source: http://www.linternaute.com/cinema/film/dossier/robots-au-cinema/6.shtml
I am reading a book The Cambridge Handbook of Artificial Intelligence (Keith Frankish and William M. Ramsey, 2014) and I would like to shortly write about some interesting ideas I came across in this text.
Artificial Intelligence (AI) is a relatively new field of study and research. It's aim is to create intelligent systems and programs. That means systems capable of rational thinking and reasoning. This may sound straight-forward, but it's a research where more than one discipline have to combined. AI research includes and combine subjects as neuroscience, psychology, philosophy, computer science, engineering, mathematics etc. There are many ways to look at AI, many potential uses and questions that have to be answered.
There are two basic directions to go in terms of creating AI. The symbolic AI is based on theory that intelligent system can be created within the capabilities of today's computer and programming languages, within the rules of today's computations. It's just necessary to find the way of how to create it. The other trend is based on imitation of human brain. Instead of computer units, we have to try create/imitate neural nets, with a basic units of neurons. This trend is referred to as connectionism.
Either if it's connectionism or symbolic AI there are abilities the system has to have. These abilities are the fundamental questions of AI. One of the most important capability is reasoning. An intelligent system needs to be able to reason what he is doing. It needs to know and have reasons for its decision to be made and the output. The basic idea of any program is that it reacts to certain inputs (stimulus) and creates an output. The imitation of neuron can be represented by the McCulloch-Pitts unit, where we have m-inputs x each with a weight of w and their sum creates the output of this unit (neuron).
The phi function is a transfer function (unfortunately I don't have enough knowledge to justify and comment on these equations, so I will just cite what the description of the transfer function is).
"In engineering, a transfer function (also known as the system function[1] or network function and, when plotted as a graph, transfer curve) is a mathematical representation, in terms of spatial or temporal frequency, of the relation between the input and output of a linear time-invariant system with zero initial conditions and zero-point equilibrium.[2] With optical imaging devices, for example, it is the Fourier transform of the point spread function (hence a function of spatial frequency) i.e. the intensity distribution caused by a point object in the field of view."
So now, we would have a machine that receives inputs, but has to create the correct reasonable outputs. That means that we have to create a machine that is intelligent and based on some knowledge rationally decides how to act. But how do we represent this knowledge? If we have a e.g. medicine robot that diagnoses a patient, he has to a great knowledge to produce a decision that would doctor deduce logically. Having algorithms for all the procedures a medicine robot would have to proceed before it reaches a decision would be inefficient, therefore we have create a logically based program.
In terms of knowledge there is another crucial question. Programs can read on the internet what does word "good" mean in thousands of articles. But after all, machine cannot think of the word good abstractly. It cannot know what it actually means and represents. By creating a machine with the human level of intelligence, that could think abstractly, logically and rationally think we would make what is called strong AI.
There is a way how to represent knowledge, so far that we know of. We can make list of concepts (e.g. good, beautiful, charming, evil) and link these concepts. Therefore a machine receives inputs and can logically decide based on the connections and links each concept has. Mathematical operations can be used to produce a logical output. There is fuzzy logic, where truth and false isn't strictly based on 0 and 1, but a value can obtain truth value in range of 0 to 1 (e.g. 0,75 true). Other operation is propositional calculus, which can transfer human language to mathematical logic propositions.
"In mathematical logic, a propositional calculus or logic (also called sentential calculus or sentential logic) is a formal system in which formulas of a formal language may be interpreted to represent propositions. A system of inference rules andaxioms allows certain formulas to be derived. These derived formulas are called theorems and may be interpreted to be true propositions. Such a constructed sequence of formulas is known as a derivation or proof and the last formula of the sequence is the theorem. The derivation may be interpreted as proof of the proposition represented by the theorem."
One of the most important parts of AI is machine learning. In history it was thought that programs could always work only the way they are programmed, therefore the can't generate something new and creative. But machine learning changed this statement. In 1959 Arthur Samuel created a checker playing program that was learning. After a while of playing he couldn't beat it.
My idea of machine learning is this. I would think of it as if of children. When they are born, they have few instincts (algorithms) they are driven by, and through out childhood they are learning thanks to basic human senses: sight, hearing, touching. If we could make a machine learn this way, just by having parts ("organs") for receiving input and in some way recognizing the meaning of it and learning from it, we would have a system that is self-learning and it would a complete imitation of a human.
I find it absolutely incredible just to trying to think of a way how could machines think. How could they actually know what "good" means. How could they learn. This article was based on the book mentioned in the beginning. It is a book containing many essays and texts from top universities professors on AI and looks at every aspect of AI.
Lukas Cerny, 7. 6. 2014
Žádné komentáře:
Okomentovat