[FoRK] The AI breakthrough

Gregory Alan Bolcer greg at bolcer.org
Sun May 7 18:14:43 PDT 2017


The biggest advance for AI is in the 1990 paper on the theoretical 
ability for hardware to do parallel computing.  (I'll have to track down 
the source). I learned about it in Kai Hwang's USC class in computer 
architectures.

In 1989, DARPA shut down all Machine learning funding as they determined 
that AI techniques had outpaced the ability for computer hardware to 
benefit from them for the next 50 years.  Every single AI/Machine 
learning person then went into software engineering. This 1990 
breakthrough then trickled up through everything including multi-core 
processors versus single core clock speeds up to modern day GPGPU 
computing as a platform for AI and machine learning.

The second biggest advance was circa 2007/2008 where a small little 
company called Kerosene and a Match (http://kerosenetech.co/) discovered 
that you could use GPGPU computing to do AI combined with image analysis 
and reverse image search acceleration. It was quite controversial and 
even discouraged at the time.

The third biggest advance was 2000/2001 and later 2004/2005 advances 
where someone discovered that you could also use GPUs, aka video cards, 
to accelerate cryptographic cracking which served as the basis for 
mapping complex AI problems onto computing resources in a way that 
scaled beyond anything that had come before it.

Any other advances that come after that pale in comparison, though open 
source Tensorflow and CNTK seem to be the biggest advances in the world 
in 2016.

Greg


On 5/7/2017 4:10 PM, Centroids wrote:
> Is that an assertion, or an observation? Is there a link?
> 
> My take is that the success of machine learning *invalidates* decades of research into algorithmic AI. :-)
> 
> E
> 
> 
> 
> Sent from my iPhone
> 
>> On May 7, 2017, at 12:33, Lucas Gonze <lucas.gonze at gmail.com> wrote:
>>
>> The mainstream-level breakthrough of AI / machine learning in the last 4-5
>> years is a watershed in computer science. It validates decades of research.
>> _______________________________________________
>> FoRK mailing list
>> http://xent.com/mailman/listinfo/fork
> 
> _______________________________________________
> FoRK mailing list
> http://xent.com/mailman/listinfo/fork
> .
> 


More information about the FoRK mailing list