[ExI] i got friends in low places

[spike at rainier66.com]

…> On Behalf Of Adrian Tymes via extropy-chat
Subject: Re: [ExI] i got friends in low places

 

On Thu, Aug 25, 2022 at 4:54 PM spike jones via extropy-chat <extropy-chat at lists.extropy.org <mailto:extropy-chat at lists.extropy.org> > wrote:

…

>>…Doesn’t that seem paradoxical? 

 

>…No, that seems like getting hung up on the various definitions of "intelligent".  There is not one single definition, and many of them are not simple yes/no things where you can say something (such as a processor running a Kalman filter) is or is not (that version of) intelligent.

 

 

OK no worries, I have an idea.  Again, skip to the last part if you are in a hurry, because it is a cool conclusion methinks.

 

We might be getting tangled up in conflating artificial intelligence with artificial human-level intelligence.  Our reasoning on that is understandable: if we create software that is human-level intelligent, we can set arbitrarily many of them to writing and optimizing software 24/7 and the result is… the singularity.

 

But not necessarily.  If we run with Adrian’s notion that the definition of intelligence matters, then we could create human-level intelligence to do stuff for us without causing the singularity.

 

Reasoning: humans are human level intelligences, but not all humans can write software.  Only a fraction of them can code worth a damn, so… perhaps we can get human level AI with no will, no desire to take over the world, no self-awareness etc.

 

The idea: we just go with our favorite machine learning technique and set it to work.  Mine is the Kalman filter because that one is used a lot in controls engineering and is my own tool of choice.  So I set up a bunch of Kalman filters with a list of perhaps a few thousand observable metrics.  

 

A practical iterative covariance matrix with today’s computing power can only deal with about 50 x 50-ish rows and columns, so… we have the AI choose an output, such as drought in England, floods in Mississippi, ACE in the Atlantic, average SAT score at the local high school for instance, then let the processor grind away, tossing out the irrelevant factors by noting which row/column factors have zero correlation with the output.  Toss those out as they reach the bottom, say 50th place, replace them with a randomly chosen factor from the long list of possibly relevant metrics.  Over time, the algorithm should learn which of the 50 metrics on the list of a few thousand observables are most relevant to predicting the outcome of whatever it is studying.

 

Then… we load all our resulting covariance matrices onto a top-level processor, which would then know how to predict a lotta stuff.  Then we don’t worry about whether the Boss Pross is intelligent (it isn’t) but it still knows a lot and can make good predictions based on measurements we give it.

 

Adrian, whaddya think of that idea?

 

spike

 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20220825/2d2787a6/attachment.htm>