How did Helen Keller learn to communicate? Partly, she learned by guessing. Not understanding how to equate sign language to objects she guessed at what things made what signs. Always guessing wrong. Eventually she guesses correctly and makes a connection. water. Because she received a response that differed with the correct guess. But she had to get to stage where responses themselves could be meaningfully understood. (arguably many human responses to stimulation are built-in through the DNA regulated developmental process)
When you did the exercise of randomly naming things, you were doing a sort of inverted guessing. Instead of guessing at the names of things, you just gave them random names, as if you were guessing (although, there was likely a pattern to the naming which you may have began to see). Guessing games, like 20 questions are great examples of how structure, narrative, and categorization work with representation and awareness.
Computers can't play 20 questions because they have no representations to start with. Any computer that starts playing twenty questions starts with a question decided by a person. A conscious AI will make up it's own questions to start with.
Helen Keller had lots of inputs, and had already made lots of representations about how to act and what to do in certain situations to get what she wanted.
But to move past that level, she had to have those representational ideas broken so that new ones could be formed. And at first, all the new representational models she had were guesses. Until she made a connection and saw a repeatable pattern, all her actions at signing would be guesses. Until she connected a tactile experience with a sign action, she could only make random signs. Once a sign could be repeated to refer to the same thing, then she could stop guessing.
She connected the sign action for water to the tactile experience of water.
water is 'water sign'.
AW:water = water = water;(sign for water) = (sign for water) = AW:(sign for water)
Beginning with an aware AI, is like beginning with a very simplistic Helen Keller. The AI has to make guesses and connect some objects of awareness to other objects of awareness. What made Helen Keller's experience all the more frustrating is she could not ask questions until she could sign.
Questions go along with guesses. We test our representations. But one of the most effective ways to test representational guesses is to ask questions. Question making is a representational testing act.
As an example, dogs ask questions, usually regarding permission. When a dog "implores" an owner about some future action, the dog and the owner are engaging in a similar awareness/representation experience. Going for a walk, going for a ride, eating a treat. Dogs ask questions about permissibility.
Helen Keller could not engage in any of the facial expressions people make regarding permissibility. She could not receive feedback about her facial expressions, and hence could not ask questions simply with her face.
She could receive tactile feedback, but it was inconsistent. And negative responses would contain and convey little subtlety. Facial expressions can be very subtle, and a negative response in a facial expression can change to reflect increased negativity or increased acceptability.
Whereas with tactile negative responses, they are pinches, hitting, tugs, etc. But like the social faux pas that elicits an under-table kick, the receiver is often baffled by the assault because there is no association to the faux pas or to the severity of the response.
An aware AI must have the ability to ask questions and work with valuations of actions to create representations from it's guesses(it's arbitrary association making). Like a child, an aware AI should demonstrate instances of over-acting and lack of attention. These should generate negative responses so the AI can construct a workable representation of actions. It is not any different than a child learning about the world and how to act in it.
Subsequent AI's could simply clone this data set, but the initial one has to learn. Cloning such a dataset would nececessarily clone any undesirable learning as well.
Asking questions is the power to circumvent the slowness of the negative/positive response process. By making a representation of some action with words, then checking that representation with another person, the action can be assessed without having to process the whole experience. The representation of the question stands in for the action for the questioner. The other person converts the question into the implied object or action and then renders a response based on that association.
Tthe response may vary. It may be inaccurate, but as a process of testing representations and actions this process is fairly effective. Of course, questions may be misinterpreted.
The classic genie story of 3 wishes is a case where different interpretations can make wishes undesirable. Contracts and much rule-making is about eliminating interpretative confusion. And both are usually processes of asking questions in advance of events. Eg, what happens if...
question making is all about representational constructs.
Questions are really associative tests. is x = x;y? That question is really one of, are you aware of x;y? eg is there a link from
aw:x = x = x;y = aw:(x;y) A negative response is where x != x;y
Because x != x;y there is no link to aw:(x;y) from aw:x Without representational linkages as objects of awareness, those objects that can be experienced simply don't exist. [see note]
When we look at Helen Keller we see a movement in her experience. From a world where there was no communication and complex representations could NOT be made, to one full of ideas and representations. Linkages between objects and representations of objects were made, and awareness grew. An AI must engage in the same process. It must make representations in awareness.
[Note: Let's examine questions of association. When we ask is some x the same as x as y x =? x;y we already recognize that x = x;y. When we say some x != x;y, it is not that x;y does not exist. What we mean is more subtle. We mean that x;y is itself associated to another object "true" and "false" ie. (x;y) ; false is what we mean by x != x;y
we can nest these relationships, because there must be an (x;y) ; false and there is an (x;y) ; true. Thus we ask which one of these relationships are "true" and "false" and this process can recurse infinitely but only because we substitute existential facts for conditional facts. We will see later that recursion does stop when the representation making itself stops. these representations ( (x;y) ; false) ; true do not recurse infinitely in practice, because we do not MAKE these representations infinitely. We constrain representation making and thus constrain the possible representations made.
When Helen Keller is guessing and making up words she is making representations. It is only upon the recognition of water to the sign for water that Helen makes a sign that she can reuse. A sign that is associated to water CONSISTENTLY.]