismet thing is NOTHING compared to humans! Deux Ex Humana!"
That's just my thoughts. It's late. I'm tired. I have to get up in four hours. Joy.
Send me comments. Or don't.
Once you pull the pin, Mr. Grenade is no longer your friend.
but the interaction of automata with humans is something that can and has to be judged within a human's value system as well.
Human can and will judge anything and everything, of course, but that doesn't mean that the judgement is valid. In this case it's not valid, because there is no cultural context in the automaton+human universe against which a system of values can be built.
Judging something out of context is only for the brain-dead, just like judging that an apple is a very poor quality orange. Not having such a context available is not a valid excuse for applying a previous inappropriate context in its place. The logical approach is to let a new context develop naturally over time, lead where it may.
But with 95% of humans happy to follow "moral leaders" like sheep, we're in for a field day of a-priori moralistic universalism, especially when nanotech begins to change the rules built up over millenia.
"The question of whether machines can think is no more interesting than [] whether submarines can swim" - Dijkstra
I wonder if they have implemented a 'familiarity' algorithm in the robot,...
I visited the MIT AI Lab in early June (I took a one week course -- it was fun). I asked Cynthia this question, and the answer is no, the robot does not remember. It identifies facial features and reacts according.
I believe she said it was on the to do list.
Steve M
This makes me worry that our first consumer-friendly robots are going to be like Marvin, the Paranoid Android.
Personally, I'm going to arrange some Kismet faces to make a "How Do You Feel Today?" montage...
Emotions are a function of one of the oldest parts of the brain, and almost certainly are hard-wired evolutionary adaptive "knee-jerk" responses to external stimuli.
Would it be any more fake if Kismet was programmed to find "baby faces" (big eyes, rounded face, etc) cute, than humans finding Kismet cute?
This isn't spoofing. This is the way emotions work.
Ben
I suppose that means it won't be long before they train it to post to Slashdot.