Scientists coax computers to think like people

By Will Dunham
WASHINGTON, Dec 10 (Reuters) - For artificial intelligence
and smart machines to really take off, computers are going to
have to be able to think more like people, according to experts
in the field. Researchers are now making important progress
toward that goal.
Scientists said on Thursday they had created a computer
model, or algorithm, that captures the unique human ability to
grasp new concepts from a single example in a study involving
learning unfamiliar handwritten alphabet characters.
This work as well as research like it has the twin goals of
better understanding human learning and developing new, more
human-like learning algorithms, New York University cognitive
and data scientist Brenden Lake said.
"We aimed to reverse-engineer how people learn about these
simple visual concepts, in terms of identifying the types of
computations that the mind may be performing, and testing these
assumptions by trying to recreate the behavior," Lake said.
The algorithm was designed to make a computer able to learn
quickly from a single example in the way people do.
"You show even a young child a horse or a school bus or
a skateboard and they get it from just one or a few examples,"
Massachusetts Institute of Technology computational cognitive
science professor Joshua Tenenbaum said.
Standard algorithms in machine-learning require tens,
hundreds or even thousands of training examples to yield similar
results, Tenenbaum said.
In the study, computers boasting the new algorithm and human
subjects were presented with selected characters among a data
set of about 1,600 handwritten characters from 50 alphabets from
around the world. They even included a fictional alien alphabet
from the animated TV show "Futurama."
Among other tasks, the human subjects and computers were
directed to reproduce various characters after being given a
lone example. Human judges were then asked to identify which
characters were reproduced by a computer. The judges found the
work produced by the computers to be virtually indistinguishable
from that of human subjects.
University of Toronto computer science and statistics
professor Ruslan Salakhutdinov said he hoped this new work would
help guide progress in artificial intelligence by leading to
next-generation intelligent machines "that hopefully will come
close to displaying human-like intelligence."
The same approach used in the study might be applicable to
machine learning for many other tasks like speech recognition
and object recognition, Lake said.
The research was published in the journal Science.