From: David Hart (firstname.lastname@example.org)
Date: Sat Aug 10 2002 - 04:12:29 MDT
Last year I modelled complex environmental behaviour (algal blooms in
the Murray-Darling river system in north NSW, Australia) in a visual
flow-based programming system (POWERSIM). The model took the form of
numerous sub-models specified and defined by a group, each implemented
by a different programmer and integrated (rather easily I might add) by me.
Based on extrapolation of the experience, I'm fairly confident the
exercise above could easily take place on scales of larger and more
It probably is indeed inefficient to apply visual programming to heavily
algorithmic environments, but it might be vastly better suited to other
programming paradigms. Can this idea be dismissed so quickly?
I again urge readers to ignore the emotive statements and arguments on
the website and focus on the idea of a temporal-signal-based approach to
software and the technical merits of the proposed implementation (COSA).
I agree that pure signal-based systems would not be the best tools for
most software tasks. I think the challenge comes in figuring out where
they could best be applied (which may not always be obvious) and then
figuring out how to integrate them with existing algorithmic systems.
A new question for thought: It seems that a complex signal-based system
that would be difficult for humans to conceive could be conceived by
sufficiently specialised AI software (e.g. a codic cortex). When, in any
programming paradigm, does it become impossible for humans to inspect a
self-modifying AI? When the times comes, won't it be very difficult for
AI programmers to 'let go'? What are the ethical implications of
allowing an AI 'out of the box' when programmers can no longer inspect
the AI (either due to complexity or deliberate obfuscation)?
Samantha Atkins wrote:
> [snip] The truth sold here is partially useful for some types of
> problems but is seldom useful by itself except on relatively small
> subsystems. Visual Programming has been tried in many systems and
> again is usually found unworkable except for fairly small problems.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:40 MDT