I don't see anywhere in the article where they call it AI. It is described as an algorithm, which seems appropriate to me.

In the caption of the video "WATCH: AI helps clear cannabis conviction backlog".

The word is associated with the cutting edge whereas the actual code itself I imagine is mundane and hacky

The software does perform text mining (presumably via some sort of NLP?), an area that is firmly a subset of AI.

It could just be very simple keyword matching, especially if this is looking at court records.

It seems that this might indeed be the case. After writing the comment I’ve had a look through the (obsolete) Ruby code base of the project that somebody linked, and it doesn’t actually perform (what I’d call) text mining as far as I can see. Instead it parses the court documents, which seems to be a mostly structured format. No natural language processing.

I think that's fake code in the images ;)

I was trying to work out what language it is - also switching on strings like that doesn't seem like a great idea.

The code in the screenshot is valid Go (switching on strings, no parentheses around the if condition, incrementing a map value using ++) and could even be the output of go fmt (with the { on the same line as the if).

The app itself appears to be written in Ruby (the github link is posted elsewhere in the comments), and looking at the image included in the article further down I assume the code is from a Go app that takes the Ruby app's logs and generates some statistics.

The project is hosted on Github [0], I'm not sure if the entire source code is available though.

[0] https://github.com/codeforamerica/autoclearance

Hi, engineer on the project here! That's an old version of the code that turned out to be too slow for the volume of records we need to process. The current version, which you see in the images from the article, is written in Golang.

You’re doing incredible work. Keep it up!

(Also use consts so you don’t make typos - or so your typos are at least consistent :)

Can you go into that a little more - the switch to Golang and the old version being too slow?

I would hope the entire source is available. We should always be wary of secret algorithmic decisions in court cases. The algorithm should be openly available for the public to see in the same way that the law is to be openly available.

Yes, that was painful to see. The strings are even capitalized in a particular way so it doesn't look like it had already been processed to normalize the formatting.

Well it’s relevant to the story, so if it’s fake they put a surprising amount of effort into it...

If you go to 1:51 in the video, you can see the same code on what appears to be an engineer's screen. It seems like it would be a lot of effort (and questionable journalism) to fake that.

for case in cases: if "cannabis" in case: case.flipverdict()

There u go ,crazy machine learning algorithm :O

Now you've created bugs where you overturn "not guilty" verdicts, expunge verdicts of cases that tangentially involved cannabis, and fail to expunge verdicts of cases where cannabis wasn't in the text but that still fell under the law change.

Haha if it's not correct to start with it must be some sort of "machine learning".

Besides, how many innocent verdicts for these crimes can there be? Prosecutors don't care near as much about pot possession as cops do (because they're not the ones getting the federal money), and they plea out possession when there's any doubt about convicting on more serious charges.