FRIHOST • FORUMS • SEARCH • FAQ • TOS • BLOGS • COMPETITIONS
You are invited to Log in or Register a free Frihost Account!


My prediction: Reductionist AI by 2039.






What do you think of my prediction?
Too Optimistic
66%
 66%  [ 2 ]
Too Pessimistic
0%
 0%  [ 0 ]
Very Reasonable
33%
 33%  [ 1 ]
Total Votes : 3

ocalhoun
Based on a very reasonable model of how the brain works at the cellular level, the number of neurons in a human brain (100 billion), and the fact that memory capacity and CPU throughput double every 18 months, I predict that we will have reductionist AI (on high-end desktop workstations; supercomputers may accomplish it sooner) by 2039.
(Reductionist AI makes a model of the brain, and predicts how it would react to given inputs, from the bottom up. An emulation of a real brain, if you will. Unlike holistic AI, which would emulate the actions of the brain, but use completely different structures (algorithmic code, rather than simulated neurons) to do so.)

Memory Requirements:
100,000,000,000 neurons
2 bits to measure pulse strength
320 bits (5 bytes to describe each neuron connection, 8 connections per neuron (40 bytes))
800 bits to describe the connotation of each neuron (for example, one type of visual cortex neuron might represent the color yellow)

51200 Tera bits of memory (6,400 Tera bytes)

CPU Requirements:
100,000,000,000 neurons
7 Floating-point operations per nerve pulse
700,000,000,000 Floating-point operations per thought

700Gfl for the appallingly slow rate of 1 thought per second. (1.4 Tera hertz)

Today's High-end workstations:
4GB memory,4Gfl CPU (2.0Ghz * 4 cores)
Double Every 18 months:
8,8
16,16
32,32
64,64
128,128
256,256
512,512
1024,1024
2 TB memory,2 Tera flops CPU
4,4
8,8
16,16
32,32
64,64
128,128
256,256
512,512
1024,1024 <-- At this point the CPU is powerful enough for one thought per second. (In the year 2034)
2 Peta bytes memory, 2 Peta flops CPU
4,4
8,8 <--Powerful enough to emulate a human brain. (With 11.42 thoughts per second) (In the year 2039)


378 months (18 months * 21 iterations) = 31.5 years.
Gagnar The Unruly
Maybe you accounted for this already, but...

Bear in mind that the human brain works on a 3-dimensional network, and that multiple neurotransmitters are used that have dynamic interactions and concentration-specific functions. The synapses in a brain are far more complex than simple binary transistors. Also, though signals travelling between the synapses are propogated by a change in electrical potential, quantitative differences in the strength of the signal affect the way the signal is recieved and processed, and that in a synapse-specific manner.

A typical 2-d binary processor, even an incredibly powerful one, would have to perform an enormous number of extraneous computations to emulate the neural network of a living organism. It would be like building a nuclear reactor with stone-age technologies.
ocalhoun
^The working parts are mostly on a (relitively) flat (2mm thick) surface, though. The rest exists to forward impulses from one part of that to another, something that can be done in a model without all the machinery in the middle to accomplish it, by using relationships like those in a relational database. The model of how it works that I'm working with is a system of geometric feedback patterns using a Darwinian type system to eliminate extraneous impulses and jitter. The geometric feedback patterns are especially useful, because they limit the amount of important connections per neuron. (one for input, one for output, and 6 for feedback)
Gagnar The Unruly
What does it take to make a modern set of transitors emulate the activity of a single synapse?
ocalhoun
^I would have the synapses be virtual, not represented with dedicated trasistors.
EanofAthenasPrime
very nice, but how do u know for certain it will be conscious?
ocalhoun
^Even if not truly conscious, it would be able to imitate human reactions to given stimulus to such a degree to be indistinguishable from a human (supposing that 11 thoughts per second is fast enough to do that). Even if you don't define that as real consciousness, it would still be enormously useful in many situations.
EanofAthenasPrime
ocalhoun wrote:
^Even if not truly conscious, it would be able to imitate human reactions to given stimulus to such a degree to be indistinguishable from a human (supposing that 11 thoughts per second is fast enough to do that). Even if you don't define that as real consciousness, it would still be enormously useful in many situations.


na it would only be useful if it was better than human, if it weren't, it would just be novelty
Gagnar The Unruly
That's one of those things that I think would be really cool and that I hope happens in my lifetime. The big one is discovering life on other planets, but if we could create a digital 'model' of the human brain I think that it would be awesome. And I also think it's really interesting to think about whether it's conscious or not. I feel that it would be.
breebree
ocalhoun wrote:
51200 Tera bits of memory (6,400 Tera bytes)

Is'nt long term memory capacity potentially unlimited
EanofAthenasPrime
breebree wrote:
ocalhoun wrote:
51200 Tera bits of memory (6,400 Tera bytes)

Is'nt long term memory capacity potentially unlimited


yep our minds aren't physical there are telepathically connected to the ever expanding universe... Rolling Eyes
breebree
EanofAthenasPrime wrote:
breebree wrote:
ocalhoun wrote:
51200 Tera bits of memory (6,400 Tera bytes)

Is'nt long term memory capacity potentially unlimited


yep our minds aren't physical there are telepathically connected to the ever expanding universe... Rolling Eyes
I dent realy see what you're getting at.
powers1983
I don't think the memory calculation was to remember things but just to store the model of the brain.

The problem with predicting computing power in the future is that the current technology is getting to the physical limits and so to continue increasing power, designers will have to use a different type of computer (quantum, neural etc.) and the growth rate of those is likely to be very different.

Also, as someone mentioned, the human brain is not digital, it is analogue. It does not rely on 0's and 1's but an infinte amount of variations.

i think your estimate would be reasonable for a working model but I don't think such a model would have any of the higher functions of intelligence (love, hate, greed etc.).

In that sense I think that holistic AI will achieve true intelligence faster as the human brain has to deal with a lot of things that an AI won't (controling the physiology etc.)

David
EanofAthenasPrime
breebree wrote:
EanofAthenasPrime wrote:
breebree wrote:
ocalhoun wrote:
51200 Tera bits of memory (6,400 Tera bytes)

Is'nt long term memory capacity potentially unlimited


yep our minds aren't physical there are telepathically connected to the ever expanding universe... Rolling Eyes
I dent realy see what you're getting at.


was being sarcastic lol
EanofAthenasPrime
Um...any AI can have hatred and greed

that is irrelevant

However I think what you sub-consiously meant to say was that

There is no technology that enables us to determine the level of consciousness

that the Machine has
Related topics
unbeliviable
Graphistes - J'ai besoin de vous!
Algum Brazuca ai?
DotA - Alguem ai joga?
un site sur des trucs que j'ai à vendre
Help - SVG to AI or CDR Converter
Help - SVG to AI or CDR Converter
Deism vs Atheism
não consigo que aprovem meu lance
Are you reductionist or holistic?
Codeur php/sql/js? j'ai besoin d'un coup de pouce
AI?
J'ai fais un nouveau site avec mes petits doigts maigrelets
Will human resistance prevail over computers :)
Reply to topic    Frihost Forum Index -> Science -> General Science

FRIHOST HOME | FAQ | TOS | ABOUT US | CONTACT US | SITE MAP
© 2005-2011 Frihost, forums powered by phpBB.