Re: AI: Relative difficulty (Was: SCI:BIO: raw genome length)

Anders Sandberg (
Sun, 19 Jan 1997 20:48:13 +0100 (MET)

On Wed, 15 Jan 1997, Eliezer Yudkowsky wrote:

> Computer vision *is* easy... relative to general planning!

:-) Well, that depends a bit on perspective. In the happy 60's, everyone
thought that general planning was just a problem of graph-traversing and
computer vision was a graduate student problem. We know somewhat better

In some sense both problems can be seen as equivalent; I'm buying more
and more into Furschters model of the brain divided into an
input/perception part and a planning/motor part.

> So I stand by my statement that if we have the software, computing
> power, and raw intelligence to draw lines from DNA to the brain, hacking
> up a virtual world takes an afternoon. The environment accounts
> probably for most of our personalities, but very little for our
> problem-solving abilities (as opposed to strategies).

Hmm, I'm not so convinced about this. Remember Piaget's various stages,
where the child seems to lack certain abilities before certain ages, and
gradually develop concepts such as object constancy (things exist when I'm
not looking at them) or conservation (if I move things around but do not
take anything away, then there is the same amount of stuff). Most likely
these have to be reinforced by experience, and may be completely due to
the environment (at least the later). The way to test it is of course to
bring up a child in a virtual reality with different laws...

> Though I could be
> wrong; visual visualization might derive its rules from the environment,
> with the genes just setting up the visual cortex to learn from it.

A classic example is the formation of ocular dominance stripes in cat
visual cortex. In normal cats, the axons from the lateral and
contra-lateral LGN at first branch out homogenously over the cortex, and
then after the kitten has opened its eyes, tend to die off clumpwise
(probably due to darwinnowing of some kind) in a way that leads to the
zebra-stripe-like dominance bands. But if the poor kitten doesn't get
visual information this doesn't happen, and the bands never form =>
incomplete vision.

There seem to be quite a few parts of the brain that has to be trained
over time to develop well; the basic circuitry exists, but then parts of
it have to be removed or adjusted in complex ways.

Anders Sandberg Towards Ascension!
GCS/M/S/O d++ -p+ c++++ !l u+ e++ m++ s+/+ n--- h+/* f+ g+ w++ t+ r+ !y