Seven Levels of AI: Putting Mental and Metal to Work, Part 2

by Evan Kirstel (Linkedin) & @evankirstel (Twitter)

To cope with the changes that will no doubt take place in making AI a reality, industry-wide standards would be beneficial. Proposing standards initiates a long list of arguments about what elements should exist where, but it is understood that, as the result of an evolutionary process, these standards might be rearranged many times before agreements are reached. To keep this process relatively simple, I propose seven levels of interaction:

1 - Physical: the physical recognition of an experience

2 - Interpretive: putting such experiences into recognizable patterns

3 - Ambiguous: adjusting those patterns for focus or open-ended manipulation

4 - Thinking: associating information with other experiences

5 - Cognitive: active manipulation and integration of symbols

6 - Contextual: prioritization of symbols, issues, and action items

7 - Philosophical: none of the above

In creating meanings for each of these levels, it is understood that knowledge can move from one level to another without any intervention from the others. Also, it is a given that each of these levels has a positive and negative element (for example, +10, 0, -10). It is possible that humans might deal with ambiguity by cheating their way through an uncertain or unknown activity. Within each level, elements such as repetition and similarity allow humans to create mental replays of past events without physically experiencing them in the present, which would lead to confusion and ambiguity about real versus imagined.

The human machine has the ability to vary the elements of a particular event, thus allowing additional and more diverse experiences to occur in the mind. The interpretive level focuses on the first level of understanding, beyond simple recognition of the experience or object. For example, the physical recognition that the object ahead of us is a grizzly bear can be more important to our survival than a philosophical discussion of a bear's eating habits. At the same time, interpreting the hundreds of details that occur in an hour while working at one's desk requires a deeper understanding of input-output issues than is required while eating dinner.

Ambiguity is similar to interpretation, but it is a narrowing or widening process. Sometimes, interpretation needs to be rapidly narrowed to make a quick decision. Visit any hamburger stand. You are faced with entering the building, interpreting and narrowing the choices, and making a decision about what to order.

You can build AI with ambiguity skills such as cheating, wild cards, and tricks that make the AI interpretive process faster or more enjoyable. There are other mental power tools that can be compared to a vast array of AI programs. Each of these programs contains a test to see if the activity can be accomplished successfully. This standards approach to packaging of skills allows us to perform many activities without thinking about them. Even complex tasks such as driving home from work are relatively automatic once they have occurred with some frequency.

Remember that, to a point, activities can skip to higher levels. In some cases, events can be considered philosophically before any physical action takes place. As I pointed out in Part One, proposing standards initiates a long list of arguments about what elements should exist where, but it is understood that as the result of an evolutionary process, these standards might be rearranged many times before an agreement is reached. Standardized systems like alphabets, words, grammar and language serve as models for making machines more human-like and for helping humans understand how machines process information.

This is an exciting subject and I invite your comments — email me at [email protected] and follow me on Twitter at https://twitter.com/evankirstel.