Slashdot Mirror
Teaching Robot Learners To Ask Good Questions
garthsundem writes "I disagree with this article's opening line: 'Within a decade, personal robots could become as common in U.S. homes as any other major appliance.' Haven't we been promised this since the 50s? But I'm fascinated by the rest — how do you teach humans to teach robots? Or, more precisely, how can you teach robots to teach humans to teach robots? The idea that designers can put a flexible platform in a robot, allowing users to determine functionality, is pretty interesting. The lead researcher for this project said, 'People are not so good at teaching robots because they don't understand the robots' learning mechanism. It's like when you try to train a dog, and it's difficult because dogs do not learn like humans do. We wanted to find out the best kinds of questions a robot could ask to make the human-robot relationship as 'human' as it can be.'"
I think the only thing keeping them out of homes NOW is no used market yet, and the high cost ($200-$5000) of entry. Basically the same problem that faced real personal computers in 1979.
Only THAT will change the real adoption of robots into the average American household.
SJW: a person who perceives an injustice, and while correcting it, commits a greater injustice.
I knew a Chinese girl who tried teaching a web-bot once. She gave up because instead of listening to her it kept trying to peak down her browse.
"That's the way to do it" - Punch
Have the dogs teach the robots.
Shut up and clean my floors!
The real question is "how to build robots that can learn." Dogs can learn. Cats can learn. Birds and bees can learn. People ... the results are not so good ...
Let's call it what it is, Anti-Social Media.
Before I can teach a robot, it must be able to feel basic emotions: pain, joy, disgust, etc. Until it is capable of associating environmental interactions into at least positive and negative stimuli, it cannot be taught anything... at least not in the same fashion we learn.
All our basic learning functions are derived from emotional responses; And much of our learning is by observing others engaging in certain behaviors/tasks/activities and noting their emotional display (positive or negative). A robot that cannot understand and express emotions in response to environmental and social stimulus is not capable of learning in a human fashion.
I'm sure I'll have a bunch of people waving their AI degrees at me and telling me their bogodynamical model of humanolinguistic interdynamics says it can be done, and I won't argue with them on it... but it won't be as efficient, and it won't be as natural to interact with.
Emotions are the key to human learning.
#fuckbeta #iamslashdot #dicemustdie
It's a Roomba, it vacuums my floor on schedule and does a decent job, also drives back to the charging post by its self, other than cleaning its brushes it's entirely autonomous.
"If any question why we died, Tell them because our fathers lied."
Gets my mail, answers my phone it I want it to, looks for music I might like, l lets me know of up coming social event. Play chess against it.
It's called a 'smart phone'. Hell, with minimal effort I can have it follow me on it's own accord.
I mean, some Lego motors and gear, and it can us it's camera to follow me. Might take a day.
The Kruger Dunning explains most post on
To some extent I disagree that dogs do not learn as humans. Its a reward and/or pain/discipline system and it works well on both. Robots on the other hand, may not feel reward or pain, so something new does need to be developed.
I will agree with his disagreement on the first line. As a retired hobbyist I am hugely directly connected to the current state of robots. They are stagnant, dead, and useless with the exception of the Vacuum cleaner bots. There are some super high end stuff going on, but it is far more akin to smart remote control. Computers are not anywhere near fast enough at present. And as long as we stick to the von neumann model for their design probably will not be for 20-30 more years.
But i have been an advocate of having moral discussion now, before it is too late. Saw a good short clip this am; http://boingboing.net/2012/03/08/disturbing-and-poignant-video.html about a robot that becomes self aware. Do we kill her, or let her be free? If we let her be free, what incentive ($$$) does anyone have to build her? If she is a slave (ie sold for profit) can we justify the treatment of any self aware being that way? And if so, why not retroactive?
Anyhow, back top the subject; we need research in that area. Today for a current project I am looking at http://mnemstudio.org/path-finding-q-learning-tutorial.htm for ideas on what is going on now. I assume it the field will advance.
slashdot troll = you make a compelling argument I do not like the implications of.
I knew a guy named Ivan and he's pretty out there. I am told that Ivan taught his nephew a little something.
Every time he wanted a cookie he was to go up to his mother and say "Trust fund".
I am wondering just how elastic these learning mechanisms will be, or will they really be pre-programmed routines where it basically learns to recognize your voice, and you can set a few parameters for which room is the living room or kitchen, but not much else.
If my wife wanted a mindless automoton that could that could take instructions in natural terse language, and can be ignored the rest of the time, she would have gotten a husband. ..... oh, wait....
"We wanted to find out the best kinds of questions a robot could ask to make the human-robot relationship as 'human' as it can be.'"
"WHAT IS YOUR NAME:"
"WHAT IS YOUR QUEST:"
"WHAT IS YOUR FAVORITE COLOR:"
From that point on its all downhill.
Sure, dogs don't learn the same as humans do, but that's because they have their own naturally-developed instinctive way of doing things. Robots don't learn the same as humans do because their technical capabilities are different, but the way they learn is still human design. The way a robot learns can be changed, and it will most certainly evolve to adapt to new technologies, but no matter what there will always be human design behind it all.
We need to backup a step and first look at how are we going to teach robot engineering students to teach robots to teach the human customers to teach the robots?
Of course, Big Software will get involved and next thing you know, Robots will do things like, "I see your trying to mop the floop, can I be of assistance?" and then proceed to not only mop the floor but the inside of the fridge and the cat, while surreptitiously installing a company maid-bot in the upstairs bedroom.
try { do() || do_not(); } catch (JediException err) { yoda(err); }
1. Fetch a beer from fridge.
2. Walk the dog to do its business.
3. Wash dishes.
4. Mow the Lawn.
5. Clean House.
6. Make robot programmable and able to share/sell programs ala app store.
People don't want to program stufff and you are not going to change that behaviour. Just make their life easier and give the innovators the tools to accomplish this and the rest takes care of itself.
Beware of those who profit off the docile and persecute the unbelievers.
Three words: Global Thermonuclear War.
Now I'm really feeling very very depressed.
Marvin
A robot, as tech has evolved, is essentially a set of limbs of some kind, or an autonomous platform of some kind, operated by a computer. Since a robot capable of doing various tasks that could help people would essentially be like a person, that is, capable of opening doors, opening jars, washing dishes, etc., it would need to be possessed of physical strength capable of doing these tasks, making it potentially dangerous. Also, computers tend to be literally minded, and still can't quite keep up with nuances of natural human speech, which would cause such a robot easily to become dangerous to the people around it. Like a giant moron, lumbering around your home, not quite understanding what is being said around it.
Take for instance "do the dishes". We know what it means, but would a robot? The logic might go:
on "do the dishes" locate all dirty dishes.
(Robot sees one of your guests holding a plate with food on it.)
identify plate: _8.5 inch dinner plate_ == IS plate
evaluate item_plate: plate has food items on it
take plate to kitchen to clean
Guest: HEY, I was still eating that!
Or, robot notices the dish on the counter with the plastic fruit in it, and cleans that. Of course, it's a decorative dish, and the robot breaks it, incorrectly assuming all dishes you have are equally durable stoneware.
Future versions of the robot's software would need to have steps added to identify when a plate is dirty, and when still in use. Even people, way more nuanced than robots, who work in the profession of clearing away dirty dishes sometimes have trouble with this one, and for safety's sake, ask "are you done with that, sir?" or "can I take that out of your way?" in restaurant settings.
This is but a single example. Would you trust your computer to have a set of limbs attached to it, lumbering around your house? Of course not, we're happy with them sitting where they are, gathering dust, waiting until the next time we want to surf the net, type a letter, watch a video, play a game, chat with friends, listen to music, etc., NOT walking around your house getting into trouble.
The lack of daily work robots in our homes are not because of lack of advances in ROBOTICS, they're because of lack of advances in A.I., ARTIFICIAL INTELLIGENCE. If we had a computer capable of thinking at or near our level, cheap enough for people to have in their homes, we'd all have robots. Maybe... one other very good reason NOT to have this fantasy vision of the future come to pass is very simple. If robots and A.I. get that advanced, they would have little trouble making the leap from domestic servant to replacing us in the workforce wholesale.
That's why we don't yet have artificially intelligent robots wandering around everywhere.
Hobbyist robotics seemed to be stuck in the mid-1980s for several decades. The hardware got cheaper, but the systems remained about as dumb as 1980s industrial robots. Lego Mindstorms is an example. In the last few years, though, there's been more movement. Hobbyist robots are starting to use SLAM, vision processing, laser rangefinders, Kinect devices, and machine learning. All that stuff can be done on low-end hardware now. (At the $1000 level, anyway. We're not down to $100 yet.)
There's now enough code available that people can use those technologies without learning the underlying math. That's what's making it happen.
I'm pretty sure I'm not a robot, but can someone teach me how to read that article... Look, when it comes down to it, it should have nothing to do with humans adjusting to how we can best teach a robot and adjusting our teaching methods to fit. It should be about designing machine intellegence to learn in a "human" way. If it's the former then no, "robots" as they say will not be predominant in the household. If its the latter and we design the "robot" to fit our needs and learn as we learn then yeah, they will actually be useful and become ubiquitous. But until then we are going to be stuck with the roomba, which unfortunately is classified as a "robot", and may well take over every living room in the country, as per the article... Much to the fanfare of the press, to me it's still not a "robot".
ogglelog
I'd like to hear more people asking good questions...
What this world is coming to - is for you and me to decide.
Or, more precisely, how can you teach robots to teach humans to teach robots?
Then perhaps we could learn to teach humans to teach humans to teach humans.
Robot teach YOU lesson!
This sig is not paradoxical or ironic.
Judgement Day
Any person using FTFY or editing my postings agrees to a US$50.00 charge
Real human maids break shit too, you know, and steal far more than robots (which, as we all know, only steal old people's medicine). Robots don't need to be better than people at these chores in order to become consumer goods, they just need to reach the point where they don't break too much shit too often - the rest will be taken care of by a Roomba-style no-robot zone.
We're very close to robots that I'd trust "lumbering around my house" - maybe 5 years out, and all that's missing is, yes, advances in robotics. (Think about Sawstop - we can already make a freaking table saw safe against accidental human contact.)
Socialism: a lie told by totalitarians and believed by fools.
...humans are oh-so-good at teaching humans.