Tumblr is currently using a neural network to identify images that contain explicit media. Neural networks are extremely susceptible to something called adversarial examples.

Adversarial examples are inputs to a neural network that result in an incorrect output from the network. It’s probably best to show an example. You can start with an image of a panda on the left which some network thinks with 57.7% confidence is a “panda.” The panda category is also the category with the highest confidence out of all the categories, so the network concludes that the object in the image is a panda. But then by adding a very small amount of carefully constructed noise you can get an image that looks exactly the same to a human, but that the network thinks with 99.3% confidence is a “gibbon.” Pretty crazy stuff!

Source:

Check that link if you want to know more.

-I'm not advocating any rule breaking, I am pointing out how underdeveloped this technology is and why it should not be rolled out on such a big scale.

The surprising reason why pharmacology is so difficult, really almost impossible.

It’s an all-too-familiar story. A laboratory somewhere announces that they have discovered a chemical formula that cures a previously intractable disease like cancer.

It works perfectly in Petri dishes, and in mice, and even in monkeys, and now the human trials begin.

Months later, we read that the formula not only didn’t work, but it made the patient sicker.

Why is pharmacology so difficult?

Bio…

This is machine learning's big problem, and I think solving it is a lot harder than many believe.

I think Bruce is mistaken about this. Humans are just as susceptible to adversarial examples: If I put up signs that indicate a lane is closed, traffic is one-way, etc. I can easily fool humans into driving into oncoming traffic.

Oh, but the machine is fooled by examples that would never fool a human? And? It still requires malicious intent on the part of someone. Just as we would blame someone for putting up malicious signs, wouldn’t we also blame someone for putting up malicious stickers?

It’s not as though we are taking about machine confusing two everyday situations (I’m sure this happens in poorly trained software, but it’s not the attack) we’re talking about a specially contrived attack. I doubt you even need new laws to create liability here, as it is pretty hard to deny that some weird specific sticker configuration isn’t intended to be an adversarial example.

I swear this is an image on a UC Berkeley website and it’s literally the most clinical statement about memes I’ve ever seen