And so, it begins...
In which I make a dramatic entrance, and then proceed to talk quite reasonably--and without too much drama at all--about AI and data ethics.
So, I decided to join Substack. And you, dear reader, have stumbled upon my first post. Allow me to introduce myself.
My name is Katherine. I’m a data scientist, author, teacher, and a big ol’ nerd. I love what I do, and I spend a probably unhealthy amount of time reading about data, Artificial Intelligence, technology, innovation, and data ethics. The most useful, interesting stuff lands on my Twitter feed, along with the odd meme about the pains of being a programmer.
Having done this for a while, I realised I’ve amassed quite a collection of resources worthy of pulling any fan of data, tech, and our future with it, into a nice, deep rabbit hole. But not all of you can fall in there with me. Many of you will have jobs. And families. And the desire to eat and sleep and maybe get out and see the sunshine, every now and then.
So I’m doing you a favour. I’m going to package up the best bits into regular ‘digests’ for you to snack on at your own convenience. I’ll try to tie each edition together into one theme - anything from the problem of regulating AI, to the joys of telling stories with fancy graphs, to the coolest, most James Bond-worthy innovations in current research. I’ll also add a little bit of commentary. Not too much, but at the very least, something for you to share on your own Twitter or LinkedIn feed, or to bring up at a social gathering of your choice.*
*I take no responsibility for any arguments that ensue. If you’re dinner guests want to insist that that Google AI chatbot really was sentient, that’s on them.
Let’s kick things off with a series of resources dealing with the topic of AI ethics. There are thousands of reasons why this has become and will remain a key talking- and thinking-point in the data technology space, but in the interest of making these newsletters short but diverse in what they cover, I’ll provide just one example:
⚠ When AI Sees a Man, It Thinks 'Official.' A Woman? 'Smile': A paper raising concerns on bias in image recognition services.
Oof, not cool. And the problem is,
🛠 "AI doesn’t just scale solutions: it also scales risk, thus making data and AI ethics business necessities, not academic curiosities," Harvard Business Review, ‘A Practical Guide Building Ethical AI’.
The point is, when we accidentally build biases into our systems, and then we deploy those systems to the masses, we are multiplying the negative effects tremendously. What’s worse, we might not even realise it. That’s why practical guides, regulation, and oversight are so important.
Many companies are aware of this, and are becoming very interested in applying tech ethically through the frameworks of explainable and trustworthy AI (whether this is out of the goodness of their heart or just for the ethical brownie points is a question I’ll save for another edition).
Two of my favourite writers in this space offer some very interesting perspectives on how companies can build ethical AI and data products:
🤝 Companies should not ask "how can we make users trust our AI-based product?" but '"how can we build products worthy of users' trust?". Marisa Tschopp on the wrong questions to ask about trust and AI.
✅ Instead of relying on explainability in AI, "opt for the trust that comes from making sure that you’re able to verify that your system does, in fact, work." From 'Explainable AI won’t deliver. Here’s why' by Cassie Kozyrkov.
At the same time, governments, NGOs and academics are wrestling with questions around how to regulate technologies such as AI and automated algorithms, and whether this is even possible. Again, this is too big a topic for my humble newsletter. But here’s one interesting resource, to get you started:
⚖ Legal cases around algorithmic discrimination must not deteriorate into a battle of statistics. A potential solution metric is Conditional Demographic Disparity. From 'Why Fairness Cannot be Automated', by Prof Sandra Wachter.
So what can you and I do about all this? As you should now expect, I’m going to offer just one resource to get you started:
🧑🤝🧑Check out DataEthics4All, and join them if you like. In their own words, ‘We learn together why Ethics Pays and how we can help scale the impact of Social Good.’
And so, my first newsletter comes to a close. Thank you for sticking with me: I hope I’ve interested and inspired you, as I’ll be revisiting all these topics and more in future editions. You can hit the subscribe button, above, and please do tell your friends (otherwise there’ll be just, like, five of us here, and that gets kind of awkward).
You can also check out my Twitter feed for the unfiltered, unstructured, un-commentated version - but with memes*.
See you again soon,
Katherine
*A meme, for the unfamiliar: