3 ms·
What is really interesting about this article is the compartmentalization of function and that this analyst was able to rise above siloization to prevent a kill
by sjg007 6y ago
What is really interesting about this article is the compartmentalization of function and that this analyst was able to rise above siloization to prevent a killing based on mistaken identity.
That's a critically important role and I think an apt description of knowledge work today. And it has implications for machine learning and predictive analytics.
Here the wrong guy was targeted based on the purple hat which was a sensor processing error based on early light levels.
While integrating all of this data is important and building machine learning models that can pick up specific markers is important, it is clear that you need some type of overall exception processing framework to confirm in this case positive id.
We have a similar problem already with law enforcement no knock warrants.
- Jugurtha 6y agoOne of the most frequent question we ask our clients is "What's the cost of a false [positive/negative]?" and "How do we know a prediction is correct?". We work with them to see the whole system and outcomes of our predictions, down to the human actions. For example, in a project where failing to predict something would result in costs up to $100 million and wasting two months, and catastrophic consequences, but a false positive would result in an engineer waking up to check, it is acceptable to have false positives. Yes, it may be inconvenient, but compared to having nothing and the engineers telling you it's okay to bother them because one of catastrophic consequences is they may lose their life, the system can cry wolf. In other projects, not so much. And in other projects, you're not dealing with consequences of that magnitude.