Amplification and dialogue as mechanisms for safe advanced AI

Add to your list(s) Download to your calendar using vCal

• Beth Barnes, Computer Lab, University of Cambridge
• Wednesday 24 January 2018, 17:00-18:30
• Cambridge University Engineering Department, CBL Seminar room BE4-38. For directions see http://learning.eng.cam.ac.uk/Public/Directions.

If you have a question about this talk, please contact Adrià Garriga Alonso.

Slides: https://valuealignment.ml/talks/2018-01-24-amplification.pdf

These techniques come at the problem of safety from a fairly different angle than the things we’ve discussed so far.

Amplification is the idea of bootstrapping a trusted core system, increasing its capabilities while maintaining safety properties. Paul Christiano and the OpenAI safety team have worked on these ideas. One current suggestion for how to do this has a lot in common with functional programming. For some more discussion see e.g. https://ai-alignment.com/alba-an-explicit-proposal-for-aligned-ai-17a55f60bbcf

This talk is part of the Engineering Safe AI series.

This talk is included in these lists:

• Cambridge University Engineering Department, CBL Seminar room BE4-38. For directions see http://learning.eng.cam.ac.uk/Public/Directions
• Cambridge talks
• Chris Davis' list
• Engineering Safe AI
• Trust & Technology Initiative - interesting events

Note that ex-directory lists are not shown.