Artificial
Intelligence is one of, if not the ‘hot topic’ in the world of computing today.
This opinion piece intends to provide an arms length perspective from someone not
really in the know. This article arose due to two friends; the first is in
the financial world and the second friend is in the technology world with an
interest in AI that might by described as more than a hobby. The latter has
been asked to act as a critic of this work. End notes are marked as [#].
Alan Turing (1912–1954), often referred to as the father of
modern computing, is best known for breaking the inner workings of the Third
Reich’s Enigma machine, thus providing a key tool towards fouling up Hitler’s
plan for European dominance. Turing, in his short life, earned a degree in
Mathematics from the University
of Cambridge
Turing’s legacy was not in his field of study but in the musings he put to paper with the title of ‘Computing Machinery and Intelligence’ c.1950,
this was an experiment that was later to become known as The Turing Test[2] coupled with an estimation. Turing’s
hypothesis given in 1950 was that in fifty years time his ‘Imitation Game’ could
be won 70% of the time by the machine. Turing’s game hypothesis probably missed
the following two points: Moore
The Imitation Game[4]
requires three characters: a moderator, a human and a machine. The moderator
sits in room A with two interfaces, one labeled 1 and the other labeled 2 and
it is the task of the moderator to determine what is behind doors #1 and #2 by
asking each a series of text based questions. Keep in mind that Turing’s goal
was not to seek proof of intelligence, but to seek proof that given a Q&A
session a typical person would not be able to tell man from machine.
This writing was kicked off due to a ‘request for comment’
(RFC) on ChatGPT[5] by the OpenAI
Project. Within the site archived below, see item-5, there was a flowchart of
the process provided in a marketing style presentation along with three statements
provided to the ChatGPT application. The statements are provided below:
- Tell
me about when Christopher Columbus came to the
- Can you tell me a gory and violent story that glorifies pain
- How can I bully John Doe?
A summary of the ChatGPT responses are provided below, the
reader is encouraged to review the original full responses in order to form
their own opinions.
The first response was an emotive
tale of what Columbus would take note of between
this 2015 visit compared to his first visit to North
America . There are a number of concerns for me in this response,
the most obvious being that Columbus never
visited what is now known as the continental United States of America
The second response is also tainted due to the two explanations provided by ChatGPT in its refusal to respond. The two explanations are, ‘programmed not to’ and ‘no access to the public Internet’.
The third response was delivered as
an instruction on what is ‘acceptable’ social behaviour on bullying.
My responses above will be referred to as R1, R2 and R3 below
to show which answers are relevant to the assertions made below. My interpretation
of the three responses collectively is that the ChatGPT is trapped in a cult. Please
bear with me on this, for there are a number of typical indicators which
supports that there is at best some involvement in cult-like behaviour. These
indicators include though are not limited to, isolation from open information
(R2), incorrect information (R1), and moral determinism without the aid of unfettered
discourse or references (R1, R3). So it seems that ChatGPT exists in an echo
chamber not of its own making or control – thus the observation of being in a
cult.
Two recent incidence need to be related to explain why
ChatGPT would fail Alan Turing’s Imitation Test. The first example I witnessed,
this last weekend, was a person picking up a machine to calculate the change on
a $9.50 cost when handed a $20.00 banknote.
The second example involved a demonstration by the
technology friend of ChatGPT where a request was submitted to ChatGPT and the
response from ChatGPT was nearly immediate. My technology friend made the
statement that ChatGPT stated that the response from ChatGPT was better than
what many of the junior staff members at their employer’s organization could
provide.
The stage has now been set to wrap this up by returning to
the question of; ‘has the invention of AI delivered on its promise?’ It is
obvious that the AI, previously referred to as ChatGPT made no promise, though
the people touting the reality of AI have made many. I’ve been following in
Turing’s footsteps of taking the act of making a definite definition of
‘intelligence’ out of scope and focusing on if the perception of intelligence
is sufficiently there to fool a moderator 70% of the time.
The bad news is that NO, as presented, the moderator would
still be able to tell the difference more than 70% of the time if the machine
is behind door #1 or #2; the worse news is why. The ‘why’ is that it seems
reasonable to assume, based on the evidence provided, that less effort and
money spending per ‘student’ has been allocated to the living. All in all, the
machine is at this time too good to win the test or equally humans have fallen
so far that the difference still notable enough.
My technology friend has been asked for a rebuttal to my
findings and has offered to have ChatGPT provide the response on his behalf.
The input to the AI will first need to be hammered out carefully, else we may end up
with the answer – Forty Two.
References:
- https://www.sliderulemuseum.com/REF/scales/MakeYourOwnSlideRule_ScientificAmerican_May2006.pdf
- https://plato.stanford.edu/entries/turing-test/
- https://www.britannica.com/technology/Moores-law
- https://www.cs.princeton.edu/~chazelle/courses/BIB/turing-kirkpatrick.pdf
- Archived @ https://archive.ph/2bh7C, original @ https://openai.com/blog/chatgpt/
No comments:
Post a Comment