"Two years ago, when Elinor Lobel was 16, a “smart” insulin pump was attached to her body. Powered by artificial intelligence (ai), it tracks her glucose levels and administers the right dose of insulin at the right time to keep her healthy."
Is AI necessary for this, or just software more generally?
I find it hard to believe that's a real story (although they do use a full name), in North America at least, vendors refuse to automate insulin pumps because of the risk of litigation (too much or too little insulin can cause life changing damage, even be fatal) - and that type of automation would be `if (sugarlevel increasing-over-time) administer-small-dose(fast acting) - resample`. Basically a PID controller (although not exactly since insulin can only reduce blood sugar, so you'd err on the side of caution in your delivery and have big alarms to get human attention if you detect crashing sugar levels). Given legislation around the world is starting to require AI backed processes to explain the decisions it makes, it seems more trouble than it's worth - this is people's lives.
This point is Machine Learning and for literature it may be indistinguishable from AI but can be just a Linear Regression / Random Forest / Basic Neural Network / Transformer implemented in 1000 lines of C or Python. The main difference in my opinion is the options we have to train those models [0] and be confident we can trim it enough to fit on the target device / get upgrades.
I think what the author should refer to is this standardized way of collecting data, running the models and shipping the weights for low power inference.
[0] https://cloud.google.com/bigquery-ml/docs/reference/standard...