Re: Humans "unique" social

The problem of machine understanding was raised by Prof Searle in his
Reith Lectures for the BBC back in the 80s. He made the claim that
while a machine could translate from English into Chinese, it could
never understand what it was doing.

This struck me as being short of the mark. To check it out I sent a
description of how machines can understand to the late Donald Michie,
of the Turing Institute. He confirmed that I was right, and that the
concept was worked out by Terry Winograd in his 1970 MIT PhD thesis,
'Understanding Natural Language'.

I believe this field is known as 'deep modelling'. Essentially, the
machine needs to take a sentence and model it.

e.g. 'The cup of water is on the table' is built as a mathematical
model, with the rules of physics etc, perhaps as it is in a computer
game. If you then give the model more information, say that the table
is on a boat at sea, then the computer will be able to predict where
the water is going etc. Such modelling would, of course, require the
computer to have an extensive database of real life objects and their
properties.

If you ask the machine to output the information, it will be able to do
so in another language - without recourse to any token turning etc. It
can think and it does understand what it is saying, as much as we do.


.

Quantcast