Nuclear bomb. Harmful or non-harmful? Let's ask the Bikini Islanders
or the ex-inhabitants of Hiroshima.
All ideas are harmful as much as they help. Its just the usage, like
any tool, that gives them a smiling face or a bloody eye to face
history with.
> Take that same neural network, and give it the ability to 'sense' some
> environment, and also give the ability to communicate to some simple
> queries. Then, allow it to go into its 'dreamlike' state for a while,
> and finally, ask it some questions to find out how it interprets its
> own flurry of activity. I would be interested in seeing what it says.
Why repeat a process that's already been done? The AI autonomous
agent architecture, Maximus, watches Mac-based email software as a
user operates it to learn their patterns and tendencies, attempting to
learn well enough to second-guess their needs. At a given time,
typicall at night when usage is low, Maximus goes into a `learning
consolidation state,' in which it considers the states its observed
during the day, compares them to older states and reworks its tables
to update for recently observed behaviour. Does this not mimic
`dreaming' well enough and hint at one possible service dreaming may
provide?
[As a sidreal note, typically you cannot `ask' a NNet about its
activity; they are `black boxes' for the operations they pursue, which
is one of the drawbacks of such a system. Systems of emergent module
interaction or of symbolic nets can generally tell you every what and
wherefore of their choices, however. Maximus is one of these
systems.]