Thursday 21 February 2008

Artificial intelligence

When I was walking back to home yesterday evening, I thought about Artificial Intelligence and how it is going to develop.

Well, we are coming nearer and nearer in these latest years.

What do I think it is? It will START (the development will be decided year by year) as something that will be able to associate ideas, concepts and connect them, as well as recognize them.

It will begin like something like a just born baby. Able to learn, but without any knowledge.

At the present, this is very possible, as well as the Knowledge, is the one that now we have on Internet, in community and applications like Wikipedia, the biggest Encyclopedia of the world.

EG: An AI enabled software can be able to connect automatically photos from flickr and getting informations about them connecting their tags with wikipedia knowledge in "some kind of smart way" (that could be implemented in millions line alghoritms).

Again, an AI software can recognize a photo comparing it with all the ones it has on the db (just checking average colours on a grid based shape) and connect the input photo with the information taken by the compared ones.

So, imagine now what about 3d models. When 3D scanners will be as advanced as to recognize 3d models "on the streets", they can be able to compare them with some kind of database of 3d model and identify the 3D area around them. At this point, a "robot" will be able to recognize everything around it and connect everything to its database informations, just like a baby see an apple and think "an apple, juicy, sweety apple".

2 comments:

Anonymous said...

Good analogy with a new born baby! Grow up, get some experience, go to kindergarden, go to school, etc - to eventually get this so called intelligence. For people it works, huh? Intelligence = 11 years of school + 5 years at uni [+ perpetual studying for the rest of your life :)]..... But what do we get from this studying? A range - extramegawide, but anyway, just a range - of questions and answers (problems and solutions) along with the understanding of "similarity" that allows you to answer similar questions and solve similar problems. If you don't know the answer, you go for a brut force.

Let's look closer at this idea of a new born bAIby: who are the teachers for a human infant? Parents, other "old" people, then other kids - for a child's intelligence it is important to communicate with other children as well.

So, in a nutshell, I just feel that this environment for AI breeding is much more than allowing the insatiable infant AI to merely get the data. Maybe this environment for AI is as important as its this ability to learn. Maybe even more important, who knows :)

Enrico Foschi said...

Well, I agree. We have to consider anyway that the knowledge base (wikipedia+photo resources+video resources+3d model resources+other ones) will just be the base, something like a powered knowledge of a 1 year old baby. Then, with this base, he should be able to learn.

AI is mainly the ability to learn, to get experienced and to choose a choice from different ones (that choice will be the best one because of different evaluations).

Then, how can an AI software learn? By experience. I still have to understand where and how he can get experience, but one thing is sure, and this makes AI software faster then humans: they can share experience with a cable.

Once an AI has learnt, everybody else has.

Unluckely we don't have a usb cable to connect our brain :(