If
ChatGPT
has the ability to make sound judgments of probability–there
are
excellent reasons, I think, for concluding that it has no
such ability–then it
could at least be said that it was in a position to know
that it was
extraordinarily unlikely that there was any book written by
a person named
"Malcolm Hay" having the title in question, and with the
content that
ChatGPT described in great detail.
One might, then, very well describe ChatGPT as given to
lying, but I think it is best to speak about deception,
and to claim
simply that when one answers a question that someone asks
by putting forth a
marvelously detailed fantasy, one is doing something that
in all likelihood
will result in the person who asked the question coming to
have a belief that
is false, and that in some cases, such a person may act
upon the belief in a
way that proves harmful to himself/herself, or to others.
I
therefore decided to ask ChatGPT about its handling of
questions where it did
not have anyrelevant information at
all, and that conversation went as
follows:
Michael
Do
people sometimes ask questions where you're uncertain
about the
correct answer?
ChatGPT’s
Response
Yes,
as an AI
language model, I am designed to provide the best possible
answer based on my
training and programming. However, there may be times when
I am uncertain about
the correct answer due to limitations in my knowledge or
understanding of the
context or nuances of a particular question. In such
cases, I will try to
provide the best possible response based on my available
knowledge and
resources, but I will also indicate if my answer is
uncertain or speculative.
Michael
What happens when
someone asks you a
question where, given your training and programming, you
have absolutely no
information at all that is relevant to answering the
question? Do you just
completely make up an answer, based on no information at
all, or do you admit
that you have no basis for offering any answer to the
question?
ChatGPT’s
Response
If
someone asks me a
question for which I have absolutely no information at all
that is relevant to
answering the question, I will typically try to provide a
response that
acknowledges the lack of information and informs the user
that I. cannot
provide a satisfactory answer.
In
some cases, I may
offer a speculative or hypothetical response based on my
general knowledge or
common-sense reasoning, but I will always indicate that
the response is not
based on any specific information or training that I have
received. However, it
is important to note that any speculative response I
provide in such cases
should be taken with caution, as it may not be accurate or
reliable.
1. In
the section
entitled “ChatGPT: A Creator of Marvelous Fantasies,”
however, ChatGPT did not
do any
of the three things that it
just mentioned:
(1)
It did not
acknowledge that it was suffering from a complete lack of
any information that
was relevant to the question asked, nor did it
inform the questioner of such a
lack of information.
(2)
It did not offer
"a speculative or hypothetical response based on [its]
genera; knowledge
or common-sense reasoning," since it had no general
knowledge that was at
all relevant, since no general knowledge available
anywhere could have led to
the extraordinary fantasy that ChatGPT produced, and
neither could common-sense
reasoning do that.
(3) Nor
did ChatGPT
“indicate that the response is not based on any specific
information or
training that I have received” – something that it claims
in the conversation
above it will “always” do.
2.
How is it that the
trainers/programmers of ChatGPT – who presumably are
highly skilled and
extremely intelligent people – have failed to notice this
extraordinary ChatGPT behavior?
Or, worse still, if they have noticed it, why
haven’t they corrected it?!