Instagram Pro – IT Pro – News to testify before US Senate on impact on teens

It is completely thrown out the window and only develops from intimidation by the community and our fellow citizens. “AI” is incomprehensible and we tell each other that it is pure nonsense. (Source: I MSc in AI, PhD in NLP)

First and foremost: instructions, structures, of course written by one person. You can not say “OK computer, create an AI” – it has a lot of programming and theoretical approaches. The field of AI has been (theoretically) working on these types of things since the 1950s-60s. So the math behind it No. A “black box”, oh often asked. If you want, you can create your own neural network – entirely on paper! How a neural network is trained is nothing more than mathematical formulas. This is all math, not magic.

Second: You could argue that it is not always clear (to us humans) why the computer gives a specific output (eg cat) based on a given input (e.g. a photo). The machine “learns” subtle differences (features) in images that we humans cannot see immediately (may be wrong!). However, that does not mean we are not Understand How is that or we can not interfere. Again, you can do the math yourself and you will get the same result. Those subtle differences I talked about are our problem as human beings. We can not always understand it. Fortunately, there is a movement in AI that seeks to further explain such hidden features.

Third: it is “interfere”. It is certainly possible, and it has often been shown that it is possible. Some political parties gave priority to others or others to certain products. It often has the human-in-the-loop. But even if it is not: “Even in meta, no one can tell you why some posts prioritize others over time”, it is utter nonsense. They use a comprehensive set of features: what you last clicked on, what you saw, what your friends talked about on Messenger, and more. Anyone can find it. (That’s how recommendation systems work.) These features are pushed through the model and output is output. As with any research in this field, those features are carefully selected, and they can certainly see locally which aspects have the most predictive power. If they want, it is easy to disable any of these features.

In other words: it is definitely possible for Facebook to show you the reason why you see a particular news / ad. But they choose not to.