Ethics for AI — A quick overview
☯️ What is ethics for AI? Why is it important?
Ethics are, according to Oxford Languages “moral principles that govern a person’s behavior or the conducting of an activity”.
For an AI, ethics can thus be defined as the set of rules that determines its “behavior” (or decisions).
Actually, the AI in itself is neutral. It is a sequence of mathematical operations, that gives a result its creators are interested in. However, its implementation and the data used to train the AI are subject to bias — and that’s where the problem lies.
Indeed, because AI is becoming increasingly complex, some models become less interpretable. Basically, it means that it is difficult to understand how the AI computed the outcome. Then, since AI is used for decision-making it can systematically, inadvertently or not, harm a category of people.
💸 Here’s an example. Let’s say you are a banker developing an AI to decide whether you should grant a loan to your clients. One client, A, living in neighborhood B comes to you.
You don’t have a lot of data on that specific area, but the only people you know who lived there were not able to reimburse their loan. Then, it’s possible that the AI could systematically refuse the loan to A.
Is that fair? Can you judge someone based on where he lives? Sometimes, unfortunately, only the AI knows.
🧭 5 sources of bias
As we’ve just seen, biased data can lead your AI to make unwanted predictions.
According to Harini Suresh, a researcher at MIT, there are 5 potential sources of harm (or bias) in the AI process.
- Historical bias: bias that already exists in the world.
- Representation bias: bias that happens when the data used to build the model is not representative of the final population.
- Measurement bias: occurs when the features you use to predict an outcome that can be incomplete or contain noise for some groups.
- Finally, Evaluation and Aggregation biases are a result of choices made to build the AI.
If the topic interests you, we highly recommend that you read Harini Suresh’s article.
🤕 Actual harms AI could make…
Now that we’ve seen how bias can appear in AI, and why it is important, let’s look at the potential harms it could make. For this, there is no better illustration than this table from Future of Privacy Forum, which you can see below.
Source: Future of Privacy Forum.
Ethics in AI is a huge topic, and we’ve only scratched its surface today.
In the coming years, we can expect governments and companies to face more pressure to regulate ethics in AI, as these algorithms will take an increasingly important place in our lives.
That’s it for today… if you’re still hungry for cool tech stuff, we invite you to check out our weekly newsletter “The Seed”. It provides easy-to-read content focusing on Entrepreneurship, Innovation, and Technology.
