People in different disciplines speak different languages. Well, not really but the same term can have many meanings. No rocket science. However, the extent of these differences and how difficult it is to spot them is tremendous.

In this post, we’ll look at a specific example, try to understand why it happens, and what we can do about it.

What’s the problem?

Let’s talk about accuracy. In science, it has a clear definition and a corresponding equation. Generally, it measures how often you get something right on average. But outside of my mathy circles, accuracy is what you’d call any measure that tells you how well something works. Accurate weather prediction, accurate betting, accurate cancer screening etc.. A mathy person reading this is already thinking about precision, recall, odds ratio, and other metrics that are more suitable for these tasks.

This divide was apparent when the EU AI regulation proposal was published. The machine learning community started splitting hairs over the abovementioned accuracy. You’d hear a lot of comments about legislators having no clue about software or statistics. It misses the forest for the trees. They know exactly what they’re talking about, they just mean something else.

Heck, I’m guilty of it myself when I bombard my security colleagues with machine learning lingo, or talk to my friends about my work.

How did we get here?

Consider a privacy-preserving machine learning project. How many different professions could be involved?

1. product managers
2. legal
3. privacy team
4. data scientists
5. software engineers

Put them in the same room and watch it burn. All these people look at this project from a different angle. They have varied educational backgrounds. They use words assuming that the others understand them just as they do. It’s all very implicit.

Cherry on top, even if they share the same goal, i.e. a new useful thing that people can use, the success criteria are different. After all, correct code in production doesn’t mean much if the company is sued for breaking the law, or there was no product-market fit in the first place. This extra piece of friction adds even more to this lack understanding.

Now what?

The takeaway from this is that we’d try to understand our colleagues and business partners. Crucially, be explicit, and ask for clarifications. Make sure that you share the vocabulary. A neat tool is to ask people to repeat back to you, though in their own words, what you just told them.

Obviously, no one has the capacity to understand every conversation at the expert level. A software engineer with top-of-the-line legal knowledge doesn’t exist. If they do, they probably charge a lot. So try to hire people with multidisciplinary skills. Encourage your team to dabble in things outside of their core skill set. The goal is reduce the friction and make it easier to communicate across departments. At the beginning, it introduces some overhead but down the line, it’s worth it.

I know it’s easier said then done. The first step to making things better is realising that this could be a problem in your team.