Volition in Artificial Intelligence
From ResearchID.org
Contents |
Formulating Problems
Problems must be formulated by first deciding on and setting goals and purposes. On the other hand, "Design conceptualization" is the act of mind in the action of designing something for the accomplishment of a purpose that has already been chosen. Therefore, Problem Formulation is not the same as Design Conceptualization.
Artificial Intelligence developers have so far not considered volition to solve "The Frame Problem". The Frame Problem is considered to be already solved by some developers. But, until we have AI systems that can formulate problems for themselves, they will never be considered fully intelligent no matter how deeply they can see into problems like playing chess or navigating through space.
Correct problem solving alone is not needed in order to indicate volition, but, it does show consistent volitional behavior when going in the sequence from:
1. problem formulation,
2. to problem investigation and testing research that generates new information,
3. to research of known related information (current knowledge and technology),
4. and on to problem solving and Design Conceptualization of the same problem from step 1.
Please notice that step 1 is the easiest step for humans, but hardest for us to get a computer to do. But why is it hard for us to get a computer to formulate new problems? Maybe it is because programmers fail to recognize a need for volition.
Ability to Change Rules
As an example of why problem formulation is important, consider the game of chess. Capturing the opponent's King is "the problem" in the game of chess. This problem has already been set as the game's objective by the programmer. Even though the programmer may not have invented the game, he non-the-less must provide that knowledge of the game's objective to the computer. He could also have formulated any other objective that he wished. For instance, capturing the opponent's knights could have been formulated as the game's objective instead of checkmating the king) or changed the size of the playing field to 50x50 or even make it 3-D (8x8x8). He could also have decided to make the computer navigate a ship through space by reading star positions. The point is that all of these objectives and rules of the game are formulated by the volition of humans.
If a computer decides to formulate a new problem that we know that it has not been programmed for, that would be a demonstration of some level of volition. Random purposeless movement (this would appear to be like someone having a convulsion) is easy to generate; however, specific successive movement that leads to the attainment of a specific goal (even in a virtual world environment) would show purposeful (Teleological) volition. One thing about computers is we can directly access and monitor everything that it does that is outside of it's programming. Its ability to formulate and act on new problems all amounts to self-programming. Developers can access and monitor that self-progamming using special hardware and software that can be designed for that purpose.
Gradual Environment Expansion
The volition of an artificial intelligent program like Deep Blue can be tested by gradual environment expansion. How well can the program continue to play if the playing field is gradually expanded? If we were to add another row onto the chess board, a human player can exercise his general purpose volition to take advantage of the new space by formulating improved strategy and tactics. However, a special purpose program is usually written to only stay with the board size that it is programmed for. It may not even have the ability to perceive of the existence of the new expansion space!.
There is a big problem that comes together with the ability to perceive large environments. The problem is in what can easily be a very overwhelmingly large amount of unimportant information and noisy disturbances that must be dealt with. AI developers have often placed very rigid volitional constraints on their "learning" AI systems so as not to have to deal with the far greater resources that will need to be designed into their systems.
Seed AI
"Seed AI is the concept of building an AI system with a limited, but carefully chosen, set of capabilities that allow the system to dramatically increase its knowledge and skills through self-directed learning and adaptation. One concentrates on carefully designing the seed of intelligence, and then nurses it to maturity - bootstrapping intelligence. In my AGI design this has two distinct forms/ phases:
1. Coding the basic skills that allow to system to acquire a large amount of specific knowledge
2. The system reaching sufficient intelligence, and conceptual understanding of its own design, that it can deliberately improve its own design."
(For more see: http://www.adaptiveai.com/faq/index.htm#seedAI)
Love Required for Volition?
In the movie "Artificial Intelligence" it was "love" that was considered the ingredient necessary for AI systems to be able to take on new tasks passionately without first being assigned to the new tasks by humans. If AI researchers can one day create artificial love in their AI systems, they may then find that "volition to formulate new problems" is also present in that same system. If research can show that love is required to formulate new problems, this will add love to the required internal dynamics of a designer. But, regardless of whether or not love is involved, it is clear that some force "X" is very much required for the true intelligence that AI systems need to formulate new problems. And whatever we call that "X-force", if it allows for formulating new problems and also allows for the AI system to assign itself to the new task of solving the newly formulated problem, then X-force will still match any definition of volition.
A.I. Artificial Intelligence (2001)
See: The Frame Problem
Back to Defining Intelligence: Volition
