> We don't need the global human knowledge for an AGI, just human-level
knowledge: 10^9 bits, $ 1 million to $10 millions (using your data).

>
> Actually $10K to $100K per worker replaced, because 99% of the 10^9
> bits you need are shared with other people and can be copied. There is
> 10^7 bits that makes you unique. In order for an AGI to serve you
> properly, it needs to learn these personal preferences.
>

Ok, 10^9 is like a fixed cost, plus a 10^7 per person. I can agree with
that number if we are building a AGI to have manking knoledge on day 0.


>
> I assume 99% sharing based on a U.S. Labor Dept. estimate that it
> costs $15K (1% of lifetime earnings) to replace an employee on
> average. Notice that in the first case I am describing how much other
> people know (or should know) about you, and in the second case, how
> much you uniquely know about other people. Summing over the
> population, it's the same knowledge.
>

I agree. But ...the whole point of an AGI singularity is that you only need
to reach humal-level inteligence AGI so AGI can improve itself.
Can you agree that building a human-level AGI will cost around 100 billion ?

If so, this 100 billion is the initial development cost.
What's the operational cost of this system ? Computer maintenance staff,
electricity and replacement hardware.
I don't have any information about supercomputer total running costs but I
can make some gross estimates :

Titan hardware costs US$ 97 M.
Let's assume that all parts needs to be replaced in 10 years. That's 9.7 M
in hardware costs per year.
How big is the support team ? 10 full-time workers ? That's ~ 1 M per year
How about the electricity bill ? US$ 100/h , another 1 M per year
How about the physical space ? Titan needs 404 m2, we can have this space
for a sub-million price, we can assume 1M for simplicity.

The total running cost is bellow US$ 13 M/yr, that's ~ US$ 1M/month
Assuming a computer works 24h/day and a regular worker 8h
Assuming that a regular people get easily distracted and computers can have
~ 100% working time
I guess that 1 AGI supercomputer can replace the total work output of 5
people.
Assuming 5 people working together  on a problem produces like 3 people
working on different problems.
An AGI machine can replace a team of 8 workers.

8 workers for US$ 1M/month is doable for companies like intel , ibm or
google.

Leonardo Stern



>
>



-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to