In 1982, Carnegie Mellon University professor Scott Fahlman suggested using:-) for humorous comments after his colleagues took a joke about mercury seriously. On September 19, 1982, Carnegie Mellon University computer science research assistant professor Scott Fahlman posted a message to the university's bulletin board software that would later come to shape how people communicate online. His proposal: use:-) and:-( as markers to distinguish jokes from serious comments. While Fahlman describes himself as "the inventor or at least one of the inventors" of what would later be called the smiley face emoticon, the full story reveals something more interesting than a lone genius moment. The whole episode started three days earlier when computer scientist Neil Swartz posed a physics problem to colleagues on Carnegie Mellon's "bboard," which was an early online message board.

Recurrent Cascade-Correlation CRCC) is a recurrent version of the Cascade(cid:173) Correlation learning architecture of Fah I man and Lebiere [Fahlman, 1990]. RCC can learn from examples to map a sequence of inputs into a desired sequence of outputs. New hidden units with recurrent connections are added to the network as needed during training. In effect, the network builds up a finite-state machine tailored specifically for the current problem. RCC retains the advantages of Cascade-Correlation: fast learning, good generalization, automatic construction of a near-minimal multi-layered network, and incremental training.

In an earlier paper, I described in some detail how a system based on symbolic knowledge representation and reasoning could model and reason about an act of deception encountered in a children's story. This short position paper extends that earlier work, adding new analysis and discussion about the nature of deception, the desirability of building deceptive AI systems, and the computational mechanisms necessary for deceiving others and for recognizing their attempts to deceive us.

This supposedly makes representing exceptions (three-legged elephants and the like) easy; but, alas, it makes one crucial type of representation impossiblethat of composite descriptions whose meanings are functions of the structure and interrelation of their parts. This article explores this and other ramifications of the emphasis on default properties and "typical" objects. While I believe this to be an important point, this article was never meant to be the definitive work on logical distinctions in knowledge representation. Some of the notions mentioned here in passing (e.g., analyticity) are perenially problematic. In addition, I have not really attempted to bring the body of the article up to date from its original form. The article is also generally nonconstructive. However, there is now ample evidence that this kind of analysis can lead to constructive suggestions for knowledge representation systems. In work pursued after the original version of this article was written, some ...

Scone is a knowledge-base system developed specifically to support human-like common-sense reasoning and the understanding of human language. One of the unusual features of Scone is its multiple-context system. Each context represents a distinct world-model, but a context can inherit most of the knowledge of another context, explicitly representing just the differences. We explore how this multiple-context mechanism can be used to emulate some aspects of human mental behavior that are difficult or impossible to emulate in other representational formalisms. These include reasoning about hypothetical or counter-factual situations; understanding how the world model changes over time due to specific actions or spontaneous changes; and reasoning about the knowledge and beliefs of other agents, and how their mental state may affect the actions of those agents.

The first class of network adaptation algorithms start out with a redundant architecture and proceed by pruning away seemingly unimportant weights (Sietsma and Dow, 1988; Le Cun et aI, 1990). A second class of algorithms starts off with a sparse architecture and grows the network to the complexity required by the problem. Several algorithms have been proposed for growing feedforward networks. The upstart algorithm of Frean (1990) and the cascade-correlation algorithm of Fahlman (1990) are examples of this approach.

The first class of network adaptation algorithms start out with a redundant architecture and proceed by pruning away seemingly unimportant weights (Sietsma and Dow, 1988; Le Cun et aI, 1990). A second class of algorithms starts off with a sparse architecture and grows the network to the complexity required by the problem. Several algorithms have been proposed for growing feedforward networks. The upstart algorithm of Frean (1990) and the cascade-correlation algorithm of Fahlman (1990) are examples of this approach.

The first class of network adaptation algorithms start out with a redundant architecture and proceed by pruning away seemingly unimportant weights (Sietsma and Dow, 1988; Le Cun et aI, 1990). A second class of algorithms starts off with a sparse architecture and grows the network to the complexity required by the problem. Several algorithms have been proposed for growing feedforward networks. The upstart algorithm of Frean (1990) and the cascade-correlation algorithm of Fahlman (1990) are examples of this approach.

Existing metrics for the learning performance of feed-forward neural networks do not provide a satisfactory basis for comparison because the choice of the training epoch limit can determine the results of the comparison. I propose new metrics which have the desirable property of being independent of the training epoch limit. The efficiency measures the yield of correct networks in proportion to the training effort expended. The optimal epoch limit provides the greatest efficiency. The learning performance is modelled statistically, and asymptotic performance is estimated. Implementation details may be found in (Harney, 1992).

Existing metrics for the learning performance of feed-forward neural networks do not provide a satisfactory basis for comparison because the choice of the training epoch limit can determine the results of the comparison. I propose new metrics which have the desirable property of being independent of the training epoch limit. The efficiency measures the yield of correct networks in proportion to the training effort expended. The optimal epoch limit provides the greatest efficiency. The learning performance is modelled statistically, and asymptotic performance is estimated. Implementation details may be found in (Harney, 1992).