Monday, November 24, 2008

Is it a good thing for science to try to re-create the brain's power?

This blog post was inspired by something that was shared with me by TimesToCome.

IBM gets DARPA cognitive computing contract (The blog subject came from this article)

The world is trying to recreate the brain. Is this a good thing? I think it is a good thing. I don't know where it will end up. I may, in the end, feel like working on this kind of research was a mistake. If things end up in a Terminator kind of world, that would be bad.

I would like to understand why human brains take so little power and yet can dwarf my laptop in terms of flexible processing capabilities when my laptop heats the entire room when taxed. I think there are other models for processing that will yield very efficient results. Many of the biggest advances in industry are the result of someone taking the processes or procedures from one industry and adapting them to another. Can we take some of the manufacturing floor processes and procedures and port them over to help us understand and manage our email flow? Seems like a good idea to me.

If humans can create a brain that is simple and yet able to learn and efficiently carry out some tasks, that should give us more insight into the workings of our brains. More insight will lead to better understanding and improved ability to repair. I believe it will also give us some new computing models. Massively parallel influence based models. I think people will be surprised by a few things. 1. The brain's inputs and outputs are simpler than we currently expect. 2. The processing model is massively parallel. 3. The brain and input devices (eyes, ears, ...) are not as all seeing and all computing as we think they are. One example that I created during a discussion with someone near and dear, was that of peripheral vision and analyzing objects. People think that they constantly understand and can see everything around them. You know that there is a white van over there and think you know just what it looks like. It even feels sharp in your vision while you are looking somewhere else. Stop. Pay keep your vision firmly focused on one spot. Pay attention to the van. Can you really see it? Are the lines sharp? I think we are excellent parallel processing and filtering machines. We glance around quickly, stop on the van for a millisecond or two. Determine that it is a white van and move on. This object is now tagged. We have not really looked at it for a while. There are tons of details that we could have seen if we spent serious time looking at it. In the background, we are scanning objects that we are tracking. Have they changed? Try having a conversation in a parking lot. Pay attention to where your eyes go. I found that mine keep hopping from object to object making sure that they are still there, have not changed, and trying to decide if there is more to look at.

Evolving the entire human brain would be interesting. I think we would gain valuable insights. These days, I think I would just take small improvements. There are many logical actions, based on the context, that computers can safely take for us. Why aren't they? So much of our lives goes through the computers and it feels like we have to baby them through step by step. There has to be a better way. If computers were not just gophers that were sent to do specific tasks manually, and instead actually took appropriate actions based upon your wishes, I think world productivity would increase.



herself said...

Of course we should try, and at some point I'm sure we'll succeed.

There's so much we could do, replace TSA officials with robots for example. That's bound to be an improvement.

But every time I do even something small in AI I can't help but wonder if we aren't putting the steps in place for a future 'Terminator' or 'Cylon'.

Jacob Taylor said...

When my AI project's first evolved predators from a random start, in the second generation, completely owned the hand coded prey that I had, it really opened my eyes. Evolution is very powerful but generally very slow. With computers evolution can be very powerful and light speed. We need to be careful with what control we give them. Input and Output are king.

BTW, TSA replacement would be an improvement. Have you read: The Things he Carried


Anonymous said...

Regarding your discussion on the white van, it reminded me of some of the first drawing instruction I received. My task was the draw a simple object. It might have been a picture, or an object, I can't remember, but it was relatively simple. The parameters were few. Spend 5 minutes or less drawing exactly what I saw. Put as much or as little detail as I wanted into it.

The outcome is probably foreseeable since you have young children. It was rough symbols for what things are. The closed book on the table became a rectangle. The apple a circle with a dimple on top and a leaf coming out of it. The face was an oval with a nose planted dead center.

That's L-mode of our brain kicking in trying to logically categorize everything. The scanning part you mentioned. It's not what's really there, but it's what our brain tries to tell us is there to speed up the process. The L-Mode is the easy part to generate in a computer program, from my perspective. It's cold and calculating. Give it these inputs, get these outputs. The R-Mode is more interesting to me though.

In drawing the L-Mode is something you have to learn to overcome in order to truly see and capture that input. In Africa, you didn't really care the length, color, and direction the wind blew the lion's mane. It was a lion, you moved. It served us well, and still does, but the R-Mode seems to be where creativity and intuition come from. Those are far more murkier, more rounded-rectangle, but the source of most inventive and creative endeavors.

Jacob Taylor said...

It is good to hear that my analysis is backed up by art teaching experiments. :)

The thing that I find most interesting is that the brain kindly ignores the lack of details. We think we see it all, thus we are not motivated to be distracted by filling in largely irrelevant details. This is exactly as you describe.