Ruminations

AI: Studying Suicide Cluster Prevention

"I'm always in awe of the incredible power of big data but I don't understand how to make the leap from identifying group trends [such as in] elections to individual behaviour, which is suicide."
"I cannot reveal details of the Cape Breton case because I can't but there was absolutely no way, zero chance, that it could have been predicted which kids were going to die by suicide from the social media messages that those kids were sending because they didn't send any."
"You can't prevent the death of an individual by finding group trends. If you are able to spot a group trend then it's not going to help predict which kid is going to die by suicide and which is not."
Dr. Stan Kutcher, psychiatry professor, Dalhousie University, Halifax

"We also have additional safeguards in place to make it very difficult or even impossible for our own staff to look in and see who is having issues and finding out who that person is."
"What we would like to try and understand is what are the signals … that would allow us to forecast where the next hot spots are so that we can help the government of Canada to provide the resources that are … going to be needed to help prevent suicide before the tragedies happen."
"This is not Minority Report and we are not identifying individuals who … have risk of self harm. We are not knocking on doors or contacting individuals. We have nothing that is personally identifiable about any individuals in this study."
"So many times in AI research we hear the stories about AI is going to take jobs … Big Brother is spying on us. If you can show that [suicide] rates have gone down because we have deployed this sort of study, that would be most gratifying."
Kenton White, chief scientist, Advanced Symbolics Inc.

"The problem is you could [examine] the past like that but it's not necessarily generalizable to the future. There's such a unique set of circumstances that may not quite align themselves again that led up to that clustering in Saskatchewan."
"We had no confidence in those results [from an earlier 2015 study] but it also gave us a lot of doubt about these kinds of big data approaches [to predicting suicides]."
James Coyne, Emeritus psychology professor, University of Pennsylvania

Using artificial intelligence to analyze social media trends, the company hired by the federal government claims it could predict surges in suicide rates, including the precise region where they will occur. (CBC)

In Canada, daily deaths as a result of suicide impact eleven people, making suicide the second leading cause of death for those between the ages of ten and 19, according to the Public Health Agency of Canada. The Canadian government has undertaken an experimental study to be conducted by an Ottawa-based market research firm, Advanced Symbolics Inc. The pilot project will start off by working with health officials to define "suicide related behaviours".

Following which the company plans to scan random samplings of social media extracted from 160,000 public accounts, for the purpose of determining whether red flags can be detected -- particularly with a specific area -- to give warning it may become the locus of a group of suicides. Previous suicide hot spots, such as one in Cape Breton where three middle-school students committed suicide within a short period last year, will begin the focus.

The company plans to report its findings, if it succeeds in detecting patterns in suicide clusters, to the Public Health Agency of Canada, the information to be used to understand how and where and when mental health resources should be deployed preempting tragedy. Dr. Kutcher of Dalhousie University, a youth suicide expert, worked closely with the Cape Breton community and has expressed his considered doubts that any data collected through research into social media would be effective in discovering red flags in an efficient manner, connecting resources with those deemed at risk.

Suicide, points out Dr. Kutcher, is an extremely personal issue and as such behaviours flagged as suicidal are individualistic. He also points out that the age group representing the greatest numbers of death by suicide fall into the age 50 to 54 range, a demographic, he points out, less attuned to the regular use of social media. Moreover, even if the artificial intelligence to be used does pinpoint spates of potential suicide, no proof exists that rushing in with mental health resources at the area at risk would have any effect in lowering suicide rates.

PHAC plans to consider whether to proceed with the surveillance and "market research on the general population of Canada", as a follow-up phase of the project, should Advanced Symbolics succeed in accurately predicting patterns in suicide behaviours linked to past instances. Dr. Coyne, cited above, with over 350 research papers to his credit on big data and suicide, feels doubtful that examining social media samples would lead to suicide clustering predictions focusing on First Nations' communities.

Professor Coyne had worked with a graduate student on a project that gathered Twitter data to determine whether angry tweets could connect with increases in heart attacks in a number of counties in the United States. The algorithm proposed for use in this current study was used in their effort to determine whether angry tweets could be connected to an increase in suicides. Their study resulted in the realization that it was in fact positive tweets that predicted an increase in suicides.