Read on Twitter
The #DarkRoom discussion continues!
Our initial piece (which revived the problem and reviewed some solutions) now has replies from Karl Friston, @fluffycyborg, @anilkseth, Colin Klein, @sandervdc, @drclbuckley, @BerenMillidge, & @a_tschantz. We reply to them in turn.
Our initial piece (which revived the problem and reviewed some solutions) now has replies from Karl Friston, @fluffycyborg, @anilkseth, Colin Klein, @sandervdc, @drclbuckley, @BerenMillidge, & @a_tschantz. We reply to them in turn.
If agents are just prediction-error minimizers, why don't they remain in highly predictable environments—like a dark, empty room? This is the Dark Room Problem. It's been around for a while, but some new work has revived it, as we note in a recent paper. https://www.sciencedirect.com/science/article/pii/S1364661320300589
Colin Klein (whose work we review) thinks the Dark Room Problem has been severely underestimated by proponents of PP/FEP:
"There is no free lunch. The simplifications that PP
introduces incur a debt, and it is not a given that it can be repaid." https://psyarxiv.com/vdswk
"There is no free lunch. The simplifications that PP
introduces incur a debt, and it is not a given that it can be repaid." https://psyarxiv.com/vdswk
Van de Cruys ( @sandervdc), Friston & Clark disagree (w/ us & Klein). They point to optimistic biases as a mechanism by which surprise-minimizers can escape Dark Rooms (& they make other excellent points).
Don't take my word for it—read their reply here: https://www.cell.com/trends/cognitive-sciences/fulltext/S1364-6613(20)30136-4
Don't take my word for it—read their reply here: https://www.cell.com/trends/cognitive-sciences/fulltext/S1364-6613(20)30136-4
Seth ( @anilkseth) & colleagues offer another path out of Dark Rooms: Curiosity & exploration. Maybe leaving the dark room now (& incurring a short-term prediction-error hit) leads to better predictions later.
Of course, they put it much better than I do: https://www.cell.com/trends/cognitive-sciences/fulltext/S1364-6613(20)30135-2
Of course, they put it much better than I do: https://www.cell.com/trends/cognitive-sciences/fulltext/S1364-6613(20)30135-2
Finally, we get the last word. Zekun & I think "optimism" comes with a huge cost that gives away much of what makes PP so exciting. And we think that exploration, though important, can't account for the most interesting surprise-welcoming human behaviors. https://www.cell.com/trends/cognitive-sciences/fulltext/S1364-6613(20)30141-8
Don't like paywalls? Here are some links you might find helpful...
Sun & Firestone: https://philpapers.org/archive/SUNTDR-2.pdf
Klein: https://psyarxiv.com/vdswk/download
Seth et al.: https://perception.jhu.edu/files/PDFs/20_DarkRoom/SethEtAl_DarkRoomReply_TiCS_InPress.pdf
Van de Cruys et al.: https://perception.jhu.edu/files/PDFs/20_DarkRoom/VandeCruysFristonClark_DarkRoomReply_TiCS_InPress.pdf
Sun & Firestone Rd 2: https://philpapers.org/archive/SUNOAP.pdf
Sun & Firestone: https://philpapers.org/archive/SUNTDR-2.pdf
Klein: https://psyarxiv.com/vdswk/download
Seth et al.: https://perception.jhu.edu/files/PDFs/20_DarkRoom/SethEtAl_DarkRoomReply_TiCS_InPress.pdf
Van de Cruys et al.: https://perception.jhu.edu/files/PDFs/20_DarkRoom/VandeCruysFristonClark_DarkRoomReply_TiCS_InPress.pdf
Sun & Firestone Rd 2: https://philpapers.org/archive/SUNOAP.pdf
This was a pleasure to be part of! Thanks to @TrendsCognSci for making it happen, & to the fantastic scientists + philosophers who wrote in (both pro and con). We didn't quite imagine we'd get to trade thoughts with Friston himself and so many others; it was a treat to do so.
Prediction is a fascinating lens through which to view human perception, cognition, motivation, and action. It seems undeniable that it is part of the end-story of what goes on in our heads. We don't think it's the whole story, though, and we're glad we got a chance to say why!
(Thanks also to several people who gave super insightful comments to us throughout this: @De_dicto, @NeuroPolarbear, @danwilliamsphil, @ibphillips, @jorgemlg, Colin Klein, Steven Gross [ @jhu_philosophy], and many other members of our lab.)
