3 ms·
That's not the only mechanism. Even if you want to stick to technical issues, it can be biased because you didn't train it long enough or the model is too small
by astrange 6y ago
That's not the only mechanism. Even if you want to stick to technical issues, it can be biased because you didn't train it long enough or the model is too small. And of course the entire question depends on what the model is being used for.
- d110af5ccf 6y agoEverything you say is true, but in context the important point is that the sort of biases ML models pick up are overwhelmingly related to training insufficiencies and are often incredibly difficult to spot unless you already know they exist. For a practical example, see the recent Twitter image cropping oddities (https://twitter.com/bascule/status/1307440596668182528 https://twitter.com/bascule/status/1307440596668182528). The idea (as quoted) that models are routinely picking up biases directly from researchers is complete nonsense.
- astrange 6y agoRight, that's some kind of human interest journalist fudging and it's not true. But bias/surprising wrong answers in ML is obviously a real problem and fixing the data is not always the right answer. You might not be able to tell what's wrong with the data, or where you could get any more of it, and you might be reusing a model for a new problem and not have the capability to retrain it.
- visarga 6y agoWe should only use models where they work well. Like in architecture, we should only build what will be safe for use.