When explaining #MachineLearning to people, I find it useful to describe the arithmetic mean as a simple "learning" algorithm: By taking the mean of a sample from a population, you get a ~binary classifier between "above avg" and "below avg". I think it helps people to get three concepts:
* Machine Learning is not magic, it's maths
* Some forms of machine learning are actually pretty simple
* The data it trains on gives it a bias, & that bias can reinforce itself through misuse
#DataScience
@cathal
> The data it trains on gives it a bias, & that bias can reinforce itself through misuse
- THIS
@Antanicus Yep; and it's hard to impress this on people when the algorithms seem so abstract. But it's easier to explain to people how using the mean can be abusive or reinforce expectations/norms/sampling bias. Once they grasp that, it's easier to explain how the same problem, with much higher dimensionality and subtlety, appears in all machine learning systems.
@cathal misuse is certainly the default. Modifying our behavior for the sake of "engagement". It sickens me. Users should be willing to pay for social media because the price is you otherwise. Of course federated free software social media works too.