We need to be careful with AI - and not just because it might take over the world

Increasingly, there is a need to be more careful when making "AI" (machine learning, deep neural nets, whatever).

We have already been warned by people like Elon Musk that if we're not careful, they might take over the world.

But even without that, we have other things to worry about. Our AI can easily become hurtful to people.

Take this story: theverge.com/2015/7/1/8880363/google-apolog..

When Google Photos first released a feature to automatically "tag" your photos (using a deep neural network), they ended up having their product make one of the most racist conclusions it could make.

Why? Was the AI intending to be mean? Well, not exactly. A neural network will reflect what you train it on. It seems that Google did not have enough diversity in their dataset.

That's a simple mistake, but think about the number of people that may have hurt.

And that's not the only time this has happened.

Here's another example: blog.conceptnet.io/posts/2017/how-to-make-a..

This literally shows with code that two of the most popular word-to-vector libraries (that map words to vectors based on their similarities) have some pretty obvious biases.

Again, this is a problem with the training used. The datasets were gathered from various sources all over the internet. That means it's learning from what we write.

It turns out, there are some people on the internet who are not very nice (🤯).

There are a couple of ways to make this situation better.

We could clean our datasets from those types of biases. How? These datasets are HUGE - way too large for a bunch of humans to manually read through.

Well, what if we trained an AI to do it? Make an AI that can detect racist, sexist, and other discriminating comments, and have it remove them from your datasets, so we can better train other networks.

But that's not really my point. Sure, we can deal with all these specific examples, but there's a deeper underlying problem here: we as developers need to remember that our code has consequences.

Because with the internet, our code can affect a very large number of people - that's part of why some of us like coding so much - but with great power (I'm sorry, but it's true) comes great responsibility.

We as developers need to think about these issues and make conscious choices about how to handle them in our code.