I agree that in its early stages, the SI will likely interact with
humans in the way
you describe. The reason that I use the term SI instead of AI is that my
hypothesis
doesn't depend on the exact nature of the SI. As I said, it's likely to
start as
some type of human-computer collaboration and augment itself from that
base. My feeling is that the SI will rapidly reach a point at which it
can re-derive knowledge
ab-initio faster than it could get the same knowledge from a human.
Alternatively, the SI will almost certainly solve all of the
then-current human research problems, specifically including nanotech
and uploading. It can then offer to host an upload
of any human that wishes to upload. When enough humans accept the offer,
the amount of useful knowledge known only to "meat" humans will cease to
be very attractive.
Even if the SI has goals other that increasing its intelligence, most
such goals
are more readily achieved by a more intelligent SI. Therefore,
increasing its intelligence is likely to be an intermediate goal for any
of the other goals.
I've argued that the same is true for today's humanity. Probably the
fastest way
for us to get to nanotech and upoading is to develop the SI. IMO, the
quickest way to develop this SI is by concentrating on software tools
and visual representation of
information, leading to easier-to-use computers and to human-computer
collaboration.
This will lead to the singularity within ten years.