...

This paper is substantially changed from [McCarthy, 1996] which was given at Machine Intelligence 15 in 1995 August held at Oxford University.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...sentences.

Newell, together with Herbert Simon and other collaborators used logic as a domain for AI in the 1950s. Here the AI was in programs for making proofs and not in the information represented in the logical sentences.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...externally.

Here's an ancient example of observing one's likes and not knowing the reason.

``Non amo te, Zabidi, nec possum dicere quare;
Hoc tantum possum dicere, non amo te.''

by Martial which Tom Brown translated to

I do not like thee, Dr. Fell
The reason why I cannot tell,
But this I know, I know full well,
I do not like thee, Dr. Fell.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...swimming.

One can understand aspects of a human activity better than the people who are good at doing it. Nadia Comenici's gymnastics coach was a large, portly man hard to imagine cavorting on a gymnastics bar. Nevertheless, he understood women's gymnastics well enough to have coached a world champion.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...subject.

Too much work concerned with self-knowledge has considered self-referential sentences and getting around their apparent paradoxes. This is mostly a distraction for AI, because human self-consciousness and the self-consciousness we need to build into robots almost never involves self-referential sentences or other self-referential linguistic constructions. A simple reference to oneself is not a self-referential linguistic construction, because it isn't done by a sentence that refers to itself.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...number.

Some other formalisms give up the law of substitution in logic in order to avoid this difficulty. We find the price of having separate terms for concepts worth paying in order to retain all the resources of first order logic and even higher order logic when needed.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...found.

For AI it might be convenient to use unrestricted comprehension as a default, with the default to the limited later by finding an A if necessary. This idea has not been explored yet.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...axioms.

We assume that our axioms are strong enough to do symbolic computation which requires the same strength as arithmetic. I think we won't get much joy from weaker systems.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...false. 

A conviction of about what is relevant is responsible for a person's initial reaction to the well-known puzzle of the three activists and the bear. Three Greenpeace activists have just won a battle to protect the bears' prey, the bears being already protected. It was hard work, and they decide to go see the bears whose representatives they consider themselves to have been. They wander about with their cameras, each going his own way.

Meanwhile a bear wakes up from a long sleep very hungry and heads South. After three miles, she comes across one of the activists and eats him. She then goes three miles West, finds another activist and eats her. Three miles North she finds a third activist but is too full to eat. However, annoyed by the incessant blather, she kills the remaining activist and drags him two miles East to her starting point for a nap, certain that she and her cubs can have a snack when she wakes.

What color was the bear?

At first sight it seems that the color of the bear cannot be determined from the information given. While wrong in this case, jumping to such conclusions about what is relevant is more often than not the correct thing to do.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...consistency.

Our approach is a variant of that used by [Kraus et al., 1991].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...consciousness.

Cindy Mason in her Emotional Machines home page (http://www.emotionalmachines.com/) expresses a different point of view.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...consciousness.

These conclusions are true in the simplest or most standard or otherwise minimal models of the ideas taken in consciousness. The point about nonmonotonicity is absolutely critical to understanding these ideas about emotion. See, for example, [McCarthy, 1980] and [McCarthy, 1986]

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...

2001: The Steven Spielberg movie, Artificial Intelligence illustrates dangers of making robots that partly imitate humans and inserting them into society. I say ``illustrates'' rather ``than provides evidence for'', because a movie can illustrate any proposition the makers want, unrestricted by science or human psychology. In the movie, a robot boy is created to replace a lost child. However, the robot does not grow and is immortal and therefore cannot fit into a human family, although they depict it as programmed to love the bereaved mother. It has additional gratuitous differences from humans.

The movie also illustrates Spielberg's doctrines about environmental disaster and human prejudice against those who are different.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.