Does it understand? Recognition and the question of sentience
P10.orientation.05 · Audience: guest, it-ml, language-pro
Sooner or later everyone asks it, usually after a conversation that felt uncannily like talking to someone: does it actually understand what it is saying? It is a fair question and a hard one, and it deserves better than the two answers usually on offer — "it's just autocomplete" and "it's basically alive". Both are confident, and neither is honest. This module separates two things the question runs together, so you can hold a clear position on the part that is settled and an open mind on the part that is not.
“Does it understand?” is really two questions, and almost every argument about it goes wrong by answering one and thinking it has settled the other.
| The question | What it asks | Can we answer it? |
|---|---|---|
| Recognition | Can it tell that something is wrong, absurd or harmful, and act accordingly? | Yes — and we can test it |
| Sentience | Is there someone in there — an experience of it, felt from the inside? | No — and no one has a test |
Keep them apart and the fog lifts. One is a question about behaviour, which you can probe with examples. The other is a question about inner life, which is old, unresolved, and not made easier by the arrival of machines that talk.
📚 Go Further
The distinction, in plain terms.
- Recognition — the model reliably identifies an absurd or harmful suggestion and explains why, because that norm runs through the text it learned from. Testable behaviour, learned rather than felt.
- Sentience — whether there is any experience accompanying that behaviour. No evidence, no accepted test, nothing in the design requiring it. Open.
- Practical habit: when someone asserts confidently that these systems do or do not have inner lives, notice that both claims outrun the evidence — and that neither is needed to decide whether a given system is safe to use for a given job.
That closes the orientation. You can now say what a model is, how it learned, why it has limits, what kind of novelty it can produce, and what it does and does not mean to say it understands — without mystique in either direction. The next track puts that understanding to work. Step back any time to the P10 pillar overview.
Ask the mentor about this module
Ask a question about this content. The mentor explains and grounds its answer in what you are studying; asking is recorded as a learning signal, not a grade.
Keeping your files on this device
Off by default. The mentor always gets your file; this only decides whether your own copy stays here. Copies live in this browser only - they do not follow you to another device, and clearing site data removes them.
My notes on this module
Loading your notes...