|
|
|
|
|
|
by monetus
1463 days ago
|
|
|
I wonder how this will play out in court when a neural net that underpins the tech of a company also can't properly handle health data. Start over? The right to be forgotten would have to be built into it all right from the get go, wouldn't it? Neural nets embed things, like you said. If Facebook pivoted towards being a hub for healthcare data as oracle seems want to do, could they use all of their facial recognition tech trained from profile pictures now that it is supposed to be detached from any one identity? I'm thinking of the equivalent of github's co-pilot spitting out comments alongside the code. |
|
|
yes exactly. I am having trouble articulating my underlying point but this is along the lines of it.
A facial recognition algorithm is ultimately saying "you look similar to this specific set of people". If the training set was 1 person, for example, then the algorithm would pretty much just be saying "you look like this photo of this 1 person". Scaling that up does not improve privacy - it only blends you into a population. Additionally, it is distilling out the things about you that separate you from that population (things that make you, you) since those distinguishing features are exactly what the model needs to use as the line to draw on whether or not new data points are similar to the population set or not.