- Can a Machine Think?
- Both "machine" and "think" are unclear terms.
- Turing suggests we replace the vague question about what
thinking is, with a kind of test. He first gives an example
of a test to determine, through written questions and answers
alone, the sex of someone. He then suggests we generalize,
and use such a test to determine whether the thing being
tested can think.
- Note, this test, which now is called the Turing Test,
is widely celebrated and attempted. Each year, for example,
Loebner Prize Contest occurs to run such tests.
- QUESTION: what features of intelligence (by any of our
colloquial undestandings of that word) does this test seem to
discover? What features might it overlook?
- As for "machine": Turing famously gave the best
definition of this. Today we call his account a "Turing
Machine." Turing describes in this article a digital
computer -- all computers are Turing machines, so you are
already familiar with these.
- It is possible to create a universal Turing machine. This
was a big insight in Turing's day, but what it means is simply
that in principle computers with the right set of instructions
can be then programmed to do anything that any other such computer
can do. Again, we are all quite familiar with this. We surf the
web both on our PCs, Macs, and Linux boxes.
- Turing next considers some objections
- It would be terrible if machines thought
- Computational Limitations
- Arbitrary disabilities
- Lovelace's Objection
- Animal brains are not discrete state machines
- Human life is not rule-governed
- Turing points out that we could perhaps find a way to make machines
- For those interested, Andrew Hodges, a respected
Turing biographer, maintains a Turing web site.