Thursday, February 11, 2016

Levels of Machine Thinking

If making a simple decision based on stored instructions can be considered as the most fundamental form of machine thinking, what capability should be added to go to the next level of thinking? Is it possible to provide  control algorithms that would permit the device to make some choices on its own, perhaps based on newly stored data or upon trial and error until it gets some expected result?

Humans and animals can be influenced in such situations by being rewarded for making one kind of decision or punished for choosing the alternative. BUT.  How does one reward or punish a machine?

Computers don’t know anything. Despite having huge amounts of information stored in its memory banks, a computer just waits for instructions telling it how to retrieve and manipulate this information.  As an example, it has no concept of what a number is. Surprise! Yet numbers are the basis for all of its actions. If one Opens a word processing Application [App] and  types ‘3’, a 3 shows up on the display and a 3 can be printed by a connected printer. The keyboard contains a rudimentary computer that encodes each keystroke with a number from a look-up table [ASCII] that is sent to the word processing App. If one types ‘three’ the computer receives ’74 65 72 65 65’ [octal] and the App displays ‘three’ by converting the ASCII numbers back to text. [Octal is a shortcut way of writing a binary number. What is really sent for ’t’ is 0100 1010].  If a user Opens a Calculator App it accepts only ASCII values which represent numbers on a keyboard.  If one addresses a computer vocally; such as, “Siri, what is three?” {Without knowing the details] Siri examines a stored dictionary and speaks the entry, “three, a cardinal number, equivalent to the sum of one and two; one more than two.”  There is no App with which one can enter … Computer. Learn what three is. 

I’m convinced that just the process of storing data does not, in itself, constitute Learning. We need to learn what a Learning Algorithm should be able to do.  Certainly ‘understanding’ is part of Learning and probably a constituent of Consciousness, but how would one code it?

No comments:

Post a Comment