Can you pin down what you specifically mean when you say that “our civilization” will “fall” like the Roman Empire?
By when will it fall? How do you define fall? What is “our civilization”, do you mean the US? The west? All of post-enlightenment culture worldwide?
Like Bryan says - I’d be happy to take the other side of this bet if you can operationalize it in a clear way. I think you’re extrapolating too much from a temporally localized phenomenon.
I’ve reached a private v0.1 demo checkpoint for MirrorOS.
MirrorOS is a governance-before-action runtime for AI agents.
The core idea: AI models can propose actions, but they should not automatically receive authority to execute those actions.
MirrorOS is designed to sit at the execution boundary, evaluate proposed actions against policy, decide whether they are allowed, denied, or escalated, and preserve an audit trail.
The private demo now shows both sides:
unsafe modifying action → denied/escalated and audited
safe read action → allowed and audited
I’m preparing trusted reviewer conversations now, especially with people who understand AI governance, compliance, security, enterprise software, or public-sector risk.
I am committed to really diving into your work further. The dynamics of organizations (and the dominant trend of hyper-bureaucracies devouring our society) are an element that is too little analyzed. We often seem to take the changes for granted, while they instigate dramatic cultural changes. After 20 years these become a 'new normal,' rather than what they would've been a generation ago (florid pathology).
I hope you'll read my recent essay and give me your thoughts. I would actually like to communicate about some of this directly. I'm going to send you a message right now.
> Besides participating as suppliers, customers, employees, or targets of such orgs, there are two other key ways we engage such orgs: politics and governance.
I think this framing elides the fact that politics and governance are also org-based activities, and we interact with politics and governance primarily as employees or targets of those orgs, with donors and/or voters as analogs for the customer role.
Decisions made on an org-by-org level are what govern the course of political events. Even leaving aside the simple ones like local/county/state/national parties, there are a bunch of orgs that make up the things we in the US just lump together the national-level of the party; separately there is a large number of orgs for fundraising, for lobbying, for polling, for voter research outside of polling (ie social media and purchased market data, on account of these being different expertise). In the U.S. these are primarily non-profit private entities, though service providers like analysts, consultants, etc. work the same way as they do in the private sector (with the added wrinkle of both major parties supporting largely independent networks of these organizations). Governance questions add orgs like campaign regulators, media regulators, commissions for things like redistricting, placement of polling places, how votes are counted, etc.
It seems to me like the research on questions of regulation of the private sector should apply pretty directly in many cases.
Deliciously depressing, most important insights. You've had me working on 'Meta-Governance' since reading your fab thougths this morning - and how we might introduce such Meta capability into organisations without 'top dogs' feeling threatened.
Sortition-based governance is the answer, imo. Any institution or organization will inevitably become dominated by those who are good at politicking rather than those who are good at fulfilling the organization's formal objectives. Stochastic selection processes dramatically reduce the expected value of politicking. I think deterministic hiring/promotion should rarely be used. With the exception of extremely specialized roles, it's better to choose at random from a pool of qualified candidates, as it weakens incentives to subvert the selection criteria.
Recent experiments with sortition-based governance have performed well. Paris and Belgium now have permanent citizens assemblies with rotating membership. Having moderated such an assembly in the US and seen the discussions up close, I can affirm that randomly selected citizens are capable enough of rapidly becoming well-informed on fairly complicated policy issues to make far better judgements than the ignorant voting public, and usually better than elected officials.
I think it doesn't make much sense for small organizations where coordination is easier. Sortition in the sense of citizens assemblies for governance probably only makes sense at the scale of >100k people. Though I think stochastic hiring/promotion would be useful even in organizations of a few hundred or less.
That said, I think leaving decisions to chance doesn't come naturally to most people, and indeed most people even seem averse to it. I could speculate on why that might be, there are all sorts of reasons. Maybe I'll write about this.
Hence the need for our AI overlords. Complex organizations need a central processing unit for some range of tasks, coordination and conflict management.
Artificial super intelligence isn’t just a potential threat, it is also a potential savior.
The cells in our finger nails shouldn’t expect to hold our brains accountable. The animals in the San Diego Safari Park don’t hold the organization that runs it accountable.
> and seek new better to choose
"New better" what?
fixed; thanks
Can you pin down what you specifically mean when you say that “our civilization” will “fall” like the Roman Empire?
By when will it fall? How do you define fall? What is “our civilization”, do you mean the US? The west? All of post-enlightenment culture worldwide?
Like Bryan says - I’d be happy to take the other side of this bet if you can operationalize it in a clear way. I think you’re extrapolating too much from a temporally localized phenomenon.
I’ve reached a private v0.1 demo checkpoint for MirrorOS.
MirrorOS is a governance-before-action runtime for AI agents.
The core idea: AI models can propose actions, but they should not automatically receive authority to execute those actions.
MirrorOS is designed to sit at the execution boundary, evaluate proposed actions against policy, decide whether they are allowed, denied, or escalated, and preserve an audit trail.
The private demo now shows both sides:
unsafe modifying action → denied/escalated and audited
safe read action → allowed and audited
I’m preparing trusted reviewer conversations now, especially with people who understand AI governance, compliance, security, enterprise software, or public-sector risk.
I am committed to really diving into your work further. The dynamics of organizations (and the dominant trend of hyper-bureaucracies devouring our society) are an element that is too little analyzed. We often seem to take the changes for granted, while they instigate dramatic cultural changes. After 20 years these become a 'new normal,' rather than what they would've been a generation ago (florid pathology).
I hope you'll read my recent essay and give me your thoughts. I would actually like to communicate about some of this directly. I'm going to send you a message right now.
https://jmpolemic.substack.com/p/what-can-be-done
> Besides participating as suppliers, customers, employees, or targets of such orgs, there are two other key ways we engage such orgs: politics and governance.
I think this framing elides the fact that politics and governance are also org-based activities, and we interact with politics and governance primarily as employees or targets of those orgs, with donors and/or voters as analogs for the customer role.
Decisions made on an org-by-org level are what govern the course of political events. Even leaving aside the simple ones like local/county/state/national parties, there are a bunch of orgs that make up the things we in the US just lump together the national-level of the party; separately there is a large number of orgs for fundraising, for lobbying, for polling, for voter research outside of polling (ie social media and purchased market data, on account of these being different expertise). In the U.S. these are primarily non-profit private entities, though service providers like analysts, consultants, etc. work the same way as they do in the private sector (with the added wrinkle of both major parties supporting largely independent networks of these organizations). Governance questions add orgs like campaign regulators, media regulators, commissions for things like redistricting, placement of polling places, how votes are counted, etc.
It seems to me like the research on questions of regulation of the private sector should apply pretty directly in many cases.
This was the good part of LBOs
Deliciously depressing, most important insights. You've had me working on 'Meta-Governance' since reading your fab thougths this morning - and how we might introduce such Meta capability into organisations without 'top dogs' feeling threatened.
"masters don’t mind and even governance changes" - missing a word?
fixed; thanks
Sortition-based governance is the answer, imo. Any institution or organization will inevitably become dominated by those who are good at politicking rather than those who are good at fulfilling the organization's formal objectives. Stochastic selection processes dramatically reduce the expected value of politicking. I think deterministic hiring/promotion should rarely be used. With the exception of extremely specialized roles, it's better to choose at random from a pool of qualified candidates, as it weakens incentives to subvert the selection criteria.
Recent experiments with sortition-based governance have performed well. Paris and Belgium now have permanent citizens assemblies with rotating membership. Having moderated such an assembly in the US and seen the discussions up close, I can affirm that randomly selected citizens are capable enough of rapidly becoming well-informed on fairly complicated policy issues to make far better judgements than the ignorant voting public, and usually better than elected officials.
Why do you think there aren't more sortion-based small businesses?
I think it doesn't make much sense for small organizations where coordination is easier. Sortition in the sense of citizens assemblies for governance probably only makes sense at the scale of >100k people. Though I think stochastic hiring/promotion would be useful even in organizations of a few hundred or less.
That said, I think leaving decisions to chance doesn't come naturally to most people, and indeed most people even seem averse to it. I could speculate on why that might be, there are all sorts of reasons. Maybe I'll write about this.
Some might see the rise of populist movements in the US and Europe as an attempt to move toward better government.
That seems like a change in which master and task alliances win, not in how well they are held accountable.
Populism is intra-master/tasks competition, me thinks
Hence the need for our AI overlords. Complex organizations need a central processing unit for some range of tasks, coordination and conflict management.
Artificial super intelligence isn’t just a potential threat, it is also a potential savior.
How do you figure that AI masters won't act like human masters?
Being incomparably smarter, they may not be benevolent, but they will be much more capable of solving coordination problems if that is a goal.
I’m not preaching irrational optimism, just saying that AI is both a potential problem and potential solution.
Masters coordinating more seems to make it harder to hold them accountable.
The cells in our finger nails shouldn’t expect to hold our brains accountable. The animals in the San Diego Safari Park don’t hold the organization that runs it accountable.