Re: The relevance of impredicativity

[Tim Gwinn <***>]

Steve,

Comments interposed below.


SJ:
> I think there is too much focus on impredicative
> modelling in discussing Rosen's legacy to the
> exclusion of more immediately useful areas of
> exploration.


TG: What ?more immediately useful areas of exploration? do you intend? And
how does Rosennean complexity play a role?


SJ:
> I see Rosen's legacy as falling into these broad
> categories:
>
> 1) importance of relational modelling
> 2) insufficiency of state-based Newtonian paradigm,
> 3) anticipation,
> 4) epistemology of modelling and measurement
> 5) the argument that life required impredicative
> models, closed causal loops etc.
>
> I feel like only item (5) has to do with causal loops,
> impredicativities and the line.


TG: I could not disagree more. Rosennean complexity is about systems which
possess impredicativities. In causal terms, this means causal loops.
1) Relational models are important because they can model impredicative
organization in complex systems.
2) The insufficiency of the Newtonian paradigm is in its inability to model
impredicativities of complex systems.
3) Anticipatory systems with internal predictive models are complex because
they possess impredicative organization.
4) Impredicativity is present within the epistemology of modelling and
measurement, intimately tied to the subject-object boundary.


SJ:
> So we can't model impredicativies on a computer. So
> what? We can't model real numbers either. We just take
> an approximation and in many cases it turns out good
> enough.


TG: "Good enough" for what? If all someone wants to do is simulate behavior
and create a computer program that will generate quantitative predictions,
then they probably have little if any reason to hang around this discussion
list. It depends what questions you want such an approximation to answer. If
you want to ask "why?" questions of them, they will generally not provide
useful answers.


SJ:
> I think Rosen's anti-computer ire was mostly directed
> against certain media lab types who make silly robots
> with a camera attached to a neural net and then
> pontificate whether or not it is conscious.


TG: Which way do you want to have it? Above you say approximations are good
enough, but here you seem to say they are not. Why are the AI
approximations - their "silly robots" - not good enough for discussing
consciousness? Answering this kind of question leads to Rosennean
complexity. This is not an "anti-computer ire". Rosennean complexity is
simply a logical conclusion related to the kinds of inadequacies inherent in
using such mimetic approaches.


SJ:
> Who is to say that no kind emergence is possible when
> highly complex software system reaches certain
> threshold?


TG: What do you mean by "emergence"? Threshold....of what parameter or
value, specifically? No amount of operating a clockwork will make it
something other than a clockwork: no finite amount of computation produces
anything that is not a computation, not an algorithmic result. Maybe your
definition of 'emergence' falls within there somewhere; but for me, a
symptom of 'emergence' is when a result is not entailed algorithmically from
prior states (or more generally, as Rosen stated it, that emergent phenomena
are not possible to describe using a single set of states [FM p. 91]). That
means the system would be complex, which computer algorithms are not.


SJ:
> Yes it will not be LIFE but that does not
> mean it will not be INTERESTING. Just look at the
> Internet. As a human/computer chimera it might well
> lead to a meta-system transition of a novel kind.


TG: Well, you are talking apples and oranges here. A chimera of [complex
systems (i.e., humans) + computer algorithms] will always be complex,
because it already is thanks to the human component. This entails nothing
about non-complex computer algorithms themselves being able to become
complex. They can't.

Regards,
Tim