herwin's Journal: The Future of Robotics

If it won't be Vinge's Singularity, where do I think we'll
be in 50, 100, 150, 200, etc., years? I suspect I can probably
extrapolate based on the historical pattern of technological progress.
It won't be magic, but given enough engineering and hard
work, it may look like magic.

Let me take an area that I know something about--biomimetic
robotics. We're currently trying to map out the locations of higher
mental phenomena and use that to put together an architecture
for a robot brain. One of my PhD students (and a lot of other
people) are trying to understand the processing performed by individual
modules in the neocortex. Yet another of my PhD students is looking
at the connectivity of the auditory cortex to see how these processing
modules might be wired together to correlate sound sources to
elements of the acoustic scene. I'm looking at how various modules
in the auditory brainstem do their processing to feed those cortical
processes. At the current rate, we'll have this worked out for
the entire brain in about 50-75 years, at which point, if the
physicalist hypothesis is correct, we should be able to build
an intelligent robot (or AI). We're trying now.

As an aside, what is the physicalist hypothesis? It's basically
monism--the belief that there is no separable soul. When you're
dead, you're dead (unless God deigns to resurrect you in the body).
That was mainstream Christian belief until the fourth or fifth
century and continues to be mainstream Jewish belief. I'm a monist.

Whether that AI can think like a human may possibly depend
on whether we can do time and spatial scale transformations like
I have a suspicion that real brains do. I'm speculating, but there's
evidence that the brain can change the timing of event sequences
by a factor of 10 or more when they're being simulated or replayed.
There may be a similar capacity for spatial scales. That's probably
doable at that point. So conscious AIs capable of non-linear thinking
in 50-75 years.

Also if the physicalist hypothesis is correct, we should eventually
be able to transfer minds to AIs. This will require a detailed
mapping of every synapse and every neuron. That's something like
10^11 neurons and 10^4 synapses per neuron = 10^15 total synapses
in a liter-plus of brain = 10^3 cubic centimeters. That basically
would involve mapping the brain with a resolution of 10^-6 meter.
That's just engineering. Allow 20 years for the technology to
be developed--that suggests mind to AI transference in 2100.

The reverse engineering will be more complicated, as it will
require local control over cell growth. On the other hand, nanosurgery
would be useful in the near future, so we're probably talking
about 2100 for robot-to-AI transference as well.

Perhaps it will be Vinge's Singularity after all.