Fragments of a Language of Thought

In designing our logical robot, we can choose whether to represent and process certain information in a serial way or a parallel way. ¹⁸In a parallel representation, a part of the information is represented by which wire the information is on in a computer system or where in the nervous system the information is located in a human or animal. When all the information is in a single memory, it has to contain labels. Our robot child will use a single memory, and therefore its language of thought will be more explicit in representing certain information than a human language of thought has to be.

The simplest way to represent a visual image in a computer is by a pointer to a pixel array, e.g. in lisp as . This can then be the value of a variable or constant and programs can communicate it in this form. This has advantages and disadvantages.

• The information is readily displayable to a human or transmissible to another robot.

• Suitable programs are required if information is to be extracted from the image.

• If relations among parts of the image are to be expressed in l.o.t., a suitable language for this is required. It is also necessary to relate parts of the image to other things such as objects. For example, we need a relation that asserts that a certain part of a certain picture taken from top to bottom represents Tom's left arm. It might be written
  
  

• This representation is unsuitable for mentally generated images, whether they be invented ab initio or modified from previous images. They will not be complete pictures.

These considerations suggest that robot l.o.t. should not represent images primarily by pictures, although pictures might be an auxiliary data type. Instead, curiously enough, the robot child will need something that is closer to what we imagine human image representation to be.