Writing the Future: Multithreaded dialog
Imagine you were asked to program an android, what would you expect to do? You would expect to write endless dialogs of connected sentences, each sentence would be linked to more than one alternative target sentence so that your dialog can be multithreaded, and can be played back in different ways but all in the same context. Then you would create another context and repeat the process until you have a good repertoire of threads. Then you'd expect to test the dialog to see if it makes sense as intelligent conversation. Is this pure fantasy? Not any more, I found a software package that does exactly that, it's called George and Mary (http://www.artificialpsychology.com). George and Mary is artificial psychology software which you can download for free, its intelligence is based on placing sentences in motivational scales and linking them, for example sentences like:
1- My opinions may have changed but not the fact that I am right.
2- Spend not need not.
3- The government should protect our culture from foreign TV programming.
4- I will volunteer to monitor the elections next season.
The first will be placed in high positive Emotion scale, the second in slightly negative Feeding, third in low negative SocioSex and fourth in positive Parenting. The scales are borrowed from animal motivation; they are fear-related and easy to understand: the Emotion scale embodies the motivation of self-preservation in the immediate sense, Feeding is to protect your resources, SocioSex is the protection of social partners and clan, and Parenting is defending your progeny. There is also the cognitive divide: Externalizing-internalizing separates a motivational state into opposite reactions: aggressive and fearful. For example "We should protect our culture from foreign TV programming" is internalizing because it expresses fear, while "We should subsidize our TV programming in order to spread our values" is externalizing because it expresses aggression.
Linking of sentences is forward-looking: playing the source sentence leads to the target. And if there are several target sentences then one is selected by chance. If there are many linked sentences then filters can be used in order to narrow down the choices. The linking and editing of sentences is a breeze with the linking editors, which come with drag-and-drop convenience and status display for each target sentence with mouse scroll.
The personalities of George and Mary are dynamic; they change with time so that if George was prompted say 5 seconds after Mary's sentence then the response can be slightly different from a response after 10 seconds. And the natural trend is for states to gradually decay to a steady state value, normally at or slightly offset from zero. Because of the dynamics, there are two parameters to figure out before starting to program the android: the time constant, or speed of decay, and the trigger value, or the state increment each time a sentence is played.
The first step is to write a sequence of sentences with pauses between them. You select the start and link it to a target in a single scale with zero time constant and trigger values. Then you can expand into more scales, and different parameters. When you are done you can sit back and enjoy the chattering of George and Mary, or try to improve the dialog by editing the links and sentences.
Source: http://www.ArticlePros.com/author.php?Khaled Naser
|
