The present is a proto-past to multiple possible futures — and anticipating which futures we might get is best done using classical historiographical methods, with the temporal telescope reversed
An excellent and timely framework as the AI debate right now is almost entirely trapped in the structuralist mode you warning against. Just saw an OpenAI researcher who tweeted that organisations which cede more control to AI will inevitably outcompete those that don't — competitive dynamics as destiny, the Schwartz Window containing exactly one future. That feels like modernisation theory with a large compute budget.
Your structure/agency/contingency triad is so much more holistic. The competitive logic is real, but treating it as inevitable ignores the agency of institutional design — we chose the NPT, we chose factory legislation, we chose antitrust at other times of change. None were structurally predetermined. And Brand's insight applies: the people most confidently narrowing the window are the ones most likely to be caught staring in disbelief when the actual future arrives.
I wrote about a related moment last year — when the Pope called for moral discernment on AI and Silicon Valley's reflex was mockery before reasoning caught up. The flinch was the diagnostic: not what anyone believed, but what the nervous system did before belief arrived. That's agency and contingency in your frame — the structural forces don't explain the sequence, only the terrain on which it happened.
Nils, this is a vital corrective to linear futurology. Your structure/agency/contingency triad captures the real mechanics of historical time, but I’d push it slightly into the philosophical register by bringing in Derrida’s hauntology and what I’ll tentatively call potentiology.
Hauntology describes how the past (especially unrealized or repressed futures) lives in the present as a spectral force. This maps beautifully onto your structural tectonic plates: they aren’t inert backdrops but accumulated legacies, debts, and unresolved trajectories that actively shape the terrain.
Potentiology (not Derrida’s term, but a useful extension) is the inverse: the way multiple potential futures already inhabit the present, exerting anticipatory gravity. Your “proto-pasts” and the Schwartz Window are essentially mapping potentiological space. These futures aren’t just “out there;" they're already vibrating in our institutions, technologies, and collective imagination.
On contingency, you’re right to rescue it from the residual-category dustbin. It also strongly echoes Nassim Nicholas Taleb’s Black Swan theory. Both of you highlight how retrospective narrative fallacy makes high-impact, unpredictable events look inevitable after the fact. The difference is mostly disciplinary: Taleb focuses on statistical ignorance and system robustness; you track the historical interplay of structure and agency.
One thing I’d add: the Schwartz Window isn’t just cognitive. It’s narrative infrastructure. What a society deems “plausible” is actively curated by the stories we tell, which in turn constrain or enable agency. If hauntology reminds us we’re haunted by lost futures, and potentiology reminds us we’re gestating possible futures, then structured anticipation becomes less about prediction and more about curating which specters we invite to the table.
Thanks for the Farrell/Shalizi piece which is excellent. Their argument that AI is a social technology in the lineage of bureaucracy and markets converges remarkably with a writer I follow (ChorPharn at The Cutting Floor) who arrives at a similar conclusion from political economy lens rather than complexity science — that AI doesn't arrive on a blank shore but enters a world already formatted by older technologies of rule. The convergence from such different starting points feels like evidence the insight is real.
A lot of interesting food for thought in this post! This frameworks does seem much more descriptive and useful than the previous simpler ones you mentioned.
I think one intuition from system theory would be that over the very long run, both agency and contingency might act as noise, with many of these events roughly canceling each other out, so the results would be determined mostly by structure. Conversely over the medium-short run contingency and agency are kings.
You could almost think of it in term a mathematical model in which the dynamics of the system (assuming you know how to describe it which we don't) gives the structure, while agency and contingency are unknown forcing terms.
Anyway, that said, because the system locally unstable -most likely chaotic- it is highly path-dependent, so we cannot exclude that agency and contingency could actually put us on a path that shapes the structure (so to speak) in an irreversible way.
Yes, I agree with this completely. To take a very salient example: “technodeterminism” is largely wrong in the short run (because agents can make choices and contingency affects rollouts) but largely correct in the long run, because technologies have specific affordances that somebody will almost inevitably eventually exploit.
Excellent summary. Given the extensive structural changes in the present and in the horizon, how do you see scenario planning processes now vs. 15-20 years ago (I was at GBN 2003-7)?
An excellent and timely framework as the AI debate right now is almost entirely trapped in the structuralist mode you warning against. Just saw an OpenAI researcher who tweeted that organisations which cede more control to AI will inevitably outcompete those that don't — competitive dynamics as destiny, the Schwartz Window containing exactly one future. That feels like modernisation theory with a large compute budget.
Your structure/agency/contingency triad is so much more holistic. The competitive logic is real, but treating it as inevitable ignores the agency of institutional design — we chose the NPT, we chose factory legislation, we chose antitrust at other times of change. None were structurally predetermined. And Brand's insight applies: the people most confidently narrowing the window are the ones most likely to be caught staring in disbelief when the actual future arrives.
I wrote about a related moment last year — when the Pope called for moral discernment on AI and Silicon Valley's reflex was mockery before reasoning caught up. The flinch was the diagnostic: not what anyone believed, but what the nervous system did before belief arrived. That's agency and contingency in your frame — the structural forces don't explain the sequence, only the terrain on which it happened.
https://rajeshachanta.substack.com/p/the-pope-said-something-boring?utm_source=publication-search
Nils, this is a vital corrective to linear futurology. Your structure/agency/contingency triad captures the real mechanics of historical time, but I’d push it slightly into the philosophical register by bringing in Derrida’s hauntology and what I’ll tentatively call potentiology.
Hauntology describes how the past (especially unrealized or repressed futures) lives in the present as a spectral force. This maps beautifully onto your structural tectonic plates: they aren’t inert backdrops but accumulated legacies, debts, and unresolved trajectories that actively shape the terrain.
Potentiology (not Derrida’s term, but a useful extension) is the inverse: the way multiple potential futures already inhabit the present, exerting anticipatory gravity. Your “proto-pasts” and the Schwartz Window are essentially mapping potentiological space. These futures aren’t just “out there;" they're already vibrating in our institutions, technologies, and collective imagination.
On contingency, you’re right to rescue it from the residual-category dustbin. It also strongly echoes Nassim Nicholas Taleb’s Black Swan theory. Both of you highlight how retrospective narrative fallacy makes high-impact, unpredictable events look inevitable after the fact. The difference is mostly disciplinary: Taleb focuses on statistical ignorance and system robustness; you track the historical interplay of structure and agency.
One thing I’d add: the Schwartz Window isn’t just cognitive. It’s narrative infrastructure. What a society deems “plausible” is actively curated by the stories we tell, which in turn constrain or enable agency. If hauntology reminds us we’re haunted by lost futures, and potentiology reminds us we’re gestating possible futures, then structured anticipation becomes less about prediction and more about curating which specters we invite to the table.
Easily the best read in a long time. Had (enjoyable) flashbacks to many, many historiography classes. Thank you
Very fine, Nils. Relatedly but perhaps tangentially, have you seen Farell and Shalizi's latest, on the role of AI being similar to that of capitalism, bureaucracy, or the Long Industrial Revolution? Worthwhile reading. https://knightcolumbia.org/content/ai-as-social-technology?utm_source=substack&utm_medium=email
Thanks for the Farrell/Shalizi piece which is excellent. Their argument that AI is a social technology in the lineage of bureaucracy and markets converges remarkably with a writer I follow (ChorPharn at The Cutting Floor) who arrives at a similar conclusion from political economy lens rather than complexity science — that AI doesn't arrive on a blank shore but enters a world already formatted by older technologies of rule. The convergence from such different starting points feels like evidence the insight is real.
A lot of interesting food for thought in this post! This frameworks does seem much more descriptive and useful than the previous simpler ones you mentioned.
I think one intuition from system theory would be that over the very long run, both agency and contingency might act as noise, with many of these events roughly canceling each other out, so the results would be determined mostly by structure. Conversely over the medium-short run contingency and agency are kings.
You could almost think of it in term a mathematical model in which the dynamics of the system (assuming you know how to describe it which we don't) gives the structure, while agency and contingency are unknown forcing terms.
Anyway, that said, because the system locally unstable -most likely chaotic- it is highly path-dependent, so we cannot exclude that agency and contingency could actually put us on a path that shapes the structure (so to speak) in an irreversible way.
Anyway, really interesting, thanks.
Yes, I agree with this completely. To take a very salient example: “technodeterminism” is largely wrong in the short run (because agents can make choices and contingency affects rollouts) but largely correct in the long run, because technologies have specific affordances that somebody will almost inevitably eventually exploit.
Excellent summary. Given the extensive structural changes in the present and in the horizon, how do you see scenario planning processes now vs. 15-20 years ago (I was at GBN 2003-7)?