Re: distributed AI was Re: computronium prime-oxide

Mark Crosby (crosby_m@rocketmail.com)
Fri, 20 Nov 1998 06:14:26 -0800 (PST)

---Hal Finney <hal@rain.org> wrote:
> The source of the program is irrelevant for
answering this question. If any program has the property that running it on a Turing machine produces consciousness, then Turing functionalism is true. If it turns out that the only way to make such a program is by running an interaction with the environment, fine; likewise for any other way of creating the program. The source doesn't matter. All that matters, for this question, is whether running a program can produce consciousness.

But if running an interaction with the environment is the source of the program, then it does matter.

Peter Wegner's 9705 _Communications of the ACM_ article, "Why Interaction is More Powerful Than Algorithms" (see http://www.cs.brown.edu/people/pw) provided an overview of the interactionist view:

"An object's operations are not algorithms, because their response to messages depends on a shared state accessed through nonlocal variables of operations... controlled by unpredictable external actions... Input streams of interactive machines are not expressible by finite tapes [as in Turing machines], since any finite representations can be dynamically extended by *uncontrollable adversaries*... nonterminating reactive processes, like operating systems, cannot be modeled by algorithms... Interactive histories of distributed systems ... consist of nonsequential events that may have duration and may interfere with each other...

"Raw interactive power is captured by interactive identity machines (IIMs) that output their input immediately without transforming it...
Interactiveness (openess) is nonmonotonic in that decomposition of systems may create interactive unpredictable subsystems, or equivalently, composition of interactive systems may produce non-interactive algorithms... Giving up the goal of complete behavior specification ... makes respectable software-engineering methods of incomplete partial system specification by interfaces".

Wegner's and similar approaches were discussed at a 9711 Interactive Foundations of Computing Conference at Washington University in St. Louis. Here's some more heresy against the Turing Church from Ronald Loui's lecture there (available at
http://siesta.cs.wustl.edu/~loui/jp4.html):

"Alan Turing and Alonzo Church are canonically held to be at the foundation. Computing is about algorithms. Algorithms are essentially Turing machines, infinite storage and finite control, sequential, state-transition specifications... The celebrated idea of Edsger Dijkstra is that a program just is a function transforming inputs to outputs. Programs are just cleverly compressed tables of logarithms, no matter how fancy they pretend to be...

"Then Peter Wegner, before his naked emperor, had the temerity to announce that algorithms are not what most of computing is about. Instead, the management of interaction better describes in abstract the day-to-day computing practice. It is as scandalous as the observation that scientists do not follow scientific method, or that economic behaviors are neither Bayesian nor utilitarian, or that logic is about notation, not reasoning."