George Holliday, the guy who filmed Rodney King in 91, had a handycam which he described as ‘like a size 13 shoe’. He auctioned it off last year.

In 68, the day after he said he wouldn’t run again, LBJ told the National Association of Broadcasters that TV had lost the war in Vietnam. The network news had to fly the film from Vietnam to a lab in Tokyo and then to the US to air tomorrow. But it worked – it opened a ‘credibility gap’ between official statements and brutal reality. ‘If we’ve lost Cronkite, we’ve lost the nation’, LBJ may have said.

The 60s war reporters probably used handheld 16mm cameras like this one.

In 2020 Cronkite was a samsung galaxy and the Tet Offensive was your street. Next year, the camera will be wearable. It will be mounted on your face. And it won’t just be challenging the legitimacy of central institutions, it will be solving and preventing crimes. (Cameras have been solving crimes since at least 18591, but what’s coming will be different).

The trend here is the decentralization of public safety. This hardware is coming on the market, like, now – the facebook’s ray ban partnership, xiaomi, something coming from apple. Software and networks will catch up in a year or two. Expect a wild ride.

(If you read with a soundtrack: pick any song with America in the title. I recommend Simon + Garfunkel or Childish Gambino, both of which are about control of space. The Nanci Griffith song about backseat drivers also works).

  1. Guided by voices
  2. Precrime is the wrong question
  3. Wearable HOA
  4. Private armies though
  5. A millimort for a millimort
  6. State capacity and its discontents

Guided by voices

When the pea whistle was invented in 1883, the first client was Scotland Yard. Unlike victims, when law enforcement speaks about crime it’s with a very loud voice. They decide which reports are credible, what bubbles up to databases, how information is summarized.

Wearables change that by decentralizing the collection of victim and witness reports. The dataset assembled by the new glassholes will be:

  • Collectively owned, with ownership weighted to the victims. I know 2 new yorkers with rough subway experiences who found out that those cameras are basically unusable by cops
  • Collectively produced: anything that happens in public will have multiple bystanders sharing angles of it, as well as their testimony about intent and fault
  • Realtime, such that a bulletin can spread to every headset in a geo radius; if someone is touching butts, the next victim can get a warning and flag that person’s face
  • Tagged quickly: the ontology of crime today depends on the court system, which is a multi-step process that can take months. The new data will be categorized instantly by some combination of witnesses, volunteers, and automation
  • Not dependent on a judicial process: individuals will make probabilistic assessments of whether something is a crime or not, using a combination of evidence and case history data

I’m not, you’ll notice, making any claims about the impact of this technology on crime rates, close rates, anything. I read a bunch of PDFs about the impact of policing on crime rates and came away understanding very little. Maybe the data will be wrong. Maybe it will stir up awful mobs.

For all the risks, decentralized safety networks will also empower people to accurately report how their lives are, to be heard by others around them, and to form groups to solve real problems.

I always associated Gandhi with his nonviolent resistance work, but in researching this post, I learned protest was only half his project. ‘My real politics,’ he wrote to a colleague, ‘is constructive work’ – the creation of new institutions that advanced his social values. Decentralization, whatever else it represents, is a factory for new institutions.

Gandhi’s root philosophy, Satyagraha, means ‘firmly grasping the truth’, and includes the idea that structural unfairness can be opposed with true information. It’s a synonym for Foucault’s parrhesia, which means radical truth-telling. It’s also, if we’re lucky, a synonym for collectively owned safety data.

Precrime is the wrong question

Public safety technology has a bad rap because for a while now the equation has been technology + police = precrime. Precrime, in case you missed both the 70s and the 2000s, is when you arrest or otherwise investigate someone before they commit a crime, often using gamey population-level stats.

In fiction it was invented by Philip K Dick, the author whose heroes spend most of their time lamenting their messy apartments and taking antipsychotics so they can ignore the physical dissolution of their world. I wonder sometimes whether the IRL precrime boosters are PKD fans or PKD characters.

Your standard precrime system starts with a person’s criminal record, then adds some random other data, and then releases a trained model. In 2018 researchers beat one using a simple linear algorithm on age and criminal record. The vendor claimed that as an independent verification. It helps to have no shame.2

Predicting if a crime will happen is problematic. Deciding whether a crime has happened, on the other hand, is less difficult and less fraught, especially if there are multiple witnesses, especially if you have neutral humans physically present and tagging the data, especially if you have video from multiple angles and time scales.

Pervasive wearables don’t enable precrime, which has never worked. They enable something like ‘wearable jury’, a system for bystanders to agree that something is wrong, to have reasonable certainty that the situation is as it seems, and to report it or act to defend people from harm.

This isn’t all gravy. You need to be sure it doesn’t turn into a witch pursuit thing. More than one social platform has fired up manhunts for the wrong person. The due process and privacy concerns here are real. Mobbing is real. At scale, the UN is blaming facebook for being the media platform for the Rohingya genocide. We can’t build more of these platforms without advancing the tools for moderation.

But empowered bystanders can do a lot. I’ve seen impromptu groups gently break up fights on that scary long stretch of the L between 1st Ave and Bedford. Bystanders can be amazing when they have certainty about the situation.

Wearable HOA

The 2011 occupy protests in nyc mostly ended when the owner of Zuccotti Park shut it down for cleaning. It would have been hard to remove protesters from a public park, but Zuccotti is a ‘POP’, a privately owned public plaza maintained by developers in exchange for zoning waivers. In a sense it’s a suburb, a public space with a low density of ownership that gets to set its own rules of decorum.

This post is about public safety, but we don’t always agree about what is ‘public’ and what is ‘safe’. People were swapping BJs for narcotics in Washington Square Park last month and apparently that was either too public, or not safe enough, because the city set a curfew and cleared the park at 10pm.

The NYPD also ran a block meeting in a local church basement to take the temperature of the neighbors. Gothamist got a quote from the 6th precinct community council president saying basically ‘we’re open minded people but we also go to sleep on time’, which is (1) totally fine and (2) something you say in a suburb.

Decentralized safety leads naturally to tagging, as witnesses and victims categorize what they experienced or saw. It’s a small jump from tagging to talking, to real or virtual church basements where they decide that a problem requires action.

If you put together 250 people who have seen or experienced a similar bad event in a narrow geo radius, you have a pretty loud constituency. These groups take hard work to organize today, but in a decentralized safety future they’ll emerge organically from the ether.

Maybe they’re locals, maybe they’re commuters, maybe they don’t own the street, but they own the problem. And ownership + voice = suburb. A suburb, remember, is a place where fewer people make more rules.

Private armies though

That of course leaves the question of how these decisions are enforced. If by force, is it a public army or a private one. If by other means, which? Uber’s star system is an example of a non-violent norm system that uses social credit. If you suck, you get reduced access to the service, lightweight banishment.

The ‘public army’ case you know. Zuccotti + Washington Square were cleared by police. BIDs and ESDs in various cities are non-profit orgs run by local businesses which can collect taxes, employ police and private security, and in some cases run courts. (Although the ESD courts were shut down by actual courts in St Louis). The ILSR claims big box stores rely heavily on the police to run their shops.

The ‘private army’ case isn’t always bad. When there was a rash of anti-Asian violence this year, for example, neighborhood watch groups formed and joined the fight: the Main Street Patrol in Flushing, Compassion in Oakland in CA. NYC mayoral hopeful + catfluencer Curtis Sliwa ran one of these in the 70s.

The modern watches are nonviolent and use the 5D method from Hollaback, an anti-harassment org. If you’re interested in being a helpful bystander yourself, click through.

Self defense, whether in the moment or with organized patrols, is a powerful argument because governments can’t deny it without looking bad. But frequent force, even in self defense, is not a good look. I started Hofstadter’s American Violence last month; he thinks that the need for force to keep order, regardless of who is doing it, is a threat to legitimacy.

Weber thinks states by definition need to be the sole provider of force in their borders, but had a complicated loophole about individual self defense. Also, was he ever in a street fight? I suspect he underestimated the hierarchy of semi-violent turf protection happening every day in a normal city. Janet Yellen of all people wrote a paper about why communities tolerate gangs and claimed that we have a non-zero floor beneath which gang violence is tolerable. I think hers is the better take.

To the extent decentralized safety networks enable consensus and norm setting, they also enable posse formation. That’s good when it’s for like picking up garbage together. It’s less good when it’s used to deploy force.

I’m not bullish about private use-of-force groups. The SS began as private security for party meetings. Woodrow Wilson’s AG Tom Gregory gave DOJ letterhead to the APL, a private army that hunted communists et al during WW1. It can happen here. It has. The bumper crop of armed militias rumbling at the fringes of our political system grew IRL on the last generation of social + forum tech. The bad actors in the decentralized model will mimic the shape of the network, will be smaller and less formal.

A millimort for a millimort

I didn’t see that golden arm show on Quibi but heard it was the best thing on there. I bring it up because in reality, some people have actual golden arms, or some other special status giving them an awesome career, and we let them sue for future earnings when something happens to that career. (There’s an ALAB podcast about problems with the formula).

The extreme opposite of golden arms is the guy or lady who drives a polaris trike past my place sometimes with the radio up. They also frequent Hoboken, which means either there are multiple polaris trike clubs in the tri-state area or the one has a very long route. The polaris website says they’re ‘impossible to ignore’. Can confirm.

Trike pilot isn’t breaking a golden arm. They’re disrupting hundreds of people over say 20 blocks for like 5 minutes each. Basically a papercut, but times a hundred. Noise hinders problem solving. Noise increases heart rate and causes heart disease with long-term exposure.

The trike driver is stealing a statistical week of life or work; just not from any single person.

Decentralizing the collection and categorization of crime reports won’t just amplify victims of serious crimes. It will also amplify the hundreds of small complaints against my trike driver by clustering them together, which enables collective action.

The tricycle is just a nuisance, but we’re also in the middle of a national debate on mask wearing in public. To the person who’s mask-less on a crowded subway car, who may or may not be vaccinated, who may or may not be sick anyway, who is certainly breaking the rules, everyone staring daggers at you now has a camera on their face. They’ve killed 1/1000 of a person, or 1/10000, not 1, but plausibly not 0.

Flattening the long tail of tiny risks is essential state capacity for fighting an exponential threat and we’re sucking at it. Decentralized safety is the right shape peg for this hole.

State capacity and its discontents

You know what else keeps the streets safe? Post offices apparently (pdf). Not directly, but as a proxy for other provisions of a strong state: formal dispute resolution with fair outcomes and the ability to enforce judgments. The development literature claims these capabilities protect countries against collapse.

Backdrop: as I was writing this, the Taliban was taking back Afghanistan. The US isn’t a failed state yet, though we failed there. Abdicating the role of ‘world police’ doesn’t mean abdicating policing our own cities. But whatever safety we made there relied on our continued occupation. Does my safety here rely on continued occupation? Jane Jacobs hopes not.

At peak lockdown last year, I had flashbacks to 80s New York – something in between the sewers of TMNT (yes the live action VHS version) and the scary Wiz version of Hoyt Schermerhorn. I know now that broken plate glass sounds like dry snowfall, something I didn’t know before.

I’m not saying crime is up, or that it isn’t up, or that it matters. I know the perception that crime is up can be politicized. But there are changes coming: technological, social, economic when the QE runs out.

Government at all levels here is slow to adopt technology; I heard a take that this isn’t because of a lack of talent, nor even fully because of corrupt or broken government purchasing, but rather because upgraded institutions bring updated values, and we can’t agree on what those should be.

Too bad, whatever the reason, because we sorely need the upgrades. Cross your fingers private sector innovation can pick up the slack. I’m not asking for decentralized institutions. I just want ones that work.

I opened with legitimacy and public data. But another challenge to legitimacy is incompetence. My generation has shrinking tolerance for the growing gap between state capacity and the state of the art. Decentralized institutions are a reaction to that, but if there’s a way to fix what we’ve got, that would be good too.


If you made it to the bottom you should subscribe to this blog here so you can continue to enjoy awkward 3000 word think pieces on public safety, periodic rants about terms of service, book reviews of late 60s sci-fi, the disastrous saga of my pro se lawsuit, something about muzak, I don’t know, there’s a long list of these but they’re sure not getting done fast! Eusocial continents, calculation debates, AI middle managers.


  1. A Brief History of Forensic Imaging, in the journal of biocommunication, gives the 1859 number and describes the case. The Law and Science of Evidence, in American Quarterly, seems to say the photo was used to examine and assemble evidence already submitted to the court, not as a piece of evidence itself. Also this gem: “The sun-pictures of various kinds have been in popular use for a generation; but many years passed before they were adopted into the apparatus of courts of justice,” noted an 1879 article in the Western Jurist. “The law of the land is a wary old fox, and scrutinizes a new invention a long time before extending the paw to appropriate it.” 

  2. More info: the US sentencing commission says criminal history is a major factor in committing future crimes and this is correlated with long jail terms (not sure they say if jail term is a factor independent of severity of crime). See also Pasco County’s attempt to use grades to do this. See also COMPAS as covered in propublica. This history of the NYPD’s compstat system talks about the link between statistics and preventive policing. You’ll note they’re doing statistics on maps + neighborhoods, where the sample size is high enough to direct patrols, rather than attempting to do statistics on individuals like the bananas precrime projects.