STM to use AI to spot potential suicides
The STM is to use artificial intelligence to spot the patterns of people contemplating suicide on the metro platform.
Incidentally, Quebec has been told it needs a law to control artificial intelligence while everyone who can make use of it is doing so already.
Nicholas 10:47 on 2024-02-06 Permalink
Unfortunately platform screen doors, which physically prevent access to the tracks and are being installed worldwide, were cut from the Blue Line extension and are too expensive here because we can’t control costs, why we can’t have nice things.
Kate 11:10 on 2024-02-06 Permalink
The AI will help if it means oncoming drivers can be warned, and trained security people sent to the platform to talk to the person, as it says in the piece.
I’ve always been against platform doors. While I’ve never visited a city that has them, I’ve seen photos, and can see they would make the metro experience so different, like walking down a narrow hallway. Our stations with the most architectural presence would be damaged forever – the large arched volumes would be blocked off from view.
That said, I’ve never lost anyone to a metro suicide.
Uatu 11:22 on 2024-02-06 Permalink
There’s subway gates that I’ve seen that are around chest high. Just enough to make it hard to jump and also prevents access to tracks, dropped cell phones etc.
Roberto 14:20 on 2024-02-06 Permalink
What a slick way to increase surveillance…under the guise of ‘suicide prevention’. This only serves eager tech bros claiming innovation. You want to prevent suicides? Look at root causes.
Kate 14:40 on 2024-02-06 Permalink
Roberto, while we’re waiting for utopia, it can’t hurt to dissuade people from obeying a momentary impulse that can end their lives or leave them alive with permanent disabilities. Studies have shown that if you deter a person from a suicidal impulse, generally they don’t immediately seek an alternative, but turn aside and find a way to go on living.
We can’t place watchers on every platform but if we can mobilize AI to recognize patterns and issue relevant alerts, I think it’s a good use of the technology.
DeWolf 14:50 on 2024-02-06 Permalink
Stations retrofitted with platform screen doors don’t have the fully enclosed platforms you see in newly built stations. They have something that looks like these:
https://en.wikipedia.org/wiki/Platform_screen_doors#/media/File:Bulgaria_metrostation.jpg
Personally I don’t think that detracts from the station architecture and it doesn’t make things too enclosed.
And with new-build stations, you can design the platforms with the screen doors in mind. The REM stations are fully enclosed and yet they feel quite bright and airy, and even Central Station, which is underground, feels quite spacious.
DeWolf 14:52 on 2024-02-06 Permalink
Also, the screen doors aren’t just to prevent suicides. When the STM was actively considering them before the pandemic, stations like Berri were getting so crowded there was a risk of people inadvertently falling onto the tracks.
And then you have the unfortunate cases of people being deliberately pushed into the path of oncoming trains. Luckily that doesn’t seem to have happened here, but it has happened an alarming number of times in New York.
Kate 18:25 on 2024-02-06 Permalink
It did happen here, DeWolf – a woman was pushed off the platform by a stranger at Guy metro, which I remember reading about, but I can’t find the story online – it was some time ago. Looking for it I found this 2004 incident in which a bunch of teens scuffling on the platform knocked a woman onto the tracks, probably not on purpose; the sad story from 2013 of the young woman who fell between two cars of one of the old trains; the 2015 story of the woman who fell onto the tracks at Jarry station.
walkerp 19:58 on 2024-02-06 Permalink
LOL @ AI being able to detect potential suicides. Some slick salesmen upsold the software buyer at the STM. AI is the Blockchain of 2023-24.
Kate 20:55 on 2024-02-06 Permalink
walkerp, if AI can be used to screen x-rays, which it is, it can be used to look at a lot of things human beings don’t have enough hours in the day to look at. We can’t afford to pay people to watch the inbound end of every platform of each of 68 stations for 18 hours a day, but we can train an AI to look for typical lurking and wavering behaviour and alert a human being to assess it and act if necessary.
What concerns me here is bias. I’ve been doing illustration work using Midjourney and other AIs, and if I ask for images of people, they almost always turn out young and white. If I want an image showing people of different skin tones or different age ranges, I need to specify this – the AI won’t usually volunteer it.
You don’t have to work with AI for long before figuring out that it mirrors the biases in our society, but you can also learn to work around it.
If we were asking the AI to look out for criminal behaviour, I’d be concerned that it would absorb the biases that make police assume all young black men are shady. But in this case, we’re not.
Tux 17:30 on 2024-02-07 Permalink
I foresee people getting surrounded by cop wannabes for being potential jumpers who just like standing at the end of the platform. Also, AIs need to be trained – do we have the data we need to do an even half decent job training the model? Everyone’s jumping on the “AI” bandwagon these days…
Ian 22:30 on 2024-02-07 Permalink
As I recall several years ago there was a similar initiative, basically looking for people at the leading edge of the platform behaving in an “agitated” manner. Everyone is talking about AI now but we shoudln’t forget that there has been video analysis technology for a long time now. They used to call it “assisted analysis”.
Suicide pattern detection was being researched here as early as 2018 in prisons. pdf link
Intelligent video surveillance for real-time detection of suicide attempts
LICEF research center, TELUQ University Montréal (QC), Canada
École de technologie supérieure, Montréal (QC), Canada