Re: Why invest in AGI?

From: Richard Loosemore (rpwl@lightlink.com)
Date: Mon Jan 23 2006 - 21:03:12 MST


Not any time soon: nobody has the detailed information necessary to
actually *do* that. Assembling a computer with the same number of
processing units and wires as a brain is pretty easy: making all the
processors behave exactly as they behave in a particular brain, and
making all the wires exactly follow the neural pathways in that same
particular brain, are devastatingly hard with current technology. I
mean, the word "hard" is a silly understatement here! I don't think it
will be done before a superintelligent AGI is built.

Which is why I castigated the IBM brain simulation project in a previous
post. You hear a lot about the fantastic things achieved with fMRI
technology: don't believe the hype. It's just marketing BS.

Place a call with your friendly local neuroscientist today: ask if the
IBM project is going to supply details of every neuron, the route of
every axon path, the strength of every synapse and the spatial structure
of every dendritic tree in a brain. Listen carefully to the answer.

Richard Loosemore

H C wrote:
> Well, it is arguable that IF you simulate an entire human brain, then
> you could create an AGI, in a sense (I guess you would be simulating an
> actual person). Thus, your Singularity take-off happens.
>
> -hegem0n
>
>
>> From: Richard Loosemore <rpwl@lightlink.com>
>> Reply-To: sl4@sl4.org
>> To: sl4@sl4.org
>> Subject: Re: Why invest in AGI?
>> Date: Sun, 22 Jan 2006 21:18:02 -0500
>>
>> There is only one problem with your story: I very much fear that it
>> is not true that if we got brain-power hardware, we would get an AI.
>>
>> If you gave Microsoft a set of four Blue Gene-L machines, they *still*
>> would not be able to deliver a bug-free version of Word any time in
>> the next century.
>>
>> We probably have the hardware power right now. What we lack are the
>> right theoretical approach and software techniques. More
>> particularly, I think we lack the right software-construction tools.
>>
>> You might respond that it does neverthless make a compact story to
>> give to an investor: I'm not sure, though, because I think they know,
>> intuitively, that it has a false ring to it.
>>
>> Richard Loosemore.
>>
>>
>> Dani Eder wrote:
>>
>>> My simple story for potential investors:
>>>
>>> The human brain has 100 billion neurons, each with
>>> 10,000
>>> synapses firing at an average of 100 Hz, for a total
>>> synapse
>>> firing rate of 100 x 10^15/sec.
>>>
>>> A modern CPU chip (Athlon 64 X2 4800+, 2.4 GHz) has
>>> two cores
>>> each processing an average of 1.5 calculations/cycle x
>>> 64 bits.
>>> This gives a bit rate of 460 x 10^9.
>>>
>>> There is some question about how much data a synapse
>>> firing
>>> equates to, but assume 1 bit/synapse firing for now. Thus
>>> it would take 217,000 of these CPU chips to equate to
>>> a human
>>> brain.
>>>
>>> The most powerful computer in the world (Blue Gene-L)
>>> has
>>> 40% fewer CPU chips, and they are each 39% as powerful
>>> as the
>>> Athlon above, for a total of about 1/4 of the required
>>> power.
>>>
>>> So the failure of AI to date can be explained by the
>>> lack of
>>> adequate hardware.
>>>
>>> Special purpose AI accelerator chips, similar to
>>> graphics
>>> accelerator ships, may buy you a factor of
>>> 10 improvement. Clever programming may buy you
>>> another
>>> factor of 10, and the expected improvement in
>>> computers
>>> in the next 5 years will get you another factor of 5.
>>>
>>> This would bring the number of blade servers required
>>> down
>>> to ~430, which is a reasonably small number. So an
>>> investment
>>> in accelerator chip design and AI programming, coupled
>>> with
>>> the expected improvement in computers overall, could
>>> yield
>>> true AI in 5 years.
>>>
>>> DRN (I've previously signed my messages 'Daniel', but
>>> Daniel Radetsky signs his messages the same way. To avoid confusion
>>> I'm now using the initials of my
>>> historical
>>> re-enactment persona 'Daniel of Raven's Nest', which
>>> is
>>> where my email address comes from. Dani Eder is my
>>> real
>>> world name)
>>>
>>>
>>>
>>>
>>>
>>> __________________________________________________
>>> Do You Yahoo!?
>>> Tired of spam? Yahoo! Mail has the best spam protection around
>>> http://mail.yahoo.com
>>>
>>>
>
>
>



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:55 MDT