Re: Halting Problem for Humans

There is a crucial difference between the case you propose and the
halting problem. The halting problem is a well defined question while
your questions to Peter and Daryl are not.

Each of them is circularly defined. Trying to complete the text of the
question to Peter (which I suppose you think contextually completed) we
would obtain:

'Will Daryl answer 'yes' to the question whether you will answer 'no'
to the question whether Daryl will answer 'yes' to the question
whether...?'

Of course, you can block the infinite regress by substituting 'this
question' for the infinte phrase 'the question whether you will
answer...' and obtain:

'Will Daryl answer 'yes' to the question whether you will answer 'no'
to this question?'

Now the self-reference only makes the circularity still more evident.

The halting problem contains no circularity in its definition because
it poses a purely syntactical or mechanical question with no semantic
or intentional dimension. Mechanical or merely syntactical states of
affairs (states of affairs about machines, symbol strings, etc.) are
always possible objects for any thinking subject whoever, they are
always out there, so to say, already objectified.

In contrast, and using phenomenological terms, we can say that no
intentional act of thinking can be its own intentional object; call
'PNS' that proposition; so according to PNS, it is not the case that
any intentional act is a possible object for any thinking subject.

This is why your questions to Peter and Daryl cannot be well defined
questions for none of them: trying to answer them correctly would force
them to consider their own attempt to answer and this would be
phenomenologically impossible.

Maybe we've met here an essential difference between thinking behaviors
and machines: only the latter are always possible objects for the
former.

Regards


Daryl McCullough wrote:
The Halting problem is ultimately about the failure of prediction.
No program can have perfect ability to predict the future behavior of
all other programs. Here I'm going to illustrate the problems with
perfect prediction using humans instead of programs.

Suppose that we have game in which two players, call them "Peter" and
"Daryl" are asked yes/no questions. The rules stipulate that the only
legal answers are "yes" and "no" (if neither answer seems appropriate,
then the players must respond with silence). Assume that the players
are intelligent and will attempt to answer the questions correctly,
but most importantly they will never intentionally answer a question
incorrectly---if they are not sure about the answer, then they should
refrain from answering, rather than risk making an incorrect answer.

The rules are that both players get to see both questions. Afterwards,
they are taken to separate answwer rooms to figure out their answers. No
interaction between the players is allowed. So there is no possibility
that one player's answer can influence the answer given by the other.
Each player must answer only using the knowledge that he brings into
the answer room.

Okay, that's the background. Now here are the questions:

Peter is asked: Will Daryl answer "yes"?
Daryl is asked: Will Peter answer "no"?

I think it is clear that there is nothing paradoxical about
the questions. Peter must use his knowledge about Daryl in
order to answer the question, and Daryl must use his knowledge
about Peter.

We can analyze the possibilities here (and presumably Peter and
Daryl can do the analysis).

1. If Daryl and Peter give the same answer (either both "yes"
or both "no") then Daryl is mistaken, and Peter is correct.

2. If they give opposite answers (one "yes" and one "no") then
Peter is mistaken, and Daryl is correct.

3. In light of 1&2, if both answer, then one of them is mistaken.

So, it is clearly impossible for both Daryl and Peter to
have perfect ability to predict the future behavior of the
other.

This formulation seems to be completely about the limitations
of knowledge and reasoning ability. There is nothing paradoxical
going on. And in particular, there doesn't seem to be the
escape clause: Daryl (or Peter) knows the answer, but he can't
say it. If he knows the answer, there is nothing at all preventing
him from saying it. One player's answer has no causal effect on
the other player.

This game is similar to the following one-player game: Someone
asks Daryl to answer the following question with "yes" or "no"
(and to make no other responses)

Will the next answer you give be "no"?

In this case, Daryl cannot correctly answer the question, but
it is possible for him to *know* the answer (and keep it to
himself). He can refuse to answer at all, in which case, he
can know that the correct answer is "no".

In my two-player version, there is nothing preventing a player
from saying the answer if he knows it.

--
Daryl McCullough
Ithaca, NY

.

Relevant Pages

  • Re: Halting Problem for Humans
    ... natural language in a way that is very difficult for the TMs ... "Daryl" are asked yes/no questions. ... then the players must respond with silence). ... Peter is asked: Will Daryl answer "yes"? ...
    (sci.logic)
  • Re: Halting Problem for Humans
    ... The halting problem is a well defined question while ... Imagine that Peter and Daryl are both robots, ... Will Daryl's response to string S be "yes"? ...
    (sci.logic)
  • Re: Halting Problem for Humans
    ... Peter is asked: Will Daryl answer "yes"? ... of knowledge and reasoning ability. ... that the stipulation you are making is not an additional rule of the ...
    (sci.logic)
  • Halting Problem for Humans
    ... then the players must respond with silence). ... Peter is asked: Will Daryl answer "yes"? ... have perfect ability to predict the future behavior of the ...
    (sci.logic)
  • Re: Halting Problem for Humans
    ... Peter is asked: Will Daryl answer "yes"? ... of knowledge and reasoning ability. ... rewards and punishments, we can say that each is rewarded ...
    (sci.logic)