• 1 Post
  • 133 Comments
Joined 1 year ago
cake
Cake day: June 19th, 2023

help-circle

  • I’m not sure we, as a society, are ready to trust ML models to do things that might affect lives. This is true for self-driving cars and I expect it to be even more true for medicine. In particular, we can’t accept ML failures, even when they get to a point where they are statistically less likely than human errors.

    I don’t know if this is currently true or not, so please don’t shoot me for this specific example, but IF we were to have reliable stats that everything else being equal, self-driving cars cause less accidents than humans, a machine error will always be weird and alien and harder for us to justify than a human one.

    “He was drinking too much because his partner left him”, “she was suffering from a health condition and had an episode while driving”… we have the illusion that we understand humans and (to an extent) that this understanding helps us predict who we can trust not to drive us to our death or not to misdiagnose some STI and have our genitals wither. But machines? Even if they were 20% more reliable than humans, how would we know which ones we can trust?


















  • I only have a limited and basic understanding of Machine Learning, but doesn’t training models basically work like: “you, machine, spit out several versions of stuff and I, programmer, give you a way of evaluating how ‘good’ they are, so over time you ‘learn’ to generate better stuff”? Theoretically giving a newer model the output of a previous one should improve on the result, if the new model has a way of evaluating “improved”.

    If I feed a ML model with pictures of eldritch beings and tell them that “this is what a human face looks like” I don’t think it’s surprising that quality deteriorates. What am I missing?