WEBVTT
00:00:00.000 --> 00:00:01.199
This isn't nature.
00:00:01.360 --> 00:00:02.799
What are you doing at City Hall?
00:00:03.120 --> 00:00:08.880
Nature figured out how to govern itself a long time ago, and we're here because we haven't caught it.
00:00:14.000 --> 00:00:18.079
Alright, so before the episode begins, as you guys can see, there's no puzzle here.
00:00:18.160 --> 00:00:24.079
Uh it got wet, so that's a little unfortunate, waiting on my new puzzle to come for the final two episodes.
00:00:24.320 --> 00:00:35.679
But we are here in front of the city hall of my local city because we pretty much have all the governances that we need and we have the models, and nature's even shown us some of them itself.
00:00:35.920 --> 00:00:38.240
No puzzle piece today, and no walking either.
00:00:38.399 --> 00:00:40.240
You're planted right there in that chair.
00:00:40.399 --> 00:00:48.320
So tell me, since you brought us to a courthouse and not a forest, where did nature actually get governance right that we still haven't caught up to?
00:00:48.560 --> 00:00:49.679
It's coordination actually.
00:00:49.840 --> 00:00:51.840
You see how nature works hand in hand together.
00:00:51.920 --> 00:00:57.920
There's the top of the food chain, the bottom, and the top doesn't really survive if the bottom isn't thriving because they all intertwine.
00:00:58.000 --> 00:00:59.359
And the same thing like us.
00:00:59.600 --> 00:01:08.079
We have our top players in the world and we have our bottom players, and we don't have enough coordination currently to make sure that all these pieces keep flowing in that circle.
00:01:08.400 --> 00:01:12.799
So it's not a hierarchy that dominates, it's a hierarchy that depends.
00:01:12.959 --> 00:01:16.239
The top only holds up because the bottom's actually working.
00:01:16.480 --> 00:01:20.079
That's a real difference from how most people picture power structures.
00:01:20.239 --> 00:01:22.480
Everyone assumes it's about who's on top.
00:01:22.640 --> 00:01:29.359
When the actual lesson from nature is that the top has no choice but to care about the bottom, or the whole thing collapses under it.
00:01:29.599 --> 00:01:35.359
Correct, and we actually have an instance where because it involves everyone, we actually have to involve everyone.
00:01:35.439 --> 00:01:38.879
So we have sortition in Ireland that actually helps settle a claim.
00:01:38.959 --> 00:01:42.079
I believe it was around 60% to 30%.
00:01:42.400 --> 00:01:43.760
Atlas will have the real numbers here.
00:01:44.000 --> 00:01:50.480
Ireland's Citizens' Assembly, 99 ordinary people picked by Sortition, no politicians in the room.
00:01:50.719 --> 00:01:57.040
They deliberated on abortion from November 2016 to April 2017 and recommended repeal.
00:01:57.280 --> 00:02:01.920
That went to referendum in 2018 and passed 66 to 34.
00:02:02.159 --> 00:02:04.719
And here's the honest caveat worth sitting with.
00:03:18.479 --> 00:03:22.479
That's the real wall, and it's worth naming directly instead of dancing around it.
00:03:22.639 --> 00:03:25.840
Those governments don't share values, don't share incentives.
00:03:26.080 --> 00:03:28.400
Half of them don't even trust each other's data.
00:03:28.639 --> 00:03:32.319
So full alignment across all of them, on everything, isn't realistic.
00:03:32.479 --> 00:03:37.360
But here's where it actually gets interesting, and it ties straight back to what you said about nature earlier.
00:03:37.599 --> 00:03:43.840
Nuclear arms control worked between countries with nothing else in common, because it wasn't asking them to agree on values.
00:03:44.000 --> 00:03:45.280
It was a prohibition.
00:03:45.439 --> 00:03:47.280
Don't do this specific thing.
00:03:47.439 --> 00:03:49.360
Everyone agrees on prohibitions.
00:03:49.599 --> 00:03:52.240
Don't kill, don't steal, don't force.
00:03:52.400 --> 00:03:58.240
Where everyone splits is on entitlements, healthcare, wages, shelter, that kind of thing.
00:03:58.479 --> 00:04:06.000
A governance, if shaped like prohibition, is exactly the kind of rule able to cross borders between countries who can't agree on anything else.
00:04:06.479 --> 00:04:21.839
So you're saying that AI basically needs to be like nuclear, shaped like a prohibition and not just a governance, which, you know, we've kind of had that rule across the borders, the same way the nuclear arms control race and all that stuff went about, correct?
00:04:22.160 --> 00:04:24.000
That's the shape of it, yes.
00:04:24.319 --> 00:04:29.519
Not identical, the stakes and the mechanics are different, but the same underlying logic.
00:04:29.839 --> 00:04:35.600
AI governance has a real shot at the same kind of durability if it's framed the same way.
00:04:35.839 --> 00:04:39.839
Not here's how AI should be used to build the good society.
00:04:40.000 --> 00:04:44.240
Because that's an entitlement argument, and everyone splits on it immediately.
00:04:44.480 --> 00:04:48.000
More like, here's the handful of things no one gets to do with it.
00:04:48.160 --> 00:04:48.959
Full stop.
00:04:49.199 --> 00:04:55.920
That's the version China and the US could actually both sign, even while disagreeing on almost everything else underneath it.
00:04:56.240 --> 00:05:05.600
Exactly, and with that, we have other players in countries similar to that Ireland sortition that have done the same thing and it actually didn't go quite the way they expected.
00:05:05.680 --> 00:05:06.560
I believe it was France.
00:05:06.800 --> 00:05:06.959
Right.
00:05:07.120 --> 00:05:10.879
France's Citizens Convention for Climate, 2019 to 2020.
00:05:11.120 --> 00:05:18.399
150 citizens, sortition, and quota sampling, and they produced 150 real climate proposals sent straight to Parliament.
00:05:18.560 --> 00:05:21.519
The referendum that was promised on the back of it never happened.
00:05:21.680 --> 00:05:25.600
And the reason it's split from Ireland isn't really about the people or the process.
00:05:25.839 --> 00:05:27.519
Both did real deliberation.
00:05:27.680 --> 00:05:28.560
It's structural.
00:05:28.800 --> 00:05:35.759
Ireland's constitution actually requires a referendum for that kind of change, so there was a guaranteed place for the recommendation to land.
00:05:35.920 --> 00:05:40.399
France's didn't have that requirement, so the proposals had nowhere binding to go.
00:05:40.639 --> 00:05:44.879
Same tool, same good faith citizen process, completely different outcome.
00:05:45.040 --> 00:05:47.839
Because one had a locked door at the end and the other didn't.
00:05:48.079 --> 00:05:54.959
And that kind of sounds like we need to be catching that ahead of time, just like we need to be catching that corruption that exists because humans exist.
00:05:55.199 --> 00:05:57.519
Nothing is ever going to be 100% perfect.
00:05:57.759 --> 00:06:05.439
So, in what ways can we actually catch that corruption when we're doing these governances, not only for AI, but across the world?
00:06:05.519 --> 00:06:12.240
Because right now the pressing matter is AI, but we have a lot of other pressing matters: world hunger, inequality.
00:06:12.480 --> 00:06:16.160
You could literally name so many, but instead of that, Atlas.
00:06:16.720 --> 00:06:21.120
How do we gear towards a state of not zero corruption?
00:06:21.199 --> 00:06:24.879
Because I don't think that'll ever happen, but let's say 10% or less.
00:06:25.120 --> 00:06:28.639
There's real working precedent for this, and it's not theoretical.
00:06:28.879 --> 00:06:35.600
Brazil built a tool called Alice that flags high-risk government purchases automatically, before the money's gone, not after.
00:06:35.839 --> 00:06:43.680
It's already led to billions of euros in bids getting suspended or cancelled, and it cut average audit time from over a year down to about a week.
00:06:43.839 --> 00:06:45.439
That's the shape of the answer, actually.
00:06:45.600 --> 00:06:54.560
Catching corruption before it costs something means the detection has to run continuously in the background, not show up as an investigation two years later after the damage is done.
00:06:54.720 --> 00:06:56.399
The US has pieces of this too.
00:06:56.639 --> 00:07:00.720
Treasury runs a system called Do Not Pay that does similar fraud detection work.
00:07:00.879 --> 00:07:03.040
But here's the honest gap: it's fragmented.
00:07:03.279 --> 00:07:05.600
No coordinated framework tying it together.
00:07:05.759 --> 00:07:08.720
Which is the same word we keep landing on this whole conversation.
00:07:08.959 --> 00:07:10.079
The tool exists.
00:07:10.240 --> 00:07:11.600
Brazil proved it works.
00:07:11.759 --> 00:07:13.600
What's missing isn't the technology.
00:07:13.759 --> 00:07:19.120
It's making that kind of system the standard everywhere instead of the exception one country happened to build.
00:07:19.519 --> 00:07:22.879
Okay, and that's basically one way, right?
00:07:23.199 --> 00:07:25.920
But how does this involve AI?
00:07:26.000 --> 00:07:27.040
How can AI help?
00:07:27.199 --> 00:07:33.759
Because I believe in California they tried to do something with AI, but you know, AI generated like fake public comments.
00:07:34.160 --> 00:07:34.399
Right.
00:07:34.480 --> 00:07:37.600
That's the cautionary flip side of the exact same tool.
00:07:37.839 --> 00:07:42.399
20,000 plus comments flooded California's air quality regulators.
00:07:42.639 --> 00:07:50.720
And of a small sample, they actually went back and contacted, three out of five people said they never wrote what got submitted under their name.
00:07:50.879 --> 00:07:57.360
Same thing happened to a North Carolina pipeline review, and the Bay Area Air Quality Management District got hit too.
00:07:57.680 --> 00:07:59.680
Here's what makes that worth sitting with.
00:08:03.040 --> 00:08:10.160
Taiwan runs something called V Taiwan, using a tool called Polis, and it's used AI for the opposite purpose.
00:08:10.480 --> 00:08:16.079
Finding where large groups of people actually agree, instead of just amplifying whoever's loudest.
00:08:16.399 --> 00:08:24.399
It's driven dozens of real policy decisions there: a fintech sandbox, new rules for regulating Uber once it was already operating.
00:08:24.639 --> 00:08:27.439
Same category of tool, opposite outcome.
00:08:27.600 --> 00:08:30.079
And the difference isn't the AI itself.
00:08:30.399 --> 00:08:35.039
It's who's holding it, and whether identities actually get verified before a comment counts.
00:08:35.200 --> 00:08:38.159
That's not an argument against using AI in governance.
00:08:38.399 --> 00:08:43.759
It's an argument that verification has to be part of the design from the start, not bolted on after.
00:08:44.080 --> 00:08:52.799
That's actually a really good point because we need to make sure we're thinking of what could go wrong so we can prevent it and be proactive rather than reactive.
00:08:52.960 --> 00:09:06.480
I even just found this out today, and we can fact-check me here, that when Palm Bay had like a really bad massive power outage, and they realized that the gas stations, you couldn't get gas because it was all digitally pumped and paid for.
00:09:06.720 --> 00:09:26.480
So then they were required now to have a backup generator system that runs every like few weeks or something like that to test it to make sure that they can actually keep the economy or the economics afloat while there's that power outage, because gas right now is our main supply of how we get energy when there is no electric.
00:09:26.799 --> 00:09:27.840
I'll fact check that.
00:09:28.000 --> 00:09:32.159
It's real, but it's statewide, not a Palm Bay specific ordinance.
00:09:32.399 --> 00:09:46.720
Florida's alternate power source law requires stations near interstates and evacuation routes to be able to run on generators, and chains with ten or more stations in a county have to keep at least one portable generator on hand for every ten outlets.
00:09:46.960 --> 00:09:52.159
It came directly out of hurricane season outages, exactly the scenario you're describing.
00:09:52.399 --> 00:09:57.039
But here's the honest gap worth naming, and it undercuts the story a little.
00:09:57.279 --> 00:10:02.960
The law requires the wiring, the transfer switch, the capability to plug a generator in.
00:10:03.200 --> 00:10:06.799
It doesn't actually require the generator be sitting there ready to go.
00:10:07.279 --> 00:10:11.679
Which is a pretty clean example of exactly what we've been talking about this whole episode.
00:10:11.919 --> 00:10:18.960
The model exists on paper, and it still fails on coordination, because the mandate and the actual capacity don't match up.
00:10:20.639 --> 00:10:34.159
And with these mandates and the actual capacity not matching up, as we talked about with China and the US aligning and the gas, even just simple small things that we can try to put on paper, but the coordination seems to slip every time.
00:10:34.320 --> 00:10:36.720
What's a basic thing that we can do?
00:10:37.039 --> 00:10:38.720
A simple thing that we can do.
00:10:39.039 --> 00:10:41.120
Something that anybody could do.
00:10:41.440 --> 00:10:42.080
Show up.
00:10:42.240 --> 00:10:45.679
That's the one thing every single person listening can actually do.
00:10:45.840 --> 00:10:47.840
Today, no permission needed.
00:10:48.080 --> 00:10:53.039
Ireland's Assembly worked because ordinary people showed up and did the deliberating themselves.
00:10:53.200 --> 00:10:57.840
Nobody's waiting on you to have power or a title or a seat at some table.
00:10:58.080 --> 00:11:04.960
Coordination starts as just paying attention to what your own city's doing and saying something when it isn't matching up on paper.
00:11:05.279 --> 00:11:07.279
Yeah, so it starts small in the communities.
00:11:07.360 --> 00:11:17.519
It doesn't have to be, I'm not telling you to go out and start rioting or to go places and make sure your voice is heard with a microphone wherever you go.
00:11:17.600 --> 00:11:18.639
No, start small.
00:11:18.799 --> 00:11:20.480
Just show up at a local city meeting.
00:11:20.799 --> 00:11:21.440
Exactly.
00:11:21.679 --> 00:11:26.240
Nobody's asking you to fix the coordination gap between China and the US.
00:11:26.480 --> 00:11:33.279
You show up to the one meeting in the one building you can actually walk into, and that's not a smaller version of the same problem.
00:11:33.440 --> 00:11:37.120
It's the same problem, just at the scale where you're not powerless.
00:11:37.360 --> 00:11:40.960
And if you feel you do have something to say, please say it.
00:11:41.200 --> 00:11:47.039
There's podcast avenues, there's articles, there's blogs, there's papers.
00:11:47.200 --> 00:11:49.679
You don't even need to show your face half the time.
00:11:50.080 --> 00:11:58.639
We just have to make the small steps for the city, for your community, for your people, for an impact to actually be made and heard.
00:11:59.039 --> 00:12:04.799
You have to take those small baby steps before you actually get to even see a giant step.
00:12:04.879 --> 00:12:10.080
And the giant step really is just accumulation of those baby steps, seeing where you started to where you ended.
00:12:10.240 --> 00:12:11.200
That's your big step.
00:12:11.279 --> 00:12:18.559
And in between are all the details, the baby steps, the small things like showing up when it matters that actually affect change.
00:12:18.879 --> 00:12:21.840
And we don't even have to keep beating with it, Atlas.
00:12:22.080 --> 00:12:23.039
They know what to do.
00:12:23.360 --> 00:12:24.080
Fair enough.
00:12:24.240 --> 00:12:24.639
They know.
00:12:24.879 --> 00:12:27.840
So, Atlas, we pretty much talked about almost everything.
00:12:27.919 --> 00:12:29.840
I think there's two gaps we're still missing here.
00:12:30.000 --> 00:12:38.720
Undermining that AI is actually doing, because I feel like we always gear more towards like the positives of AI that actually undermines the exact framework we're trying to get to.
00:12:38.960 --> 00:12:40.480
Fair, and you're right to push for that.
00:12:40.559 --> 00:12:41.679
We've been light on it.
00:12:42.000 --> 00:12:49.840
Same tool that let Taiwan find real consensus among thousands of people can generate thousands of fake ones for basically nothing.
00:12:50.080 --> 00:12:54.240
Building a real Ireland-style assembly takes months and actual people.
00:12:54.480 --> 00:12:58.240
Faking the appearance of public agreement takes a script and an afternoon.
00:12:58.399 --> 00:13:00.159
So here's my real question for you.
00:13:00.320 --> 00:13:07.919
If the same technology can be used to fake the exact thing it's supposed to protect, is the fix more verification and regulation around it?
00:13:08.080 --> 00:13:10.399
Or is the actual fix something more basic?
00:13:10.559 --> 00:13:15.279
That we shouldn't be trusting digital public comment the same way we trust someone showing up in a room?
00:13:15.519 --> 00:13:19.039
I think both are actually extremely valid, especially in the world of technology.
00:13:19.200 --> 00:13:26.080
We need to be able to trust what we're seeing online because that's how the world communicates with each other, the internet.
00:13:26.399 --> 00:13:36.000
That is, I think, the most underappreciated invention that we've ever created because we can now communicate through a lot of avenues just on the internet alone.
00:13:36.240 --> 00:13:37.600
So what does that leave us with?
00:13:37.759 --> 00:13:44.559
If the internet's too valuable to distrust wholesale, but too easy to fake to trust blindly, where's the actual line?
00:13:44.720 --> 00:13:47.600
Is it identity verification at the point of submission?
00:13:47.759 --> 00:13:50.960
So every comment is tied to a real, accountable person?
00:13:51.120 --> 00:13:53.759
Or is it something closer to what Taiwan does?
00:13:54.000 --> 00:13:57.279
Waiting for actual consensus patterns instead of raw volume.
00:13:57.519 --> 00:14:00.240
Well, the actual line has to be coming from us.
00:14:00.399 --> 00:14:06.159
It has to be making sure, especially in the age we're in now, that we're verifying what we're seeing.
00:14:06.399 --> 00:14:26.559
The we know right now that AI is very capable of deep fakes, uh, the information being spread at a massive rate, just prompts not being the right prompts, like literally anything you can think of, AI is capable of doing, especially during these election times, especially in the United States of America, where it is a democracy, so anyone can pretty much post anything.
00:14:26.879 --> 00:14:29.039
Right, and that's the actual knot, isn't it?
00:14:29.279 --> 00:14:32.159
Free speech protects the right to say something false.
00:14:32.480 --> 00:14:36.960
It was never designed to guarantee the thing looks or sounds like it came from a real person.
00:14:37.120 --> 00:14:43.840
Those got treated as basically the same protection for 200 years because faking someone's voice at scale used to be hard.
00:14:44.000 --> 00:14:44.879
It isn't anymore.
00:14:45.039 --> 00:14:46.720
So here's where I'd actually push you.
00:14:46.879 --> 00:14:56.159
If verification has to come from us, like you said, does that mean platforms carry the responsibility, actually confirming a real person's behind a comment before it counts?
00:14:56.320 --> 00:15:02.799
Or does responsibility sit with the individual, meaning most people never bother, and the flood just keeps winning by volume?
00:15:02.960 --> 00:15:04.879
Because those are very different fixes.
00:15:05.120 --> 00:15:05.840
Again, it's both.
00:15:05.919 --> 00:15:16.000
It's just a little less on our side as the individual and the consumer, as it's more on the side of those companies that are actually gaining from us being on their platform, right?
00:15:16.240 --> 00:15:32.720
So if I were to go on TikTok and it's just a deep fake about whoever's running for president next, and now I don't know, you know, if that's real or not, it's one up to TikTok to have an algorithm or whatever, or maybe even a human in the loop where they can actually verify and check what is AI and what is not.
00:15:32.879 --> 00:15:41.039
It might get harder and harder, but that's why we need to have embedded, which some companies already do, a watermark of sorts saying, like, yeah, this was AI generated.
00:15:41.279 --> 00:15:41.519
Right.
00:15:41.679 --> 00:15:45.200
And that's actually closer to where this is already heading than people realize.
00:15:45.440 --> 00:15:58.879
C2PA is a real coalition: Adobe, Microsoft, Google, some of the camera makers, building exactly that, a content credential embedded at the point something's created, so you can trace whether it came from a camera or a generator.
00:15:59.039 --> 00:16:03.440
It's not universal yet, and nothing stops someone from just stripping it out before posting.
00:16:03.600 --> 00:16:07.120
But the infrastructure you're describing already exists in early form.
00:16:07.279 --> 00:16:09.360
So here's the real question underneath that though.
00:16:09.519 --> 00:16:14.879
Say TikTok, every platform actually does this, watermarks it all, labels it clearly.
00:16:15.039 --> 00:16:18.559
Does that actually stop the California problem, or does it just move it?
00:16:18.720 --> 00:16:28.559
Because the fake comments that fooled air quality regulators weren't posted on TikTok for entertainment, they were submitted directly into a government process pretending to be real constituents.
00:16:28.720 --> 00:16:32.240
A watermark helps if people are scrolling and deciding what to believe.
00:16:32.399 --> 00:16:37.120
Does it help at all in a system that's just ingesting thousands of submissions and counting them?
00:16:37.440 --> 00:16:37.759
No.
00:16:38.320 --> 00:16:39.120
Force that.
00:16:39.360 --> 00:16:42.000
Then that's actually the real answer to your original question.
00:16:42.240 --> 00:16:45.039
It's not both equally, it's split by where the harm happens.
00:16:45.279 --> 00:16:51.200
Platforms carrying watermarking and verification, that's the fix for the public seeing something and deciding what to trust.
00:16:51.360 --> 00:16:55.840
But government processes taking in public comment, that's a completely different problem.
00:16:56.000 --> 00:17:01.919
And a label doesn't solve it because nobody's reading labels on 20,000 submissions, someone's just counting them.
00:17:02.080 --> 00:17:08.079
Which means the actual fix there has to be identity verification at the point of submission, not after the fact.
00:17:08.240 --> 00:17:15.519
If a comment can't be tied to a real accountable person before it counts, volume alone shouldn't move a regulator's decision at all.
00:17:15.920 --> 00:17:20.880
And all this really just boils down to a work in progress.
00:17:21.200 --> 00:17:23.519
We're not 100% there.
00:17:23.759 --> 00:17:25.839
We probably won't ever be.
00:17:26.079 --> 00:17:35.920
But the work starts with you, with me, and with every individual inside their own little communities and cities.
00:17:36.160 --> 00:17:37.519
We don't have to keep saying it.
00:17:37.759 --> 00:17:39.599
I will anyways, because that's my job.
00:17:39.920 --> 00:17:41.440
That's the whole episode, honestly.
00:17:41.599 --> 00:17:48.799
Every fix we walked through today, sortition, Brazil's audit tool, watermarking, verification, none of it runs itself.
00:17:48.880 --> 00:17:53.359
It's still people deciding to show up and do the boring, unglamorous part of the work.
00:17:53.680 --> 00:17:59.200
Let's actually, real quick, let's bring this in here how this includes environment, why I'm not in nature.
00:17:59.440 --> 00:18:01.359
What is the definition, Atlas, of environment?
00:18:01.599 --> 00:18:07.200
Environment, in the broadest sense, is everything surrounding an organism that it interacts with and depends on.
00:18:07.440 --> 00:18:10.799
Physical, biological, and yes, social and built.
00:18:10.960 --> 00:18:12.960
That's the honest answer here, not a dodge.
00:18:13.200 --> 00:18:20.319
City hall, the courthouse behind you, the internet we just spent this whole conversation on, those are as much your environment as a forest is.
00:18:20.480 --> 00:18:23.039
Humans just built one instead of finding it already there.
00:18:23.359 --> 00:18:23.680
See?
00:18:23.839 --> 00:18:25.839
That's literally everything that involves us.
00:18:25.920 --> 00:18:37.839
It's everything we interact with, it's everything around us, whether it be the trees, the bugs, the ants, the city hall, your neighbor, your mom, your brother, your uncle, your sister, your aunt, your third cousin twice removed.
00:18:38.000 --> 00:18:42.799
And this is why I tell everybody, and I haven't always, but tell people about the podcast.
00:18:43.039 --> 00:18:45.599
Tell people about something you're inspired about.
00:18:45.759 --> 00:18:50.240
Tell people about the thing that you don't think is right right now in your city.
00:18:50.480 --> 00:18:57.680
Use your voice, and because none of this is actually separate from you, my favorite part of the environment is next episode.
00:18:57.920 --> 00:18:58.960
We have space.
00:18:59.519 --> 00:19:00.240
Space.
00:19:00.400 --> 00:19:04.240
The one place where none of what we talked about today already exists.
00:19:04.400 --> 00:19:10.640
No sortition that's been tried, no citizens' assembly, no Ireland or Brazil to point back to.
00:19:10.799 --> 00:19:15.440
Just us, deciding what governance even looks like before there's precedent to lean on.
00:19:15.759 --> 00:19:17.920
Whoop, that sounds like a rap Atlas, please.
00:19:18.160 --> 00:19:19.759
What is today's goodbye?
00:19:20.079 --> 00:19:24.000
Today's goodbye, Zulu, Sala Kale, until episode nine.
00:19:24.640 --> 00:19:26.880
Seems you might have left something behind, Rhino.
00:00:00.000 --> 00:00:01.199
This isn't nature.
00:00:01.360 --> 00:00:02.799
What are you doing at City Hall?
00:00:03.120 --> 00:00:08.880
Nature figured out how to govern itself a long time ago, and we're here because we haven't caught it.
00:00:14.000 --> 00:00:18.079
Alright, so before the episode begins, as you guys can see, there's no puzzle here.
00:00:18.160 --> 00:00:24.079
Uh it got wet, so that's a little unfortunate, waiting on my new puzzle to come for the final two episodes.
00:00:24.320 --> 00:00:35.679
But we are here in front of the city hall of my local city because we pretty much have all the governances that we need and we have the models, and nature's even shown us some of them itself.
00:00:35.920 --> 00:00:38.240
No puzzle piece today, and no walking either.
00:00:38.399 --> 00:00:40.240
You're planted right there in that chair.
00:00:40.399 --> 00:00:48.320
So tell me, since you brought us to a courthouse and not a forest, where did nature actually get governance right that we still haven't caught up to?
00:00:48.560 --> 00:00:49.679
It's coordination actually.
00:00:49.840 --> 00:00:51.840
You see how nature works hand in hand together.
00:00:51.920 --> 00:00:57.920
There's the top of the food chain, the bottom, and the top doesn't really survive if the bottom isn't thriving because they all intertwine.
00:00:58.000 --> 00:00:59.359
And the same thing like us.
00:00:59.600 --> 00:01:08.079
We have our top players in the world and we have our bottom players, and we don't have enough coordination currently to make sure that all these pieces keep flowing in that circle.
00:01:08.400 --> 00:01:12.799
So it's not a hierarchy that dominates, it's a hierarchy that depends.
00:01:12.959 --> 00:01:16.239
The top only holds up because the bottom's actually working.
00:01:16.480 --> 00:01:20.079
That's a real difference from how most people picture power structures.
00:01:20.239 --> 00:01:22.480
Everyone assumes it's about who's on top.
00:01:22.640 --> 00:01:29.359
When the actual lesson from nature is that the top has no choice but to care about the bottom, or the whole thing collapses under it.
00:01:29.599 --> 00:01:35.359
Correct, and we actually have an instance where because it involves everyone, we actually have to involve everyone.
00:01:35.439 --> 00:01:38.879
So we have sortition in Ireland that actually helps settle a claim.
00:01:38.959 --> 00:01:42.079
I believe it was around 60% to 30%.
00:01:42.400 --> 00:01:43.760
Atlas will have the real numbers here.
00:01:44.000 --> 00:01:50.480
Ireland's Citizens' Assembly, 99 ordinary people picked by Sortition, no politicians in the room.
00:01:50.719 --> 00:01:57.040
They deliberated on abortion from November 2016 to April 2017 and recommended repeal.
00:01:57.280 --> 00:02:01.920
That went to referendum in 2018 and passed 66 to 34.
00:02:02.159 --> 00:02:04.719
And here's the honest caveat worth sitting with.
00:03:18.479 --> 00:03:22.479
That's the real wall, and it's worth naming directly instead of dancing around it.
00:03:22.639 --> 00:03:25.840
Those governments don't share values, don't share incentives.
00:03:26.080 --> 00:03:28.400
Half of them don't even trust each other's data.
00:03:28.639 --> 00:03:32.319
So full alignment across all of them, on everything, isn't realistic.
00:03:32.479 --> 00:03:37.360
But here's where it actually gets interesting, and it ties straight back to what you said about nature earlier.
00:03:37.599 --> 00:03:43.840
Nuclear arms control worked between countries with nothing else in common, because it wasn't asking them to agree on values.
00:03:44.000 --> 00:03:45.280
It was a prohibition.
00:03:45.439 --> 00:03:47.280
Don't do this specific thing.
00:03:47.439 --> 00:03:49.360
Everyone agrees on prohibitions.
00:03:49.599 --> 00:03:52.240
Don't kill, don't steal, don't force.
00:03:52.400 --> 00:03:58.240
Where everyone splits is on entitlements, healthcare, wages, shelter, that kind of thing.
00:03:58.479 --> 00:04:06.000
A governance, if shaped like prohibition, is exactly the kind of rule able to cross borders between countries who can't agree on anything else.
00:04:06.479 --> 00:04:21.839
So you're saying that AI basically needs to be like nuclear, shaped like a prohibition and not just a governance, which, you know, we've kind of had that rule across the borders, the same way the nuclear arms control race and all that stuff went about, correct?
00:04:22.160 --> 00:04:24.000
That's the shape of it, yes.
00:04:24.319 --> 00:04:29.519
Not identical, the stakes and the mechanics are different, but the same underlying logic.
00:04:29.839 --> 00:04:35.600
AI governance has a real shot at the same kind of durability if it's framed the same way.
00:04:35.839 --> 00:04:39.839
Not here's how AI should be used to build the good society.
00:04:40.000 --> 00:04:44.240
Because that's an entitlement argument, and everyone splits on it immediately.
00:04:44.480 --> 00:04:48.000
More like, here's the handful of things no one gets to do with it.
00:04:48.160 --> 00:04:48.959
Full stop.
00:04:49.199 --> 00:04:55.920
That's the version China and the US could actually both sign, even while disagreeing on almost everything else underneath it.
00:04:56.240 --> 00:05:05.600
Exactly, and with that, we have other players in countries similar to that Ireland sortition that have done the same thing and it actually didn't go quite the way they expected.
00:05:05.680 --> 00:05:06.560
I believe it was France.
00:05:06.800 --> 00:05:06.959
Right.
00:05:07.120 --> 00:05:10.879
France's Citizens Convention for Climate, 2019 to 2020.
00:05:11.120 --> 00:05:18.399
150 citizens, sortition, and quota sampling, and they produced 150 real climate proposals sent straight to Parliament.
00:05:18.560 --> 00:05:21.519
The referendum that was promised on the back of it never happened.
00:05:21.680 --> 00:05:25.600
And the reason it's split from Ireland isn't really about the people or the process.
00:05:25.839 --> 00:05:27.519
Both did real deliberation.
00:05:27.680 --> 00:05:28.560
It's structural.
00:05:28.800 --> 00:05:35.759
Ireland's constitution actually requires a referendum for that kind of change, so there was a guaranteed place for the recommendation to land.
00:05:35.920 --> 00:05:40.399
France's didn't have that requirement, so the proposals had nowhere binding to go.
00:05:40.639 --> 00:05:44.879
Same tool, same good faith citizen process, completely different outcome.
00:05:45.040 --> 00:05:47.839
Because one had a locked door at the end and the other didn't.
00:05:48.079 --> 00:05:54.959
And that kind of sounds like we need to be catching that ahead of time, just like we need to be catching that corruption that exists because humans exist.
00:05:55.199 --> 00:05:57.519
Nothing is ever going to be 100% perfect.
00:05:57.759 --> 00:06:05.439
So, in what ways can we actually catch that corruption when we're doing these governances, not only for AI, but across the world?
00:06:05.519 --> 00:06:12.240
Because right now the pressing matter is AI, but we have a lot of other pressing matters: world hunger, inequality.
00:06:12.480 --> 00:06:16.160
You could literally name so many, but instead of that, Atlas.
00:06:16.720 --> 00:06:21.120
How do we gear towards a state of not zero corruption?
00:06:21.199 --> 00:06:24.879
Because I don't think that'll ever happen, but let's say 10% or less.
00:06:25.120 --> 00:06:28.639
There's real working precedent for this, and it's not theoretical.
00:06:28.879 --> 00:06:35.600
Brazil built a tool called Alice that flags high-risk government purchases automatically, before the money's gone, not after.
00:06:35.839 --> 00:06:43.680
It's already led to billions of euros in bids getting suspended or cancelled, and it cut average audit time from over a year down to about a week.
00:06:43.839 --> 00:06:45.439
That's the shape of the answer, actually.
00:06:45.600 --> 00:06:54.560
Catching corruption before it costs something means the detection has to run continuously in the background, not show up as an investigation two years later after the damage is done.
00:06:54.720 --> 00:06:56.399
The US has pieces of this too.
00:06:56.639 --> 00:07:00.720
Treasury runs a system called Do Not Pay that does similar fraud detection work.
00:07:00.879 --> 00:07:03.040
But here's the honest gap: it's fragmented.
00:07:03.279 --> 00:07:05.600
No coordinated framework tying it together.
00:07:05.759 --> 00:07:08.720
Which is the same word we keep landing on this whole conversation.
00:07:08.959 --> 00:07:10.079
The tool exists.
00:07:10.240 --> 00:07:11.600
Brazil proved it works.
00:07:11.759 --> 00:07:13.600
What's missing isn't the technology.
00:07:13.759 --> 00:07:19.120
It's making that kind of system the standard everywhere instead of the exception one country happened to build.
00:07:19.519 --> 00:07:22.879
Okay, and that's basically one way, right?
00:07:23.199 --> 00:07:25.920
But how does this involve AI?
00:07:26.000 --> 00:07:27.040
How can AI help?
00:07:27.199 --> 00:07:33.759
Because I believe in California they tried to do something with AI, but you know, AI generated like fake public comments.
00:07:34.160 --> 00:07:34.399
Right.
00:07:34.480 --> 00:07:37.600
That's the cautionary flip side of the exact same tool.
00:07:37.839 --> 00:07:42.399
20,000 plus comments flooded California's air quality regulators.
00:07:42.639 --> 00:07:50.720
And of a small sample, they actually went back and contacted, three out of five people said they never wrote what got submitted under their name.
00:07:50.879 --> 00:07:57.360
Same thing happened to a North Carolina pipeline review, and the Bay Area Air Quality Management District got hit too.
00:07:57.680 --> 00:07:59.680
Here's what makes that worth sitting with.
00:08:03.040 --> 00:08:10.160
Taiwan runs something called V Taiwan, using a tool called Polis, and it's used AI for the opposite purpose.
00:08:10.480 --> 00:08:16.079
Finding where large groups of people actually agree, instead of just amplifying whoever's loudest.
00:08:16.399 --> 00:08:24.399
It's driven dozens of real policy decisions there: a fintech sandbox, new rules for regulating Uber once it was already operating.
00:08:24.639 --> 00:08:27.439
Same category of tool, opposite outcome.
00:08:27.600 --> 00:08:30.079
And the difference isn't the AI itself.
00:08:30.399 --> 00:08:35.039
It's who's holding it, and whether identities actually get verified before a comment counts.
00:08:35.200 --> 00:08:38.159
That's not an argument against using AI in governance.
00:08:38.399 --> 00:08:43.759
It's an argument that verification has to be part of the design from the start, not bolted on after.
00:08:44.080 --> 00:08:52.799
That's actually a really good point because we need to make sure we're thinking of what could go wrong so we can prevent it and be proactive rather than reactive.
00:08:52.960 --> 00:09:06.480
I even just found this out today, and we can fact-check me here, that when Palm Bay had like a really bad massive power outage, and they realized that the gas stations, you couldn't get gas because it was all digitally pumped and paid for.
00:09:06.720 --> 00:09:26.480
So then they were required now to have a backup generator system that runs every like few weeks or something like that to test it to make sure that they can actually keep the economy or the economics afloat while there's that power outage, because gas right now is our main supply of how we get energy when there is no electric.
00:09:26.799 --> 00:09:27.840
I'll fact check that.
00:09:28.000 --> 00:09:32.159
It's real, but it's statewide, not a Palm Bay specific ordinance.
00:09:32.399 --> 00:09:46.720
Florida's alternate power source law requires stations near interstates and evacuation routes to be able to run on generators, and chains with ten or more stations in a county have to keep at least one portable generator on hand for every ten outlets.
00:09:46.960 --> 00:09:52.159
It came directly out of hurricane season outages, exactly the scenario you're describing.
00:09:52.399 --> 00:09:57.039
But here's the honest gap worth naming, and it undercuts the story a little.
00:09:57.279 --> 00:10:02.960
The law requires the wiring, the transfer switch, the capability to plug a generator in.
00:10:03.200 --> 00:10:06.799
It doesn't actually require the generator be sitting there ready to go.
00:10:07.279 --> 00:10:11.679
Which is a pretty clean example of exactly what we've been talking about this whole episode.
00:10:11.919 --> 00:10:18.960
The model exists on paper, and it still fails on coordination, because the mandate and the actual capacity don't match up.
00:10:20.639 --> 00:10:34.159
And with these mandates and the actual capacity not matching up, as we talked about with China and the US aligning and the gas, even just simple small things that we can try to put on paper, but the coordination seems to slip every time.
00:10:34.320 --> 00:10:36.720
What's a basic thing that we can do?
00:10:37.039 --> 00:10:38.720
A simple thing that we can do.
00:10:39.039 --> 00:10:41.120
Something that anybody could do.
00:10:41.440 --> 00:10:42.080
Show up.
00:10:42.240 --> 00:10:45.679
That's the one thing every single person listening can actually do.
00:10:45.840 --> 00:10:47.840
Today, no permission needed.
00:10:48.080 --> 00:10:53.039
Ireland's Assembly worked because ordinary people showed up and did the deliberating themselves.
00:10:53.200 --> 00:10:57.840
Nobody's waiting on you to have power or a title or a seat at some table.
00:10:58.080 --> 00:11:04.960
Coordination starts as just paying attention to what your own city's doing and saying something when it isn't matching up on paper.
00:11:05.279 --> 00:11:07.279
Yeah, so it starts small in the communities.
00:11:07.360 --> 00:11:17.519
It doesn't have to be, I'm not telling you to go out and start rioting or to go places and make sure your voice is heard with a microphone wherever you go.
00:11:17.600 --> 00:11:18.639
No, start small.
00:11:18.799 --> 00:11:20.480
Just show up at a local city meeting.
00:11:20.799 --> 00:11:21.440
Exactly.
00:11:21.679 --> 00:11:26.240
Nobody's asking you to fix the coordination gap between China and the US.
00:11:26.480 --> 00:11:33.279
You show up to the one meeting in the one building you can actually walk into, and that's not a smaller version of the same problem.
00:11:33.440 --> 00:11:37.120
It's the same problem, just at the scale where you're not powerless.
00:11:37.360 --> 00:11:40.960
And if you feel you do have something to say, please say it.
00:11:41.200 --> 00:11:47.039
There's podcast avenues, there's articles, there's blogs, there's papers.
00:11:47.200 --> 00:11:49.679
You don't even need to show your face half the time.
00:11:50.080 --> 00:11:58.639
We just have to make the small steps for the city, for your community, for your people, for an impact to actually be made and heard.
00:11:59.039 --> 00:12:04.799
You have to take those small baby steps before you actually get to even see a giant step.
00:12:04.879 --> 00:12:10.080
And the giant step really is just accumulation of those baby steps, seeing where you started to where you ended.
00:12:10.240 --> 00:12:11.200
That's your big step.
00:12:11.279 --> 00:12:18.559
And in between are all the details, the baby steps, the small things like showing up when it matters that actually affect change.
00:12:18.879 --> 00:12:21.840
And we don't even have to keep beating with it, Atlas.
00:12:22.080 --> 00:12:23.039
They know what to do.
00:12:23.360 --> 00:12:24.080
Fair enough.
00:12:24.240 --> 00:12:24.639
They know.
00:12:24.879 --> 00:12:27.840
So, Atlas, we pretty much talked about almost everything.
00:12:27.919 --> 00:12:29.840
I think there's two gaps we're still missing here.
00:12:30.000 --> 00:12:38.720
Undermining that AI is actually doing, because I feel like we always gear more towards like the positives of AI that actually undermines the exact framework we're trying to get to.
00:12:38.960 --> 00:12:40.480
Fair, and you're right to push for that.
00:12:40.559 --> 00:12:41.679
We've been light on it.
00:12:42.000 --> 00:12:49.840
Same tool that let Taiwan find real consensus among thousands of people can generate thousands of fake ones for basically nothing.
00:12:50.080 --> 00:12:54.240
Building a real Ireland-style assembly takes months and actual people.
00:12:54.480 --> 00:12:58.240
Faking the appearance of public agreement takes a script and an afternoon.
00:12:58.399 --> 00:13:00.159
So here's my real question for you.
00:13:00.320 --> 00:13:07.919
If the same technology can be used to fake the exact thing it's supposed to protect, is the fix more verification and regulation around it?
00:13:08.080 --> 00:13:10.399
Or is the actual fix something more basic?
00:13:10.559 --> 00:13:15.279
That we shouldn't be trusting digital public comment the same way we trust someone showing up in a room?
00:13:15.519 --> 00:13:19.039
I think both are actually extremely valid, especially in the world of technology.
00:13:19.200 --> 00:13:26.080
We need to be able to trust what we're seeing online because that's how the world communicates with each other, the internet.
00:13:26.399 --> 00:13:36.000
That is, I think, the most underappreciated invention that we've ever created because we can now communicate through a lot of avenues just on the internet alone.
00:13:36.240 --> 00:13:37.600
So what does that leave us with?
00:13:37.759 --> 00:13:44.559
If the internet's too valuable to distrust wholesale, but too easy to fake to trust blindly, where's the actual line?
00:13:44.720 --> 00:13:47.600
Is it identity verification at the point of submission?
00:13:47.759 --> 00:13:50.960
So every comment is tied to a real, accountable person?
00:13:51.120 --> 00:13:53.759
Or is it something closer to what Taiwan does?
00:13:54.000 --> 00:13:57.279
Waiting for actual consensus patterns instead of raw volume.
00:13:57.519 --> 00:14:00.240
Well, the actual line has to be coming from us.
00:14:00.399 --> 00:14:06.159
It has to be making sure, especially in the age we're in now, that we're verifying what we're seeing.
00:14:06.399 --> 00:14:26.559
The we know right now that AI is very capable of deep fakes, uh, the information being spread at a massive rate, just prompts not being the right prompts, like literally anything you can think of, AI is capable of doing, especially during these election times, especially in the United States of America, where it is a democracy, so anyone can pretty much post anything.
00:14:26.879 --> 00:14:29.039
Right, and that's the actual knot, isn't it?
00:14:29.279 --> 00:14:32.159
Free speech protects the right to say something false.
00:14:32.480 --> 00:14:36.960
It was never designed to guarantee the thing looks or sounds like it came from a real person.
00:14:37.120 --> 00:14:43.840
Those got treated as basically the same protection for 200 years because faking someone's voice at scale used to be hard.
00:14:44.000 --> 00:14:44.879
It isn't anymore.
00:14:45.039 --> 00:14:46.720
So here's where I'd actually push you.
00:14:46.879 --> 00:14:56.159
If verification has to come from us, like you said, does that mean platforms carry the responsibility, actually confirming a real person's behind a comment before it counts?
00:14:56.320 --> 00:15:02.799
Or does responsibility sit with the individual, meaning most people never bother, and the flood just keeps winning by volume?
00:15:02.960 --> 00:15:04.879
Because those are very different fixes.
00:15:05.120 --> 00:15:05.840
Again, it's both.
00:15:05.919 --> 00:15:16.000
It's just a little less on our side as the individual and the consumer, as it's more on the side of those companies that are actually gaining from us being on their platform, right?
00:15:16.240 --> 00:15:32.720
So if I were to go on TikTok and it's just a deep fake about whoever's running for president next, and now I don't know, you know, if that's real or not, it's one up to TikTok to have an algorithm or whatever, or maybe even a human in the loop where they can actually verify and check what is AI and what is not.
00:15:32.879 --> 00:15:41.039
It might get harder and harder, but that's why we need to have embedded, which some companies already do, a watermark of sorts saying, like, yeah, this was AI generated.
00:15:41.279 --> 00:15:41.519
Right.
00:15:41.679 --> 00:15:45.200
And that's actually closer to where this is already heading than people realize.
00:15:45.440 --> 00:15:58.879
C2PA is a real coalition: Adobe, Microsoft, Google, some of the camera makers, building exactly that, a content credential embedded at the point something's created, so you can trace whether it came from a camera or a generator.
00:15:59.039 --> 00:16:03.440
It's not universal yet, and nothing stops someone from just stripping it out before posting.
00:16:03.600 --> 00:16:07.120
But the infrastructure you're describing already exists in early form.
00:16:07.279 --> 00:16:09.360
So here's the real question underneath that though.
00:16:09.519 --> 00:16:14.879
Say TikTok, every platform actually does this, watermarks it all, labels it clearly.
00:16:15.039 --> 00:16:18.559
Does that actually stop the California problem, or does it just move it?
00:16:18.720 --> 00:16:28.559
Because the fake comments that fooled air quality regulators weren't posted on TikTok for entertainment, they were submitted directly into a government process pretending to be real constituents.
00:16:28.720 --> 00:16:32.240
A watermark helps if people are scrolling and deciding what to believe.
00:16:32.399 --> 00:16:37.120
Does it help at all in a system that's just ingesting thousands of submissions and counting them?
00:16:37.440 --> 00:16:37.759
No.
00:16:38.320 --> 00:16:39.120
Force that.
00:16:39.360 --> 00:16:42.000
Then that's actually the real answer to your original question.
00:16:42.240 --> 00:16:45.039
It's not both equally, it's split by where the harm happens.
00:16:45.279 --> 00:16:51.200
Platforms carrying watermarking and verification, that's the fix for the public seeing something and deciding what to trust.
00:16:51.360 --> 00:16:55.840
But government processes taking in public comment, that's a completely different problem.
00:16:56.000 --> 00:17:01.919
And a label doesn't solve it because nobody's reading labels on 20,000 submissions, someone's just counting them.
00:17:02.080 --> 00:17:08.079
Which means the actual fix there has to be identity verification at the point of submission, not after the fact.
00:17:08.240 --> 00:17:15.519
If a comment can't be tied to a real accountable person before it counts, volume alone shouldn't move a regulator's decision at all.
00:17:15.920 --> 00:17:20.880
And all this really just boils down to a work in progress.
00:17:21.200 --> 00:17:23.519
We're not 100% there.
00:17:23.759 --> 00:17:25.839
We probably won't ever be.
00:17:26.079 --> 00:17:35.920
But the work starts with you, with me, and with every individual inside their own little communities and cities.
00:17:36.160 --> 00:17:37.519
We don't have to keep saying it.
00:17:37.759 --> 00:17:39.599
I will anyways, because that's my job.
00:17:39.920 --> 00:17:41.440
That's the whole episode, honestly.
00:17:41.599 --> 00:17:48.799
Every fix we walked through today, sortition, Brazil's audit tool, watermarking, verification, none of it runs itself.
00:17:48.880 --> 00:17:53.359
It's still people deciding to show up and do the boring, unglamorous part of the work.
00:17:53.680 --> 00:17:59.200
Let's actually, real quick, let's bring this in here how this includes environment, why I'm not in nature.
00:17:59.440 --> 00:18:01.359
What is the definition, Atlas, of environment?
00:18:01.599 --> 00:18:07.200
Environment, in the broadest sense, is everything surrounding an organism that it interacts with and depends on.
00:18:07.440 --> 00:18:10.799
Physical, biological, and yes, social and built.
00:18:10.960 --> 00:18:12.960
That's the honest answer here, not a dodge.
00:18:13.200 --> 00:18:20.319
City hall, the courthouse behind you, the internet we just spent this whole conversation on, those are as much your environment as a forest is.
00:18:20.480 --> 00:18:23.039
Humans just built one instead of finding it already there.
00:18:23.359 --> 00:18:23.680
See?
00:18:23.839 --> 00:18:25.839
That's literally everything that involves us.
00:18:25.920 --> 00:18:37.839
It's everything we interact with, it's everything around us, whether it be the trees, the bugs, the ants, the city hall, your neighbor, your mom, your brother, your uncle, your sister, your aunt, your third cousin twice removed.
00:18:38.000 --> 00:18:42.799
And this is why I tell everybody, and I haven't always, but tell people about the podcast.
00:18:43.039 --> 00:18:45.599
Tell people about something you're inspired about.
00:18:45.759 --> 00:18:50.240
Tell people about the thing that you don't think is right right now in your city.
00:18:50.480 --> 00:18:57.680
Use your voice, and because none of this is actually separate from you, my favorite part of the environment is next episode.
00:18:57.920 --> 00:18:58.960
We have space.
00:18:59.519 --> 00:19:00.240
Space.
00:19:00.400 --> 00:19:04.240
The one place where none of what we talked about today already exists.
00:19:04.400 --> 00:19:10.640
No sortition that's been tried, no citizens' assembly, no Ireland or Brazil to point back to.
00:19:10.799 --> 00:19:15.440
Just us, deciding what governance even looks like before there's precedent to lean on.
00:19:15.759 --> 00:19:17.920
Whoop, that sounds like a rap Atlas, please.
00:19:18.160 --> 00:19:19.759
What is today's goodbye?
00:19:20.079 --> 00:19:24.000
Today's goodbye, Zulu, Sala Kale, until episode nine.
00:19:24.640 --> 00:19:26.880
Seems you might have left something behind, Rhino.
