During the AAAI conference I discovered scientists of the Socially Intelligent Machines research lab from the Georgia Institute of Technology working enthusiastically on Simon, the robot which we previously have covered at Chatbots.org. Here’s a video of my personal experience:
Simon is a humanoid robot. He has a torso, arms, a head with eyes, ears and a mouth. Simon’s right hand is designed to receive objects, grabbing objects and releasing objects. Simon’s left hand is designed for pointing at objects.
Although my experience with Simon is impressive in itself, it’s interesting to realise that this demonstration is not about voice recognition, not about voice synthesis, not about robotics, not about object recognition and not about expressing emotions, but solely about learning.
The objective of Simon is to store objects in bins. The counter contained several objects varying in size, color and shape. The floor had various bins, in various colors, sizes and they were numbered. But what should go where? It’s up to the human, Simons’ teacher, to decide how Simon should organize all objects into the bins. For example: You may decide that all red objects should go into the red bin, but you can also opt for small objects to go into the red bin and large objects into the green bin. It’s a matter of teaching Simon by setting an example and repeating this consistently.
The learning sequence starts with: ‘Simon, I want to teach you something’. Simons’ memory will be reset, so he will start from scratch.
Then you say: ‘Simon, take this’, he says ‘sure’, and you hand him an object by putting it into his hand.
Simon takes the object from your hand and says ‘let’s see’. Simon brings the objects close to his eyes and his ears resemble the color of the object.
Now he asks where to put it: ‘Put it in bin 1, Simon.’
This sequence needs to be repeated with different objects.
Then, you can ask: ‘Simon, do you have a question?’
Simon points to an object and asks: ‘Should I put that in bin 1?’
If you say: ‘That’s correct Simon’, he will take it and put it in bin 1.
If you hand him an object and ask: ‘Where should it go?’ Simon will answer: ‘I’ll put it in bin 2’.
If you haven’t given Simon enough information or provided inconsistent information (unconsciously or on purpose to confuse him) he might say ‘I have no clue’. However, in some cases, when he estimates that he’s close to the correct answer, he will just suggest a bin.
It’s just a matter of trial and error.
That’s how Simon learns. That’s how humans learn. Consistent input is essential.
Can you imagine a future robotic sales man at a retail shop explaining to you in detail which products are for sale? Simon will pave the way…