DeepMind founder thinks AI will need to build on neuroscience
DeepMind founder thinks AI will need to build on
neuroscience
BY JAMES WALKER JUL 21, 2017
A leading artificial intelligence expert has said he
believes research in the field must start to build on existing areas of
neuroscience. Describing the two areas as having a "long and intertwined
history," he said AI will need to learn human qualities.
Demis Hassabis founded the artificial intelligence
startup firm DeepMind. The business was sold to Google for $650 million in
2014. This week, Hassabis published a research paper in which he explores the
connection between AI and neuroscience. His view is that the two fields are
already aligned but closer work is required to continue AI's development.
Most modern neural networks are predominantly focused on
problems that are solved by mathematical solutions. Areas such as speech
recognition and image identification are some of the most promising areas of
artificial intelligence. These rely on mathematical algorithms which, while
complex, are relatively simple for experienced developers to build.
The harder challenge is in creating an AI that reflects
the way the brain works. This would allow qualities such as creativity and
imagination to be imbued in the system. The development of neural networks with
these attributes could give rise to a new generation of artificial intelligence
that can balance its mathematical prowess with moral and humanitarian concerns.
In the paper, published in the Neuron journal, Hassabis
explained his belief that more study of neuroscience is required to facilitate
the coming breakthrough in AI. The report, written with three co-authors, makes
the case that we should first understand our own intelligence before attempting
to replicate it in machines.
Hassabis' argument is based on two main principles. He
believes neuroscience will serve to inspire and then validate new developments
in artificial intelligence. The evolution of conscious AI with human qualities
will eventually come to reflect biological mechanisms that have a proven role
in enabling intelligence.
"The benefits to developing AI of closely examining
biological intelligence are two-fold. First, neuroscience provides a rich
source of inspiration for new types of algorithms and architectures," the
paper explains. "Second, neuroscience can provide validation of AI
techniques that already exist. If a known algorithm is subsequently found to be
implemented in the brain, then that is strong support for its plausibility as
an integral component of an overall general intelligence system."
The remainder of the paper focuses on the parallels
between research and artificial intelligence. Arguing that a closer
relationship between the two will be essential if AI's to play the societal
role many expect, Hassabis suggested a "virtuous circle" should be
setup where advances in either field benefit the other. He acknowledged that
being an expert in either area is difficult though, let alone trying to
identify ways to make your work relevant in the other.
Comments
Post a Comment