July 8, 2022

s12e35: Wait, it's just a guess?!

0.0 Context Setting

It's Friday, July 8, 2022 and there's weather in Portland, Oregon.

---

1.0 Some Things That Caught My Attention

Wait, it's just a guess?!

Portland, where I live, is thinking about using ShotSpotter "to curb gun violence"¹, and that would be the same Shotspotter that just lost its defamation case against Vice media, after Vice pointed out that some records have been altered at the behest of law enforcement.

A journalist gets in touch with me to run a story about this potential deployment of ShotSpotter, which is the chat I've just had about half an hour ago, and it is definitely something that caught my attention. The journo asked me a few questions, and one of my favorites was "can you explain ShotSpotter", which I had a bunch of fun doing.

The thing about ShotSpotter is that it's quite simple and easy to understand. You set up a bunch of microphones and they start listening for sounds. You set up enough of them and you know where they are so that when there's a loud sound, it arrives at different microphones at different times. Then you use a bunch of high school maths to figure out, roughly, where that loud sound was, or what direction it came from.

Now comes the fun bit. What ShotSpotter does afterwards (and what any tech company will say it does, essentially) is that it will run a bunch more maths and call it AI or Machine Learning or an Algorithm, and what that will do is try to guess whether a loud sound was a gunshot.

The fun bit is where the journalist does a bit of a double take and says: "wait, it's guessing?"

And I get to say: Yeah, sure it's guessing. I mean do you know for sure, just by sound, whether something's a gunshot or just a different loud sound? Can you tell, without seeing it? No, you can't. So computers do roughly the same thing: they use maths to guess -- produce a guess, at any rate, with a level of confidence, as to whether a loud sound is a gunshot or not.

Well, that changes things quite a bit. Because the way we sell technology, and they way we like to use technology is that we want them to deliver certainties, and outsourcing certainties and making decisions to third parties is quite nice and reduces our stress. We can say somebody else or something else made this decision. The thing is, we always make the decision in the end: we chose to use ShotSpotter, and we chose to attempt to absolve ourselves of responsibility. (Which, of course, is where the whole deal with law enforcement coming in and, a few times at least, wanting to override whether ShotSpotter says a loud sound was a gunshot or not)

But that's the thing. It's a guess. Sure you can call it machine learning, but I genuinely believe that a more intuitive, more realistic and more helpful way of thinking about these systems is that, roughly, they are guessing, and some of them are better at guessing than others.

Anyway, the interview also gave me an opportunity to talk about what the point of ShotSpotter even is. Look, I'm a resident of Portland, I don't like gun violence or gun crime, I'm perfectly happy to admit that there's a nearby kids park I don't like to go to anymore because there's been a couple daytime homicides in the past couple of years, or that there have been shootings just a couple blocks a way. Not good! So I understand the need and desire that something must be done, but the thing is, what would you even do with ShotSpotter data?

Say you now get a running list of Loud Noises that you didn't get before, so your law enforcement have a bunch of ears around the city:

• Do you have the resource to run them all down? Probably not. So you need to decide which ones to investigate.
• Will you change which ones you investigate? How would you know? Should you? Are you changing the ones you investigate for the better?
• If it's not just responding to loud noises (e.g. now you have a Department for the Responding to Loud Noises That Might Be Dangerous), then how will you use that data for preventative, inclusive, equitable community policing? Would it change what you're doing?

The big question, in my mind, is whether ShotSpotter would change anything, and whether you'd know that change was good or not.

One thing I suggested was that if ShotSpotter were used, then open data would be a red line, otherwise there's not even the remotest chance of accountability, and that's supposing that you believe in institutions capable of enforcing and wanting to follow through on accountability in the first place. But the general argument was this: other cities have open data policies, I mentioned that I was looking at Washington, D.C.'s ShotSpotter data, updated quarterly, and that there'd better be a good reason why the data couldn't be updated every month, to at least make sure that ShotSpotter data was being used for something nominally useful. Would it, in the end, just provide more justification for heavy-handed, racist, policing? How would you know? How would you know if it were an adequate tool in prevention, in the first place?

So in the end it came down to pointing out that scads of data don't do anything in themselves unless there's a clear policy aim. What are you supposed to do with it? Do you know what it means? Do you know what you want it to be, and whether that's achievable? Which isn't to say that this is just a policy issue, which makes it a community political issue, but also that the technology lives and should be used in service of that. It would be a damn shame if ShotSpotter comes in and is a waste of time, even worse if it increases racism and inequity.

Logistics

I am waiting today due to logistics: there is a package that is coming for me and has my name on it, and it is the kind of package where a human has to be able to scrawl on a tablet to say that a human received the package. This could not, in general, be improved by "the blockchain". My day has been roughly shot due to a combination of waiting for this package and also the surprise appearance of a rat, which our cat helpfully caught.

The second thing about logistics is that I got a push notification that another package is coming, which is to say that a medical device is coming my way. This is good, because I've been waiting for it, and also weird in that uniquely 21st century way: I find out about the package because of the logistics provider, and not from my medical device provider. This is because the medical device provider has outsourced its delivery infrastructure (which makes sense!) and accidentally as part of that outsourced its customer/patient notification system. It is also a bit like when MyChart sends me a push notification that my labs are ready before my doctor is ready to interpret them, allowing me to slowly freak out or Google hypochondriac myself if anything looks weird.

---

Okay. A difficult day today.

How are you?

Best,

Dan