Social Robots!
There’s a huge article in this week’s NYT Sunday Magazine on Social Robots!
They link to a bunch of videos of some of MIT’s fancy new robots, including Leo, who is pretty freakin awesome.
There’s a huge article in this week’s NYT Sunday Magazine on Social Robots!
They link to a bunch of videos of some of MIT’s fancy new robots, including Leo, who is pretty freakin awesome.
(1) Does Leo respond only to one voice, or can it expand the directions it learned to other vocal inputs?
(2) How can be possibly know that “turn all buttons on” is different from “turn 2 buttons on”? (because they were given in the same context - he had a 50-50 chance of guessing which one was which)
I am pretty sure the lady speaking in the video is Leo’s trainer, so I think it is calibrated to respond only to her voice. However, voice recognition software is advanced enough that this eccentricity of Leo shouldn’t be that big of a limitation to its future applications.
However, your second question is a good one. You can see the first time she asked to turn on all the lights, he guessed that “turn all buttons on” meant just turning on the green button; and Leo was in fact wrong about that, and had to be corrected. I presume that the same holds in other generalizations- if It had tried just turning on two buttons, it would have performed incorrectly, and would have to be corrected. Eventually, the hope is, it can make the right sorts of generalizations after enough practice.
The novelty, however, isn’t just that it can make the generalizations. Rather, the novelty is that it can learn these generalizations from verbal input and indication alone: the trainer just needs to (verbally) correct it if Leo makes a mistake. So the idea here isn’t to create a fully intelligent robot, but to create a robot that learns on the basis of our own (human) intuitive social learning cues.
Good questions, though.
If you are interested in bots that can accept multiple speech commands from different users, you might want to check this out:
http://video.google.com/videoplay?docid=-3386251792468042381&hl=en
Also, the MIT Lab is always worth checking out:
http://robotic.media.mit.edu/
is leo programmed to know colors? i’m sure he is. how exactly does he tell the difference?
i thought it was cool that he stopped after pressing each button to look up at the woman. but what exactly was that? it looked like he was looking up for confirmation of some sort. and also he nodded. was he programmed to nod after commands or what?
Leo sort of reminded me of that famous horse that people thought could think and understand (who could answer questions like 2+2 using clicks of his hoof). We talked about him in a psych class, and it turned out that the way he was responding to the questions by hooftaps was looking at the people’s eyes and where they were focused. If the audience’s eyes were focused on his hoof, he would tap it, but if most of the audience looked up at him he would stop. So, people thought that he was really understanding how to do math, but really he was just reading peolpe’s responses to his taps. parth’s comment about Leo’s looking up sort of looks similar to the horse. (If I remember the name if the horse I’ll post it)