Re: Hardware Progress: $397 (fwd)

From: Dani Eder (
Date: Thu Mar 28 2002 - 10:48:03 MST

> However, computers are better viewed as amplifiers
> of human abilities
> than as replacements for humans. One human plus a
> computer may replace
> many humans at certain tasks (like generating 10^18
> or so steps in a
> Markov Process and sampling them appropriately), but
> I cannot forsee a
> time when the human is entirely removed from this
> equation until
> (possibly when) computers are truly self aware and
> can dream up useful
> or even useless work to do on their own.

My employer right now buys a lot of human thinking
power (represented as the present value of my
salary and benefits) and a little computer power
(represented by the 500MHz PC I'm typing on).
When an economic crossover occurs, I would expect
the ratio of spending to shift in favor of more
computer and less human. I don't expect the
human component to disappear, and I don't require
'real AI' for the shift to happen. For example,
my employer (Boeing) may develop a genetic
algorithm/monte carlo/optimizing wing design
program that replaces a bunch of engineers with
one engineer and a big linux cluster.

A real example is my local supermarket installing
4 self-checkout units with one cashier to oversee
them. More computers, less people.

The point of the particular crossover I am tracking
is that for 'real AI' to get deployed as more
than a reasearch project, it needs to be competitive
with the cost of what it is replacing.

> Finally, it isn't clear why these particular hard
> disk and memory ratios
> where chosen as part of the metric,

Lacking a 'real AI' algorithm that works, I used
the complexity of the human brain as an estimate
for how much storage will be required. The numbers
I used were 10 bytes/synapse x 10^4 synapse/neuron x
10^11 neurons/brain = 10^4 Terabytes, which is 100x
the 100 Tflops processing rate.

For human-like reaction times of ~1/4 second, I
assumed that you need 8 bits of active memory
for each 32 bit floating point operation, which
works out to 1 byte/Flop. To get consistent cost
trend data, I want to keep the ratios of processor
power, memory, and storage the same.

> To conclude, this seems to be a moderately naive
> metric for systems
> comparison and I wouldn't recommend that it be
> widely adopted:-)

As soon as someone comes up with a better estimate
of computer requirements for running a 'real AI'
I'll adopt them, but in the mean time I consider
it useful to track computer cost trends regardless
of whether I've got the mix of components exactly

Do You Yahoo!?
Yahoo! Movies - coverage of the 74th Academy Awards®

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT