General Artificial Intelligence is a term used to describe the kind of artificial intelligence we are expecting to be human like in intelligence. We cannot even come up with a perfect definition for intelligence, yet we are already on our way to build several of them. The question is whether the artificial intelligence we build will work for us or we work for it.
If we have to understand the concerns, first we will have to understand intelligence and then anticipate where we are in the process. Intelligence could be said as the necessary process to formulate information based on available information. That is the basic. If you can formulate a new information based on existing information, then you are intelligent.
Since this is much scientific than spiritual, let’s speak in terms of science. I will try not to put a lot of scientific terminology so that a common man or woman could understand the content easily. There is a term involved in building artificial intelligence. It is called the Turing Test. A Turing test is to test an artificial intelligence to see if we could recognize it as a computer or we couldn’t see any difference between that and a human intelligence. The evaluation of the test is that if you communicate to an artificial intelligence and along the process you forget to remember that it is actually a computing system and not a person, then the system passes the test. That is, the system is truly artificially intelligent. We have several systems today that can pass this test within a short while. They are not perfectly artificially intelligent because we get to remember that it is a computing system along the process somewhere else.
An example of Fusionex artificial intelligence would be the Jarvis in all Iron Man movies and the Avengers movies. It is a system that understands human communications, predicts human natures and even gets frustrated in points. That is what the computing community or the coding community calls a General Artificial Intelligence.
To put it up in regular terms, you could communicate to that system like you do with a person and the system would interact with you like a person. The problem is people have limited knowledge or memory. Sometimes we cannot remember some names. We know that we know the name of the other guy, but we just cannot get it on time. We will remember it somehow, but later at some other instance. This is not called parallel computing in the coding world, but it is something similar to that. Our brain function is not fully understood but our neuron functions are mostly understood. This is equivalent to say that we don’t understand computers but we understand transistors; because transistors are the building blocks of all computer memory and function.
When a human can parallel process information, we call it memory. While talking about something, we remember something else. We say “by the way, I forgot to tell you” and then we continue on a different subject. Now imagine the power of computing system. They never forget something at all. This is the most important part. As much as their processing capacity grows, the better their information processing would be. We are not like that. It seems that the human brain has a limited capacity for processing; in average.
The rest of the brain is information storage. Some people have traded off the skills to be the other way around. You might have met people that are very bad with remembering something but are very good at doing math just with their head. These people have actually allocated parts of their brain that is regularly allocated for memory into processing. This enables them to process better, but they lose the memory part.
Human brain has an average size and therefore there is a limited amount of neurons. It is estimated that there are around 100 billion neurons in an average human brain. That is at minimum 100 billion connections. I will get to maximum number of connections at a later point on this article. So, if we wanted to have approximately 100 billion connections with transistors, we will need something like 33.333 billion transistors. That is because each transistor can contribute to 3 connections.
Coming back to the point; we have achieved that level of computing in about 2012. IBM had accomplished simulating 10 billion neurons to represent 100 trillion synapses. You have to understand that a computer synapse is not a biological neural synapse. We cannot compare one transistor to one neuron because neurons are much more complicated than transistors. To represent one neuron we will need several transistors. In fact, IBM had built a supercomputer with 1 million neurons to represent 256 million synapses. To do this, they had 530 billion transistors in 4096 neurosynaptic cores according to research.ibm.com/cognitive-computing/neurosynaptic-chips.shtml.
Now you can understand how complicated the actual human neuron should be. The problem is we haven’t been able to build an artificial neuron at a hardware level. We have built transistors and then have incorporated software to manage them. Neither a transistor nor an artificial neuron could manage itself; but an actual neuron can. So the computing capacity of a biological brain starts at the neuron level but the artificial intelligence starts at much higher levels after at least several thousand basic units or transistors.