This is one of the most important, and often most painfully learned, disciplines for an engineer.

Early in our working lives, some of us were a touch overconfident. We stated something as fact and found that people listened. We’re pretty sharp, and we’ve learned everything that’s been thrown at us—that confidence is somewhat justified. But then one day, we miss something and discover that, this time, we’d got it wrong. No big deal, right? We’ve almost always been right about things.

Wrong.

It only takes a single time to seriously undermine our credibility, and rightly so. Then it’s a long road back to regaining it.

Fortunately, it’s entirely avoidable!

We build credibility by being honest about how sure we are, not by always being right (though if we can do that too, we’re on to a winner).

If there’s any uncertainty whatsoever, communicate it. Instead of presenting something as absolute, acknowledge our level of confidence (or lack of it) that we have in it.

That can be as simple as:

“I’m almost certain this is the case, but…”

…followed by why we might be wrong. Or, if we have a good sense for it, express our confidence as a percentage. Estimating certainty is another rabbit hole, and generally humans have poor intuition when it comes to probabilities, but that’s beside the point. The important thing is to communicate that uncertainty exists.

And, pretty much always, there will be some degree of uncertainty. So this should be the default.

There was a line from the TV show Billions that I always liked: it was a euphemism the characters used when they had extremely high confidence: “I am not uncertain.” In their case, it was due to an illicit edge. But for us in engineering, this is because we have the depth of knowledge, the situational awareness, and the experience to know when we’ve crossed this threshold.

Our confidence in these cases carries weight because we’ve demonstrated that we don’t overproject certainty. But we won’t necessarily always have built that credibility. At these times, we shouldn’t hold back or allow ourselves to be ignored. We shouldn’t caveat or soften our language. This is the time we go all in.

Recognising this trait in other people is also important. People who overproject certainty should become a red flag. Look out for this in general conversation as much as at work. For example, misquoting a news article, being challenged on it, then doubling down without hesitation. There are nuances here, of course. But knowing we have a set of people that we can absolutely trust when they are certain allows us to make game-changing decisions unreasonably fast when the scenario arises.

Perhaps we can see a parallel here with LLMs?

Large language models are regularly confidently wrong. But they’re also mostly right, and we interact with them a lot, so the natural balance of trust is hard to get right. Unlike people, the models suffer very little downside to being wrong. And today’s models aren’t consistently able to calibrate their own confidence levels either. So we must get good at understanding confidence in model output… and that is perhaps a good topic for next time.