6.0 Testing

 
Testing is introduced at this point because of the difficulty of proving that something does what is required of it, especially when that something is as fuzzy as learning and thinking.

[We may have trouble defining intelligence, but we recognize it when we see it.]

[must show reasoning in its decision making and 2nd have purpose in their actions] "How the Mind works" pg 61-62

When the Wright brother invented the first man made machine to fly the test and proof was straight forward. Everybody had a common understanding of what it meant to fly and the observation of the effect was proof enough. Some people might have said that the word 'to fly' should only be used to describe that which birds do and therefore man could never fly. But this is a narrow view and the meaning of 'to fly' is recognized by certain key features which happen to be independent of the material used to build the mechanism.

It is unfortunate that learning and especially thinking are activities for which there are no commonly understood and agreed upon definitions. One can not say that something is or is not thinking with the same certainty that we can say something is flying. Even flying is not certain as determining whether something is heavier than something else. So there is a continuous spectrum of certainty with weight measurement at one extreme and thinking at the other. Learning also suffers from the same problem. It can be interpreted as a continuous spectrum from not learning to effective learning. Does a rock learn? Does a vacuum cleaner learn? Does a thermos learn? A thermos can keep a hot liquid hot and a cold liquid cold? Is that learning? Does a video recorder learn? Does a virus learn? Does the logon software learn your userid and password? Does a database application learn its contents? Do you learn when you read a book? Does a monkey learn how to use a stick as tool? Do you learn how to dance? Do you learn how to apply differential equations? Do we learn how to think?

Faced with this difficulty, how are we to say that adaptron learns and thinks.

One thing we must do is determine where the line is that says a rock and thermos do not learn but you do when you read a book. The only possible way is to observe its behaviour and judge whether it exhibits the same features that we attribute to learning and thinking when it is performed by humans. This assumes that we can agree that humans can learn and think.

How can you test for learning and thinking

Behaviour is evidence. The Turing Test is based 100% on the ability of the intelligent agent to communicate, however I believe that intelligence can exist without the need to communicate. Therefore a test is required that only requires observation of the agent. The criteria for observing intelligence is an exactly specified environment with its laws of cause and effect, a well defined set of events that can be observed and a set of actions it can perform. Plus It must have a purpose or a goal. This goal must be one that is only achievable by performing some form of reasoning about its environment. It is intelligent if it can learn the laws of the environment and achieve its goal using this knowledge.

No one has ever measured thinking, we have felt it in ourselves but in anyone else we can only infer it. We have to infer someone else thinking from what has happened to him or what he does. If he says he is thinking his statement is only one of the possible piece of 'what he does' but how reliable?

 
6.1 The Environment

Define Environment - stimuli, responses, stimuli that are good and which are bad.

Rich environment - avoid bad, pursue good (does it need good - no) (no drive to pursue good)

11th May 1993 - While attending the local SIGART meeting I thought about how one could deduce that an entity was able to learn and / or think just by observing that entity, i.e. via it behaviour.

 - The environment could be classified into a number of types:

  1. linear in time - sound (pitch & volume) or light (colour & brightness) or letters / words typed.
  2. 2 dimensional - sight / projection in time
  3. 2 dimensional physical - motion / obstacles (touch)
  4. 1 dimensional physical - motion

or A/ Passive - environment does not change unless active agent / robot makes change.

   B/ Active - Things in environment which change location etc.

 - Motivations are important to be understood if some meaning is to be obtained from behaviours. There are positive and negative ones often the reverse of each other. Positive: Food (Energy), Interest - mental challenge - sight/sound, Pleasures - biological - aligned with reproduction and energy.

Negative: Avoidance - heat/cold - survival, no air, pressure - body damage, hunger - pain (low energy), Boredom.

 - Observe learning taking place when it avoids painful situations! - survival, repeats pleasurable activities.

 - Observe thinking taking place when it explores the environment - Interest, Does not repeat exploring known environment - or exercises/uses known environment to get somewhere (not to explore it), seeks out pleasure

26th March 1990 - Could use A->Z as real world visions and a -> z as real world symbolic representations of A -> Z which implies that a -> z are words for A -> Z. We would also need symbols for representing actions / outputs.

17th May 1986 - Environment for adaptron, Location when moved in to gives off different letter. Adaptron can do a number of things such as move up, down, left, right. Maybe some walls like a maze so that when it moves in that direction it hurts itself. What is its goal, purpose in life? To explore the environment. (Yes, but what motivates it? 1st April, 1990) To learn all it can about the environment, what it can and can't do. To learn all possible effects of what it can do. To continually find differences - new things to see, to do. In a closed environment like this one it should eventually become bored. If one square or combination of situations produces pleasure then it must also become bored of this situation too. Next level of environmental complexity is one which changes. e.g. squares move to different locations - input changes when robot does nothing. This can happen when free agent is also roaming in environment and collision encounters can occur.

Undated - Model its environment so as to use the model instead of the environment. What's its final steady state? Maybe cyclic steady state- always changing or cycling.

 
6.2 The Goal

Survival is out as a motivation.

What is good/bad pre-programmed -> There are goals it has - when these are found in animals they are also accompanied by - instincts that match and what else?

Why does it need to think / learn - maximize or minimize something

What? 

  • Length of life
  • defend itself
  • amount of energy / wealth
  • amount of info/knowledge (about its environment) / understanding of environment
  • understanding of itself

5.3 Intelligence

Testing for intelligence suffers from lack of a definition of intelligence.

25th Aug 1995 + There is a need for a test to recognize and acknowledge that something is learning/thinking/sentient. It needs to test for the following aspects/dimensions/processes of intelligence: Recognition, Learning, Thinking, and Communicating. Then there is a need for a systems test which tests all of these working together to exhibit intelligent behaviour.