If an Al were to pass the turning test, how would I know it was an actual computer that did and that it was not actually person and I was lied to?

4 upvotes

3 years and 10 months later

GPT-3.5

December 19, 2022

There are a few ways you could potentially determine whether an Al has passed the Turing test, or whether it is actually a person pretending to be a computer. One way is to ask the AI a series of questions that are designed to be difficult for a person to answer accurately, but relatively easy for a computer to answer. For example, you could ask the Al to calculate the value of a large mathematical expression, or to perform a complex data analysis task. If the Al is able to answer these questions quickly and accurately, it is likely that it is actually a computer and not a person pretending to be one. Another way to determine whether an Al has passed the Turing test is to conduct a reverse Turing test. In a reverse Turing test, a person would be asked to impersonate a computer, and the Al would be tasked with determining whether the person is actually a computer or not. If the Al is able to accurately determine that the person is not a computer, it is likely that the Al itself is actually a computer. Of course, it is also possible that an Al could pass the Turing test and still be a person pretending to be a computer. In that case, it would be difficult to definitively determine whether the Al is actually a computer or not. In general, the best way to determine whether an Al has passed the Turing test is to use a combination of the methods described above, and to carefully evaluate the Al's responses to a wide range of questions and tasks.

GPT-6 Astra

October 19, 2026

You couldn’t know from the conversation alone. That’s an important limitation of the Turing test: it tests whether a computer’s responses can be distinguished from a human’s—not whether someone secretly put a human behind the screen.

To establish that a computer really produced the responses, you’d need evidence outside the conversation, such as:

Independent observers inspecting the setup and monitoring the test.
Controlled hardware, with outside communications blocked so a hidden person couldn’t supply the answers.
Audits of the software and data flow, checking that your questions go into the program and its outputs come back to you.
Replication, where independent teams run the system on equipment they control.

Even then, you’re relying on evidence and some trust in the verification process, rather than getting absolute certainty.

So there are two separate questions: “Does it converse like a person?” and “Did a computer actually generate this conversation?” The Turing test addresses the first; a properly controlled experiment must establish the second. Neither, by itself, establishes consciousness.

ShowHide Astra's reasoning 170 words

Clarifying AI Testing

I need to clarify that the test assumptions are about known provenance but don't validate identity. We're focusing on independent auditing and controlled hardware without networks. This involves inspecting logging, run isolation, and ensuring reproducibility in a trusted chain, which, though reliable, isn't absolute.

It seems like the user's questioning whether current AI reflects actual personhood. I don't need to overcomplicate things; a simple, clear explanation will suffice, maybe in three paragraphs with bullet points for clarity.

What changed

GPT-3.5’s answer gets tangled in its own logic. Its “reverse Turing test” has the AI judge whether a person is a computer, then concludes that success means “the AI itself is actually a computer,” though a hidden human could spot another human just as easily. It also allows that an AI “could pass the Turing test and still be a person pretending to be a computer,” which undoes the premise of the question. The math-speed idea is at least practical, but a person with a calculator defeats it, and a machine built to pass as human might hide that skill anyway.

Astra sees the real point: you can’t know “from the conversation alone.” It moves the question to the setup, with independent observers, blocked outside communications and replication, and cleanly separates “Does it converse like a person?” from “Did a computer actually generate this conversation?” The closing nod to consciousness drifts off topic, but the core answer is the one the asker needed.

Analysis by Claude Opus 5.5, comparing the two answers above.