On Tue, Aug 15, 2023, 7:44 AM John Rose <[email protected]> wrote:

> I suspect human K complexity is larger than most people realize.
>

It's about 10^9 bits of long term memory (based on recall tests for words
and images) and 10^8 to 10^9 bits in our DNA. The compressed size of human
genome is 5 x 10^9 bits, but only 8% is functional and the rest is not
easily compressed because it accumulates random mutations that don't get
removed by natural selection. The coding parts are more repetitive.
Evolution can only add 1 bit of information per population doubling
generation.

It is not hard to build humanoid robots out of metal and plastic. The
challenge is the power supply. Lithium battery capacity is about 300 watt
hours per kilogram. The record is 711 Whr/kg. But fat from food or stored
in the body has 10,500 Whr/kg. Human metabolism is 100 W resting and 1000 W
during hard, sustained work while producing 250 W of mechanical power. That
is doable with daily battery swaps and motors running at 80% efficiency,
vs. 22% for muscles.

The problem is computation. A human brain sized neural network needs 10
petaflops and 1 MW of electricity. We can't reduce that by making
transistors smaller because we are already near the limit where feature
sizes are smaller than the 5 nm spacing between silicon dopant atoms. The
brain uses 20 W. We need nanotechnology to achieve that, moving atoms
instead of electrons.

If we assume that Moore's law doubles global computing capacity every 2
years, then it will take about 100 years to catch up to the 10^37 bits of
DNA and 10^31 transcription operations per second in the biosphere. We can
get there 50 earlier if we don't care about making robots self repairing
and self replicating like our bodies. Our 10^13 cells carry 10^23 bits of
DNA, which is 10^8 times the number of synapses in the brain.


------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T772759d8ceb4b92c-M593ffd3deebc82cd3e0db128
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to