Defining Artificial Intelligence (AI)
What is artificial intelligence (AI)? When technical experts talk about it, they use different definitions referring to a host of the functional domains and have a range of opinions about it.
In their seminal 1955 book, Artificial Intelligence: A Modern Approach, authors Stuart Russell and Peter Norvig give it an arresting description,
"...agents that receive percepts from the environment and take actions that affect that environment."
This definition pulls together a lot of important concepts:
- "receiving precepts": the question is, can a machine "perceive?" We know they can sense. However, perception is thought of as sensation influenced by a point of view or a vested interest that changes the sensation--or goes beyond the sensation.
- "take actions that affect that environment": this adds robotics and speech as well as natural language potentials to the definition--also the idea of evaluating outcomes and knowing when its aims are completed.
When we declare that artificial intelligence is here, we find that the definition of artificial intelligence has changed and splintered over the years.
What is and isn't AI?:
Handwriting recognition and voice recognition used to be called AI. However, with the availability of commercial systems to do these things, they no longer fall under the AI umbrella. The process of finding solutions in a massive array of data can still be referred to as AI, even though the training of the system using samples of training data and choosing from a variety of paradigms seems too slavish to call it intelligence. AI always seems to incorporate the unknown and mysterious.
AI is now a continuum:
- Assisted intelligence is where some machine judgment system based on predefined rules replaces many of the repetitive and standardized tasks done by humans.
- Augmented intelligence is where humans and machines feedback and learn from each other.
- Autonomous intelligence is where some adaptive and continuous machine system takes over in some cases. The decision to allow autonomous intelligence to take over certain tasks always depends on the economics of the situation, and the degree of trust people have in the ability of the autonomous system.
The idea of AI floating around in consumer culture is an idea generated in myth and movies. When we interact with an AI system, today, the feeling that this scheme is truly intelligent may strike one initially but doesn't last long. We keep looking for the soul of the machine and find nothing but repetition and half-understood word patterns.
What we can say with certainty is that handy device that will follow verbal commands and do things for us, or turn on other systems on command. These systems may be able to learn how we give it commands when we tediously train them or may be able to learn by repetition the things we ask them to do most often. The rest is advertising and myth. Artificial intelligence is principally a matter of human perception.
The myths surrounding AI may have come out of the symbolic logic demonstrations in the 1960s. In 1963, a computer system called "Logic Theorist" was able to use the rules of symbolic logic to discover proofs to theorems in symbolic logic. A system called "DENDRAL" (now called an "expert system") was able to mechanize aspects of scientific reasoning used in organic chemistry. Another program called "MYCIN" (also now called an "expert system") was able to diagnose infectious diseases interactively.
There is something flawed with the quest for AI. There are a lot of interesting theories and experiments underway, and some undoubtedly will achieve commercial success. However, the question remains, "Where are the intelligent machines like the HAL 9000?" Where are the intelligent machines that could do us harm, machines whose power comes from their having their point of view?