What are you inferring?

Have you noticed how the data centres powering the AI services we use are described as performing “inference”?

More significantly, why does this word matter?

Or asked another way:
What can you infer from the fact AI infers?

When knowledge is static and unchanging, it can be definitive and certain – like a PDF manual or a static FAQ webpage. But when information is dynamic and context-dependent, the answers are inferred rather than absolute – like every response AI generates for you.

Intelligently inferring the answers given.

Which means, it’s combining fragments of different facts into the most plausible response it can construct. Generating answers that are probably right… and therefore possibly wrong. Mostly accurate, occasionally not.

Making this an age where your skilled judgement matters more than ever. Running your own human inference to evaluate whether the machine’s inference seems reliable, or whether further verification might be wise.

Here’s one practical way:

AI often guesses (a.k.a. infers) based only on the training data it already knows – just like we do – rather than spending extra time (and energy) searching to verify that it has inferred correctly. So when it matters, increase the odds of accuracy by adding two small but very powerful words to your prompt:

“Don’t guess.”

Then watch as your AI takes greater care to search and validate before responding. Of course, that same guidance can also help increase the quality of thinking within your own teams.

Reducing inference. Increasing accuracy.

I guess.