Artificial Trust definitions

Trust

A computational shortcut, in the form of a saved memory that represents a decision that has been already been made. Trust reduces the computation needed to be certain, because it relies on computation that has already been done (either by others or by oneself).

A saved memory is considered trustworthy (or not) by the trust recognizer. In general, we trust our memories, i.e. the trust recognizer usually says “this memory is trustworthy.” This is why all intelligences have confirmation bias.

Without trust, an intelligence would become paralyzed by indecision. Any one decision could potentially take infinite effort, since every computational result that the decision relies on would need to be re-checked again and again.

Artificial trust

Artificial trust closely resembles natural trust, as it reduces the computation needed to achieve confidence in a result. But in artificial trust, this confidence is false, because the results are not reliably valid.

Artificial trust is caused by failures in the trust recognizer, either in its algorithm or in the data it is trained on. These failures may arise from poor design, or may be the result of hostile manipulation by an adversary.

Trust recognizer

A computational mechanism for deciding what can be trusted, and therefore, not re-checked.

Trusted source

A source whose information does not need to be double-checked for correctness. In machine learning, all training data is typically assumed to come from a trusted source.