[LINK] Fwd: Facebook AI is stupid

[Bernard Robertson-Dunn]

It's a long weekend Friday. This might lighten the mood.

I wrote it in the early 1990s only a few years after the Expert
System/AI/5th Generation/Multi-Function-Polis hype.

Artificial Intelligence - A State of the Art

For many years philosophers, scientists, engineers and people with
nothing better to do, have been trying to develop a machine that
displays some of the characteristics of human intelligence. These people
believe that a suitable combination of hardware and software can be made
to behave like an intelligent human being.

Some time ago I decided to jump on the bandwagon and study the cognitive
processes of mankind and the duplication of the amazing feats of thought
that we humans can achieve.

The starting point was to make a study of the literature on Artificial
Intelligence and my co-workers and I discovered that, in spite of many
claims by workers, the ultimate goal of a thinking machine has not yet
been attained.

We then conducted some of research into Artificial Intelligence and
discovered that the reason for the failure of this technology to achieve
any really useful results is because the problem has been tackled in the
wrong way. The aim of Artificially Intelligent machines should be the
ultimate goal not the first goal. It is a mistake to attempt to achieve
the ultimate objective in a single step. It is rather like the Wright
Brothers saying "OK we got it off the ground for twenty feet, now for
the Atlantic".

We have decided to take a radically new approach to the problem. We
believe that it is necessary to define a set of sub-goals and try for
the attainment of each in turn. This might eventually allow me to
achieve the difficult aim of Artificial Intelligence.

These sub-goals have been defined, along with the criteria that each
must meet. We have also indicated where machines of each type might be
utilised.

Artificial Ignorance The first sub-goal defined was Artificial
Ignorance. Ignorance here is defined as the ability to always get the
answer wrong. An Artificially Ignorant Machine would consistently reply
with incorrect information. Some of my co-workers believe that a
computer that has not been switched on, already exhibits this behaviour
but the general consensus is that this is cheating and that we need to
create a machine that actively tries to get the right answer but is
wrong. This is different from being wrong by omission or inaction.

A machine of this type could be used in weather forecasting, in
predicting the national budget or by television stations when covering
general elections.

Artificial Stupidity The second sub-goal defined was Artificial
Stupidity. In order to duplicate this behaviour we must be able to model
a class of behaviour epitomised by that of someone who believes the
promises of politicians at election time, project managers at review
meetings or computer software salesmen. Stupidity is different from
ignorance in that facts are known but actions are not logical.

If one of these machines was asked if it was willing to take on the job
of creating a computer system, whose design was specified by an end user
and do it on time and within budget, it would answer, yes. This is
obviously a stupid answer. It is also the type of behaviour often
displayed by management consultants.

Another example of stupid behaviour that we must be able to duplicate is
shown by people who believe that the computer system they have just
specified will help to make their job easier or (even worse) be useful.
We were tempted to put this behaviour into a special category called
"When I get my new system, everything will be wonderful", but decided to
leave it in Stupid.

There has been much discussion as to whether a very large spreadsheet,
used to make important company decisions, displays Artificial Ignorance
or Artificial Stupidity. We have decided that this is a special case and
should be labelled "Really Stupid".

Artificial Incompetence Sub-goal three is Artificial Incompetence. We
believe that the test for this is to train the machine in a particular
skill, test it in another and have it believe that it can perform the
second skill as well as the first. This is rather like an accountant
building computer systems or an engineer running a marketing company.

Another example of incompetent behaviour is that shown by a previously
intelligent person standing for public office. There is a hierarchy
involved here. It is believed that standing for Federal Parliament
demonstrates less incompetence than standing for State Parliament.

Our first Artificial Incompetence machine would be used to help in
writing tenders for government departments. The process would go
something like this.

1    Write the tender

2    Feed the tender into the Artificial Incompetence machine

3    If the machine rejects the tender, re-write it and go back to step 2

4    When the machine accepts the tender, submit it.

In this way we could duplicate the evaluation procedure that seems to go
on in government departments. This should give us a competitive
advantage over other companies who rely on logic and common sense when
responding to such tenders.

Artificial Intelligence The final sub-goal is Artificial Intelligence
itself. Unfortunately we have been unable to find many examples of true
intelligence that we can model. Because of this, determining the test
for Artificial Intelligence has been very difficult. Work is continuing
in this area.

Expert Systems There has also been heated debate on the topic of Expert
Systems. We have been trying to determine into which of the four
sub-goals this type of system might fit. Some felt that, while not
exhibiting intelligence, they did go some way towards attaining the
ultimate goal. Others thought that a special category should be created,
along the lines of "I'm not letting that machine diagnose my illness. I
bet it can't even spell schizophrenia".

The conclusion was that Expert Systems themselves are examples of
Artificial Stupidity but that the people who claim that these system are
intelligent, fall into the "Who are they fooling?" class.

Post Script. After a lot of deep thought, late nights and many bottles
of wine, we have now decided to abandon the whole project. We have
finally seen the light. We realised that the thought that we could
actually duplicate intelligence is a demonstration of our naivety. Not
only that, but naivety was not even included in our sub-goals, neither
real or artificial.

We also had the thought that, if it was possible to create an
intelligent machine, what would become of we humans?. We would all
probably be retrained as Chartered Accountants while the machines did
the interesting jobs and had all the fun. We have decided to go back to
creating incompetent computer systems, a demonstration of Real
Stupidity. When we get to intelligence ourselves, we might try again.

Then again we might not.

-- 

Regards
brd

Bernard Robertson-Dunn
Canberra Australia
email: brd at iimetro.com.au