School of Computer Science THE UNIVERSITY OF BIRMINGHAM CN-CR Ghost Machine

-- Local Arrangements --

Computing at School (CAS) 2014 Conference
University of Birmingham 21st June 2014

Notes on the possibility of using a variant of the Turing Test
for teaching.

Possible alternative plan for talk at the CAS 2013 Conference:
For a while I considered talking about the Turing Test
(Which Turing did NOT propose as a test for intelligence.)

The presentation could include background information on the Turing test
why it's a mistake to think Turing proposed the 'imitation game' as a serious
test for intelligence, and the relevance of chatbots to education in
computational thinking.

The Turing Test event on 7th June 2014.

After the testing process at the Royal Society London, it was announced that
one of the competing chatbots, Eugene Goostman, had been the first to pass the
(so-called) Turing test.
This produced an enormous furore. Here's a tiny sample of comments
on the world wide web:
(Click here to see lots more!)

The idea that Turing proposed his imitation game as a test for machines with
intelligence is a myth, believed mainly by people who have never read his 1950

Some people think that there could be a behavioural test for intelligence but
the so-called Turing Test is not stringent enough, so they propose alternative,
more demanding tests. I think they are all missing the inappropriateness of
attempting to provide such a test. Could any one test enable us to decide which
types of animals are intelligent? The search for a behavioural test for a
machine being intelligent is as misguided as a search for a behavioural test for
a machine being computational.

Instead we need a deep theory of varieties of intelligence and their properties
and means of implementation, and an investigation of which forms were produced
by biological evolution and which forms can and cannot be implemented on digital
computers, neural computers, chemical computers or some combination. This would
extend the deep theories developed since the early 20th Century of varieties of
digital computation, what they can and cannot do, varieties of implementation,
what they can and cannot explain, and much more. A project to develop such a
deep theory based on a study of achievements of biological evolution is
summarised here.

I suspect that that is one of the things Alan Turing would have worked on if he
had lived longer, extending his pioneering work in 1936 that contributed so much
to the theory of discrete computation, as well as his later work on developmental
processes in organisms.

A detailed discussion of the possibility of a behavioural test for intelligence,
and why such a thing would be of little scientific or philosophical interest,
can be found here:

Adam Ford interviewed me about the Turing Test, its limitations, and alternative
approaches to understanding intelligence, on 12 June, and posted the video here:
I have an extended discussion here (work in progress).

Relevance to Computing at School
The supposed Turing test is really just a test for whether a text-only chatbot
can, for a while, deceive certain types of human, into thinking the chatbot is
also human. That may be an interesting and mildly (sometimes more than mildly)
entertaining type of project, even if its scientific and philosophical
importance has been hugely over-rated. However, there is an important
educational function.

Teaching children to develop chatbots using a succession of increasingly
sophisticated programming methodologies to enhance their chatbots could be a
major contribution to computing education, teaching students many forms of
computational thinking of varying depth and sophistication.

For example, they could start off using only long lists of explicit conditional
instructions (e.g. 'if input = "Hello" then ...'), then progress to using
pattern matching, then grammars, parsers, semantic interpreters, fixed
databases, changing databases for simulating actions, rule-based systems for
combining different sorts of knowledge in generating a reply, logical inference
systems, constraint-propagation systems, neural nets, evolutionary computation,
connections to microphones and speakers, connections to cameras and
manipulators, embedding in a mobile robot, and many others.

From my experience of teaching AI to beginners at Sussex University many years
ago, I suspect it would help to attract more girls into deep forms of programming
and computer science, as well as contributing to the education of computationally
informed psychologists, linguists, neuroscientists, biologists, philosophers,
and others.
Here's a 30 year old chatbot used in that teaching process.

Peter Millican (Philosophy Department Oxford University),
who also talked at the CAS 2014 conference has a web site
Elizabeth: An Educational Chatterbot for Windows
With downloadable code:

Maintained by Aaron Sloman
School of Computer Science
The University of Birmingham