[TriEmbed] Intellectual and technical debt and machine learning cautionary article

John Vaughters jvaughters04 at yahoo.com
Mon Jul 29 10:08:10 CDT 2019

 What we (Society) call Machine Learning, I call weighted big data with Artificial Stupidity. Yes the machine is learning, NO it is not intelligent. Quite the opposite. In fact, it reminds me of the old Sesame Street game, One of these does not belong with the other. Smart comparisons based on big data inputs with massive processing. The current Machine Learning will be useful as a tool, similar to a ratchet wrench is to a car. It will help us for specific tasks, but not all tasks. Ever try to use a ratchet wrench as a hammer, right, well it does work, but can give very bad results. That is experience talking :) 
I've seen at least two articles from people in the industry stating we are headed in the wrong direction on AI. One software and one hardware. The software person used the same term I always use, Artificial Stupidity. He felt that we had to re-think the entire approach, but did not offer one. The point being he just flat out believed smart weighted comparisons are not the answer. A tool yes, but it will not lead to intelligence. The Hardware guy was somehow connected to Intel and believed the heavy processing was not the answer and instead of high electrical power with high processing power is not the solution. They were looking at low power processing with fast small calculations in massive parallel. Think video card cores. These articles were pie in the sky thoughts, so no idea if they went anywhere. All this tells me is what I have been saying for a long time, we have no clue what intelligence is or how to create it. What we keep doing is taking shots in the dark and extracting a little light to take new aim with another shot. Each shot provides great amazing tools. Object Oriented Programming came from one of those shots. I don't know about you, but that was a pretty amazing concept that lead to incredible advances in usable software. More tools are coming that will blow our minds, but it still will not be intelligent. 
Machine Learning is so complex and very unreliable, because when it fails, it can be quite spectacular. The worse part is the creators have no idea why it failed, because they cannot evaluate the neural network. This is a real Frankenstein. Enough knowledge to build it but not able to understand or control it. The phase we are in right now is to build software to help evaluate what the neural nets are doing and it is a massive task. 
I have always criticized Elon Musk for being afraid of AI, but I have backed down on him a bit. Because, if we allow some of this technology to run our world, the fear is not that it will take over, but it will fail and fail big. I have no idea why Elon Musk is afraid of AI, but I do now see a very real issue where people think their software is great and apply it in situations that can cause massive problems. For instance, imagine AI implemented in an electric grid. SCARY! Ummmm Weapons decisions. YIKES! Sadly, I have found out both of these are being looked at, hence my fear level has raised, but not due to Sky Net domination, Due to Human Stupidity allowing Artificial Stupidity to be misused. 
2 cents worth a half pence on a good day
John Vaughters
    On Monday, July 29, 2019, 10:08:42 AM EDT, Brian via TriEmbed <triembed at triembed.org> wrote:  
 On 7/27/19 5:07 PM, Mark Sidell via TriEmbed wrote:
> Favorite pick-up line: You look like a thing and I love you.

Best.  Pick-up.  Line.  EVAR.

I may have to try this one.


Triangle, NC Embedded Computing mailing list

To post message: TriEmbed at triembed.org
List info: http://mail.triembed.org/mailman/listinfo/triembed_triembed.org
TriEmbed web site: http://TriEmbed.org
To unsubscribe, click link and send a blank message: mailto:unsubscribe-TriEmbed at bitser.net?subject=unsubscribe

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.triembed.org/pipermail/triembed_triembed.org/attachments/20190729/9b4792cb/attachment.html>

More information about the TriEmbed mailing list