3 ms·
Optimization over Explanation
- everdev 9y ago> Human-constructed models aim at reducing the variables to a set small enough for our intellects to understand. Machine learning models can construct models that work — for example, they accurately predict the probability of medical conditions — but that cannot be reduced enough for humans to understand or to explain them. Some problems in life are too complex to explain in 5min to someone without knowledge of the space.
- hyperion2010 9y agoWitness the number of times someone complains about a feature of a programming language where to anyone who has encountered the problem it was created to solve it is fairly obvious that the complainer simply doesn't get it, and usually no amount of explaining will alleviate them of their complaints. They usually have to have the problem for themselves because it is not something that is easy to articulate.
- xapata 9y agoThe trouble with these black box models isn't that we're afraid of uncertainty, it's that we're worried they're using a bad proxy. Throwing more data at the algorithm won't solve an omitted variable problem or an endogeneity problem.
- xg15 9y ago> But here’s another: Accept that we’re not always going to be able to understand our machine’s “thinking.” Instead, use our existing policy-making processes [...] to decide what we want these systems optimized for. Measure the results. Fix the systems when they don’t hit their marks. Celebrate and improve them when they do. If really no one understands how the algorithm works, how could anyone meaningfully "fix" or even improve the algorithm?