Lossless and lossy compression are both useful for AGI. We use lossless
compression to test prediction accuracy. We use lossy compression to study
models of human perception and cognition.

Lossy compression can only test deterministic predictors because the
decoder must reproduce the exact same sequence of predictions. This
normally isn't a problem even though the brain is not deterministic.

Lossy compression is harder to evaluate because it requires subjective
human evaluation of quality. It can't be automated or reproduced because
evaluation is itself an AI problem. But we know from human memory tests
that video can in theory be compressed to 10 bits per second, in the sense
that if you watched the original video and its reconstruction from a script
one after the other, you couldn't tell if they were different. We have a
long way to go on this problem.

------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T5ff6237e11d945fb-M377f1d41b307314a35a44411
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to