Notes by Retraice

Re23-NOTES.pdf


Listen Later

(The below text version of the notes is for search purposes and convenience. See the PDF version for proper formatting such as bold, italics, etc., and graphics where applicable. Copyright: 2022 Retraice, Inc.)

Re23: You Need a World Model

Retraice^1

What's going on out there is hard to know and the deadlines are real.

Air date: Wednesday, 19th Oct. 2022, 10:35 PM Eastern/US.

Get serious

If (a) what's going on out there is hard to know, and (b) real deadlines are rapidly approaching, we can either wing it (the default option), or get serious.

To get serious is to get a world model.

What the hell is he talking about?

This is an understandable feeling, when dealing with generalists, who know almost nothing about almost everything. Compare specialists, who know almost everything about almost nothing, or more and more about less and less.

What is a world model?

The term `world model' is not common. In the AI literature, there is some use of it.^2

At this point, we don't have a definition. But it is what it sounds like.

Why do we need a world model?

Consider what we've already discussed on Retraice (and some things we haven't yet, namely H14-H17), and how complicated and complex it all is: o Re1: Intelligence Kinds o Re2: Telling Friends vs. Foes o Re3: Information Problems o Re4: Trust Problems o Re5: Survival Problems o Re6: AI Aspects o Re7: Goals o Re8: Machines and Life o Re9: AI Perception o Re10: Intelligence Guess-Check-Fight o Re11: Intelligence Travel o Re12: Aliens Thinking How-vs-What o Re13: Intelligence Constraints on Behavior o Re14: Omniscience Physical Animate Mental o Re15: Trust is a Mess o Re16: Trust is a Response o Re17: GOOT is Hypotheses, Current History o Re18: Hypothesis Testing by Numbers o Re19: Nature Is Not F-ing Around o Re20: The Deadline Problem o Re21: Time's Up o Re22: Computer Control o H1: Space o H2: Technology o H3: Death o H4: China o H5: U.S. Civil War o H6: Environment o H7: Better o H8: Intelligence o H9: Darkness o H10: Wealth o H11: Wildcards o H12: Computer Control o H13: Something Bigger o H14: Hong Kong o H15: Taiwan o H16: Philippines o H17: Ukraine o combinations (there are a lot) o sources (there are a lot)

Not winging it

Almost everybody is just winging it. There are very, very few people who've produced anything that resembles a world model. It's hard.

And in-my-head doesn't count. World models that only exist as neural networks inside skulls are not scrutable, not open to inspection, very hard to shine sunlight on and disinfect.

_

References

Ha, D., & Schmidhuber, J. (2018). World models. arxiv.org. [Submitted on 27 Mar 2018 (v1), last revised 9 May 2018 (this version, v4)] https://doi.org/10.48550/arXiv.1803.10122 Retrieved 19th Oct. 2022.

Russell, S., & Norvig, P. (2020). Artificial Intelligence: A Modern Approach. Pearson, 4th ed. ISBN: 978-0134610993. Searches: https://www.amazon.com/s?k=978-0134610993 https://www.google.com/search?q=isbn+978-0134610993 https://lccn.loc.gov/2019047498

Footnotes

^1 https://www.retraice.com/retraice

^2 Russell & Norvig (2020) p. 848; Ha & Schmidhuber (2018), https://worldmodels.github.io/ and Hallucinogenic Deep Reinforcement Learning Using Python and Keras

...more
View all episodesView all episodes
Download on the App Store

Notes by RetraiceBy Retraice, Inc.