Baars ("In the theater of consciousness") mentions a person playing Tetris. When the person first starts to learn it, his brain is very active. After four weeks, the person plays much better, and his brain shows only a small trace of activity.
What's happening in the mind?
In the learning phase the mind is creating a plan for playing Tetris. A great number of small problems need to be solved and we can imagine that the mind is creating partial plans for all these problems. Also, the partial plans need to be incorporated so that they form a consistent whole that has a response for each input that is provided by the game. Once the plan is complete, there is no more need for problem solving and the process can be performed automatically.
What processes are involved here?
- As soon as we are playing Tetris for the very first time, a separate activity is created in some part of the brain, a tiny agent for playing Tetris. The agent, however, doesn't have a clue yet. For everything it does or doesn't do yet, it needs help. Since it is possible to create two games of Tetris at the same time (the analogy comes from playing Chess, by grandmasters; I mean: it is possible) there is not just one agent process permanently available, processes are created and destroyed each time they are needed.
- The agent takes / is given the plan and executes it.
- The agent creates new eyes and ears. These virtual sensory organs make sure that the visual input is unconsciously filtered and sent directly to the agent, without intervention of the Working Memory. The tetronimoes should be recognized without thought. These organs themselves will be developed later, so that the agent can not just "see" the tetronimoes, it sees strategies, structures, possibilities.
- Whenever a situation occurs that the agent cannot deal with immediately, it sends a request for attention to the Working Memory. WM passes it on to whomever is capable of creating plans.
- Creating new plans is of course the distinguishing feature of intelligence. It is done by analogy. New inputs are matched the the inputs of existing plans, and these plans are then adapted for the new need. A special intelligence module takes care of it. The partial plans need to be made context dependent. A certain plan needs to be activated only if a bunch of conditions are true.
- When the agent receives the partial plan, it is woven into the plan it is building for Tetris. The plan, of course, is available to other Tetris processes as well. It is the global plan that is adjusted. This plan is stored in declarative memory. This memory is not like a library from which you lend books via the librarian (WM), all unconscious processes have a direct link to it.
You can play Tetris and talk to someone at the same time (I wanted to say "drive a car", but that is probably not advisable). Especially when you have been playing Tetris for four weeks and you are talking about the weather. However, if you are a beginning player or talking about politics, your process slows down a lot, because you need to ask WM for attention each time, and other processes need to help you.
The thing is that several processes can be active at the same time and each of these may be building a plan. This means that building the plan, or extending an exsiting plan, is something that is done by the active processes, and not by a separate "plan-building" module. Why not? Because each time a small hick-up is encountered in the existing plan that is currently being executed, the whole plan would have to be transferred to the planner module, along with all the contextual data. And then, when it is done, it would have to be returned to the executing process. This is an unlikely state of affairs. However, given the trouble that people have doing several new things at the same time, it might just be the case. On the other hand, as for neural plausibility, it is very unlikely that large parts of knowledge are transferred in the brain all the time. Therefore it is more convenient to suggest that new and autonomous processes are governed by an active process-agent, that does not just process a plan, but calls for help and processes the given help as well. That is, when you are playing Tetris, your Tetris-persona becomes active and acts autonomously.
- my agent
- 28-06-2009 28-06-2009 22:58 - Context as a state machine
- 23-06-2009 23-06-2009 14:07 - Procurios zoekt nieuwe developer
- 22-06-2009 22-06-2009 09:22 - Ternary trees
- 21-06-2009 21-06-2009 21:01 - Analysis of a context switch
- 20-06-2009 20-06-2009 16:07 - Interactionist dualism
- 19-06-2009 19-06-2009 23:20 - Articles about Procurios' semantic web implementation.
- 04-06-2009 04-06-2009 21:17 - vvvzeeland.nl
- 24-05-2009 24-05-2009 11:08 - Activity automation
- 17-05-2009 17-05-2009 22:05 - Agent architecture
- 16-05-2009 16-05-2009 16:37 - Great week for semantic web
- 16-05-2009 16-05-2009 16:17 - LillyBot
- 10-05-2009 10-05-2009 22:36 - Question space