Artificial Intelligence (A.I.)

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

ruffrecords

Well-known member
Joined
Nov 10, 2006
Messages
16,227
Location
Norfolk - UK
Read some interesting articles on AI  in the latest issue of New Electronics. A group from Manchester University have build a 1200 board system where each board contains and ARM processor which simulates 1000 neurons and one million connections. The whole assembly is said to equate to about 1% of the human brain or to be equal to the brain of a small mammal.

Current work of several researches is concentrating on reducing power consumption the ultimate target being the mere 20W consumed by the human brain. Early works has led to a single chip equal to an insects brain which apparently holds out the possibility of autonomous drones.

I guess it won't be long before we see PC based Producer applications that can take a bunch of shitty tracks and polish them into a gold record.

Cheers

Ian
 
Interesting. But neurons are only part of the picture, higher brain function has a lot to do with the workings of glial cells. I read this (very interesting and enlightening) book a while ago and highly recommend it:

https://www.goodreads.com/book/show/7170963-the-other-brain
 
I loved the ending of that movie. You know, when the aliens return to earth and find him by the blue fairy.
 
living sounds said:
Interesting. But neurons are only part of the picture, higher brain function has a lot to do with the workings of glial cells. I read this (very interesting and enlightening) book a while ago and highly recommend it:

https://www.goodreads.com/book/show/7170963-the-other-brain

Yes, you are right. The articles I read also mentioned that neurons are by no means the whole story.

Cheers

ian
 
This (parallel neuron like computing ) has been worked on for years and not cost effective (yet). Will involve a completely different computing architecture, and massive number of connections.
----
Not exactly AI, but a recent irritation for me is when you call some help line and the voice recognition maze you have to negotiate before reaching an actual human takes longer than it should take a computer to decide, but they add delay on purpose and fake clicking sounds so the unwashed think they are actually talking to a human looking stuff up on a computer monitor.  :mad:

JR
 
a recent irritation for me is when you call some help line and the voice recognition maze you have to negotiate before reaching an actual human

You ain't kidding. Having to contact Comcast, an East Coast ISP all the time I got to testing their decision tree.  Upon the last of the initial options, If I ask to speak to a representative it repeats the question. Most people give up and select from the choices and then you get put in a queue. If you ignore the options and repeat your request to speak to a representative, it apologizes but then again, repeats the list of choices of departments. Ignore the question a third time, lo and behold, it connects you to a human. So sneaky and annoying. My friend calls it "shepherding by tolerance".
 
ruffrecords said:
Read some interesting articles on AI  in the latest issue of New Electronics. A group from Manchester University have build a 1200 board system where each board contains and ARM processor which simulates 1000 neurons and one million connections. The whole assembly is said to equate to about 1% of the human brain or to be equal to the brain of a small mammal.
The strongest interest in machine learning (ML) right now is deep learning, which extends neural nets to large numbers of layers and neurons. Inspired by the human brain, but humans actually don't know a whole lot about how are own brains work. Each neuron is a simple decision making node. 
I think the distinction between AI and ML is important - but more philosophical that scientific. At what point does an algorithm that can emulate something become 'intelligent'?
A deep learning algorithm in recent years could beat humans at Chess or Go. But no machine learning algorithms have become as good as humans at navigating rough terrain (i.e. things animals have evolved to be very good at). While something like chess SEEMS like a significant thing to be good at, it isn't actually something animals have evolved to be good at. Who knows - maybe humans are actually terrible (in an absolute sense) at some of these things we consider so challenging.

Current work of several researches is concentrating on reducing power consumption the ultimate target being the mere 20W consumed by the human brain. Early works has led to a single chip equal to an insects brain which apparently holds out the possibility of autonomous drones.
There are many amazing comparisons that show what incredible technology animals are (if you want to consider them technology). Processing power, energy use, memory capacity, plus things like the maneuvering capability of insects.
Another human capability that we don't even consider challenging is facial recognition. Deep learning has been trying to do this as well as a human for years.

We are probably too far to turn back - but once weaponized drones are released the world might get a 'bit' more chaotic. Really one of the greatest imminent dangers civilization faces. I think it is safe to assume governments are putting a lot of R&D into this. And when individuals have attached firearms to remote control drones and posted videos to youtube, one can only imagine what military forces could deploy.  And if they put AI into independent, autonomous drones? What could go wrong.

I guess it won't be long before we see PC based Producer applications that can take a bunch of sh*tty tracks and polish them into a gold record.
I've seen some attempts at using 'AI' to write songs. Again it comes back to emulating versus creating. At what point does a machine learning algorithm become intelligent or legitimately creative? And if a ML algorithm made a song, would anyone really care about listening to it? Similarly, if every bunch of tracks were combined, processed, and polished into the same 'gold' record sound, wouldn't everyone want to hear something different? Could AI learn how to continually evolve or move forward to make the next hit record?
Every generation loves the music they listened to when they were 25 yrs old and it HAD to be different that the music their parents loved. Would 'AI' be able to learn how to evolve like that? 
 
JohnRoberts said:
This (parallel neuron like computing ) has been worked on for years and not cost effective (yet). Will involve a completely different computing architecture, and massive number of connections.
I'm not sure what you are talking about. Some kind of a biological computing infrastructure?
But machine learning (including deep learning) has been deployed in cost effective ways by business for years. It can run on regular computers or cloud. Google released their tensorflow toolkit which anybody can run through python. You can find tutorials online to run deep learning (i.e. keras in python). Computing power is cheap. For very large datasets there are distributing packages like hadoop.
Mostly used for advertising and user experience on FB, google, etc...
But it is also being used in the insurance industry for instance
 
Petri dish, electrodes, rat brain, plane simulator.
https://www.newscientist.com/article/dn6573-brain-cells-in-a-dish-fly-fighter-plane/

Real brain equals best AI.

 
dmp said:
I'm not sure what you are talking about.
me too sometimes...
Some kind of a biological computing infrastructure?
I stopped reading the trade journals last century but you've probably seen the same research PR I did...

The brain is a different animal (literally).. beyond parallel computing but very connection oriented..
But machine learning (including deep learning) has been deployed in cost effective ways by business for years. It can run on regular computers or cloud. Google released their tensorflow toolkit which anybody can run through python. You can find tutorials online to run deep learning (i.e. keras in python). Computing power is cheap. For very large datasets there are distributing packages like hadoop.
Mostly used for advertising and user experience on FB, google, etc...
But it is also being used in the insurance industry for instance
yup.. not what I was talking about... human brain mimics have been a computer boy wet dream for a long time... still not practical AFAIK, but every decade more becomes possible,

JR 

[edit] speaking of AI today I got a robot call with a computer begging me to not hang up....  click. [/edit]
 

Latest posts

Back
Top