Thoughts on artificial intelligence

I thought about artificial intelligence and came up with some basic concepts. I think all these may have been discovered by many others, but I think I may have a slight different perspective due to my experience with computer architecture, software and biochemistry. This may not be new, but I may have an unique way to explaining it.

There are 4 levels of intelligence, instincts, reactive, simulate and imagination.

Instincts

This is hardly called a form of intelligence, but let’s put religion aside, there has to be instincts to allow the advance of intelligence. The examples of instincts are suckle as an infant, pull ones hand off a hot stove, parents love to their children, curiosity, the will to survive and many more.

Those who believe in religion can view these as God's creation. Those who do not believe in God can view these as part of nature, which is yet to be fully understood by human. But the simple fact is that we are born with these instincts. We do not make conscious decisions on these instincts and cannot lose these as long as they are physically part of our body.

In order to build AI into machines, a lot of these instincts has to be designed and created by human. This is no easy task. It involves a lot of technology in sensors, electrical and mechanical engineering, software designs, etc. Many of these has been solved individually. BigDog, which can stand and walk on its own, is a good example of how a lot of these problems are solved. It was built by Boston Dynamics (www.bdi.com).

Reactive

Starting from reactive, we get into what commonly viewed as AI. This is the learning ability. In human this is part of us that help us learn to walk and talk.

The concept of this type of learning is simple. We have memories and we try different ways to do things, if it works, we remember how to do it again, if it hurts us, we remember to not do it again. This is basic trial and error type of learning.

I have seen some projects that already achieved certain level of this learning skill. For example, the 2005 DARPA challenge winner Stanley has the ability to learn from human and itself. However, today's technology has not provided an economic way of mass-producing this ability into common applications. The two biggest obstacles are memory and time.

The basic elements of this learning are: a specific goal and a set of criteria’s, a set of options or capabilities to tryout, and a feedback mechanism to tell if the goal is reached, or if it is closer to the goal, does it hurt or feel good. The information path back is our instincts, e.g. pain means bad and not successful. Of course, there are more complicated learning when we had to overcome short term negative feedbacks to get to a higher goal. That is the part about human and it is related to simulation.

Simulation

This can be confusing, especially in my field. This is not logic simulation in IC design. There are similarities and that is why this term is used.

The most important element of simulation in our brain is abstraction. What every simulation we run in our brain is not reality. Whether it is a simple task of picking up a cup and get water or design an airplane, we run ideas through our brain. The ideas are based on representation of the reality in our brain. A cup in our brain is not a real cup. It is some electrical signature saved in our brain to represent a cup. And water is not real water in our brain, either. However, using the abstract representations, we could plan out how do we pick up the cup, put it under a water fountain and get water into the cup. After we simulate the actions in our brain, we have more confidence in our action and does not run into as many errors.

The second element is knowledge. This kind of knowledge is different from the memory in the trial and error type of learning. This is why I used term knowledge instead of memory. The memory for TAE is very specific. It directly maps to the criteria it uses to determine if the action is 'good' or 'bad'. The knowledge for simulation is more like a formula or method. We give the method some data and or assumption as input and the go through the simulation in our brain. The brain can then use TAE to find out which input yield good or best results.

Simulation saves us from risking ourselves and at the same time allows us to move forward.

Connection

Before I talk about creativity, we should also understand that these three types of intelligence are not completely separate. In fact, they help provide input and take output to and from each other. For example, the knowledge simulation is based on can be from simple trial and error. Imagination provides new targets for us to use simulation.

Another random thought is that to achieve the same level of human intelligence, we can use very simple memory logic blocks with dynamic wiring between them, instead of more complex CPUs and computers. The key is to be able to have a huge number of these basic units and enough space and material for the wiring.

With this, the advantage will be on the computers, because one can add a few complex processors to a self-learning and creative computer. Even an old 286 processor can do math calculations faster than human, so the overhead of adding processors is not hard. The hardest would be to find the correct interface to the huge network of the basic blocks and finding out how does the intelligent computer discover the complex processors and learn to use them. Once we solve that, we can also think about adding processors in human brains.

I don't think the above are new ideas. There are a lot SiFi stuff talking about these. But I think I actually can see it happening.

I think the basic blocks can be based on today's FPGAs.

Connection 2

One part of data collection to accumulate knowledge for simulation is abstraction. Abstraction is talked about in Object Oriented Programming a lot, because that's the foundation of classes and many related concepts. However, human creates software abstraction. In order to think like human, it means a machine has to be able to do abstraction.

My current view, which could be a lot simpler than what it really requires, is that abstraction is really just data analysis. I got this idea after I thought about how people deal with large amount of data. We first need to collect data. But almost at the same time, we start to look for similarities and difference between the data we have. How did a baby learn the concept of "people"? First we keep point to different people and tell the baby they are people. Inside of the baby, it starts to compare different objects we call "people." Eventually, they would notice that people all stand on legs, with 2 arms, a face with 2 eyes, a nose and a mouth. The abstract idea of people is gradually solidified after many positive enforcements and corrections. This type of abstraction or rather definition of concept is never ending in our mind.

The above also means that the abstraction depends on basic trial and error learning, because to get positive or negative feedback on an abstraction, we need to express our understanding (try) and then see what response we get.

What is intelligence?

After watching Nova's short series on the Sting theory and the multiple dimensions, I start to think that maybe intelligent is something on a different dimension.

The intelligent I talk about here is not the ability to calculate, but the creativity part or maybe people can call the soul. If it is coming from a different dimension, it maybe pointless to pursue true AI, unless we can communicate with the other dimension. Of cause, just having a soul is not going to solve problems, so the pursue for better computers is still needed.

The way I see this is that if physics laws using energy and matter can explain how a human being is created, where does intelligent come from. Can energy be converted into intelligence? If yes, then intelligence should cause an energy lost as part of human being development. If no, then intelligent cannot come from our 'world', because everything has to be from earth. Or if intelligent does not take anything to create, but that's hard to believe that something so important to define us as human is so cheap.

I am not able to use mathematics to prove intelligence actually can transform into energy or other forms of this physical world, but I think it is a possibility that something like the string can be used to prove it.

One thing in the string theory that I cannot understand is how they theorized the gravitons could move between membranes of dimensions. If that is the case, we should detect a lost in energy when that happens. They may have other explanations that are hard to explain in a TV series, but that's one of the odd things I cannot understand.

But to me this multiple dimension thing is a very possible answer to what soul is and how intelligence may come about. The TV show also clearly stated some physicist's skepticisms on the string theory, calling it a philosophy rather than physics, because the fundamentals of theory cannot be tested in a lab or observed by our instruments. But I am fine with it being a philosophy at this time. Eventual prove by experiment would be better, but a good or tight mathematical explanation is ok, as long as the logic is tight. God itself is a theory that no one can prove or disprove, because it is outside our four dimensional world. If string theory can point out the possibility of such a unified theory, maybe we are close to actually seeing God. But I maintain skeptical about the whole heaven and hell things that Bible or any formal religion teaches, because those are human interpretation of what God's worlds like, not God's own words. Same thing, the possibility of time travel, parallel universes and beginning of time and space is all human interpretation. That is why I am still interested to see experiment proving these theories.

Another possibility is that we got the whole thing wrong. We may interpret the world the wrong way. Yes, we can use Newton's law to predict motion, but the whole behavior can be explained in a totally different way, as precise and simple.

Creativity

Now let's talk about creativity. Before I watched the String theory TV show, I view creativity as some kind of random generation in the brain, maybe random wiring of different things. A lot of them may not be real or possible, but the few that came out to be true propelled us to the greatness. After the Nova program, I think it is quite possible creativity is not so random. One problem with my original thinking is that if everything were random, then there would be far more useless imaginations than useful ones. However, some thing keep steer us to the 'right' direction. This is the same problem with evolution, where the possibility of mutation to turn out horribly wrong is much higher than creating a new stable and elegant being. Personally I don't believe in the intelligent design theory. I think it is not science at all. I have not heard a single good argument for it, e.g. how does intelligent design explain the Newton's laws. If they really want to study, maybe they should consider using the string theory to show the possibility of a higher being from a different dimension.

My way of thinking is also shaped by the main concept behind Matrix. This is the concept that we may not be 'looking' at the reality the way it really is. Our problem is that each person only sees the world from our little window, which are our senses. Our body and mind may not necessary have to be together. I think this carries some thought from Buddhism, which enlightenment will bring us the reality we cannot see.

Move back to my original idea of how creativity would work in a computer like system. Creativity by itself is useless. Because there is so much randomness, most of the ideas would be wrong. Although we do not seem to deliberately rule out most of the bad ideas, it is very clear we use simulation to try out ideas in our minds before we go out do things. Most of the engineering projects require this type of simulation in our brain. We need to “imagine” what need to be done, what are the possible out comes, how do we plan to solve some of the problems, how do all the parts work together, etc. Today we have powerful computers to help us to do very focused simulation, so we can save our brain for more creative thinking. However, we still do a lot of simulation in our brain before any idea is recorded on paper or on disk drives.

Welcome! This is my first post. I do not have time to write a lot now, so I just want to say that I hope to share some of my ideas here. I do not use this as a place as a personal journal.