Who can use the Technical AI Forecasting Dictionary?

The Dictionary is distributed under the MIT license, which means that it’s freely available to use by any project.

Are you trying to establish a consensus interpretation of terms among AI researchers?

No, for two reasons. First, we are simply trying to provide a standard of interpretation to serve a very particular purpose (forecasting). What matters here is not what the “one true definition” is, but rather whether our definition captures at least some natural category of things. We are also interested in highlighting when a term is used differently by different groups or individuals, or seems unclear to many people, so that forecasting questions can make sure to either choose a specific definition (whether the one in the dictionary or not,) or avoid the use of the unclear or dsiputed terms.

Second, the Dictionary is targeted at researchers and practitioners of forecasting, and this set is not co-extensive with the set of AI researchers, even though there is overlap.

Nonetheless, the Dictionary is open-source and if others find it helpful we welcome their usage.

Who created the Dictionary and why?

The Dictionary was created by Jacob Lagerros and Ben Goldhaber (Parallel, LLC) in collaboration with users of forecasting platform Metaculus AI, funded by the Berkeley Existential Risk Initiative. David Manheim is currently actively working on developing it further.

It is a piece of infrastructure for the public good. We created it because we care about society getting the transition to AI right, and because we think forecasting can help with that.