I watched the Netflix Documentary American Nightmare. It feels like standard fare at this point, where we dip back a decade or two to some crime story that had a few twists and turns in it and make it freshly popcorn worthy again. This was the “is this the real life Gone Girl?” case. Denise Huskins was very literally kidnapped and the bumblefuck cops spent all their time blaming first her boyfriend then her, until they were smacked in the face with enough evidence they were forced to admit it was real.
In the last (3rd) episode, a then-new woman detective gets involved in another case a few hundred miles away, and they catch the guy. She’s smart, and sees a blonde hair in the creepy taped up swim goggles in the criminals crime-hole by Lake Tahoe. She’s like… but our victim wasn’t blonde… so there is probably another victim out there we haven’t connected yet. It took all kinds of long-term dedicated commitment to this idea before she was able to connect enough dots to get it back to the Huskins case.
The cases had a heaping helping of similarities. The white Mustang, the taped up swim goggles, the home invasion, the flashlight taped to the gun, the zip ties, even the mannerisms of the criminal.
With the backdrop of AI we’re all living in, it was impossible not to think about how computers should have connected the dots on this in two seconds. This took far too much human gumption and luck to solve this case. I don’t even know if AI is what’s needed here, but maybe it would help? The natural language thing cops could use like “show me other cases where the perpetrator used swim goggles” feels like it would be helpful.
I’m sure there are all sorts of dangerous implications to having AI crime models (see Minority Report, Judge Dredd) but it feels like there has gotta be some better way to Do Computers to solve crimes.
All that makes me think about my kids ultrasound the other day. She had been sick a lot of days, and they wanted to make sure it wasn’t appendicitis (it wasn’t). This required using the jelly and the whole thing. Then one (1) singular doctor looks at those pictures and decides what’s up. Shouldn’t those pictures instead go through some AI model that has been trained on every single ultrasound ever taken with diagnosis and outcomes, in order to assist that singular doctor in what she is seeing?
I’m hoping the coming AI revolution helps us all be smarter in ways where “collective knowledge” is crucial.
First of all, I’m sorry to hear your child went through that! I’m wondering if the concern is that the AI models just aren’t quite there yet. I seem to recall John Oliver had a segment on AI a while back. While attempting to train it to recognize skin cancers, it learned to identify rulers as a sign of malignancy, since the healthy skin photos lacked rulers and every cancerous photo contained a ruler. I’m sure it’s come a long way since then, but when we’re relying on such crucial health diagnoses, where false positives can cause a lot of anxiety and unnecessary stress/extra unnecessary tests, I can appreciate the caution at this current time.
Hi Chris – sorry to hear about your daughter!
This kind of AI is already in use in a number of hospital systems and imaging centres in the USA and around the world. (I work for a company that helps deploy such apps.) For a whole lot of reasons, it’s a bit of a lottery at the moment as to whether your chosen healthcare system will use them or not. But I can’t imagine that they aren’t going to become a lot more common than they are at the moment.