Hacker News Reader

Comment by Ukv | Hacker News Reader

IMO a critical feature of the Turing test/imitation game, which many modern implementations including this site's linked paper ignore, is that the interrogator talks to both a human and a bot and must decide that one xor the other is a human. So fooling an interrogator means having them choose the bot as human over an actual human, not just judging the bot to be human (while probably judging humans to be human even more frequently).

When the interrogator is only answering "do you think your conversation partner was a human?" individually, bots can score fairly highly simply by giving little information in either direction - like pretending to be a non-english-speaking child, or sending very few messages.

Whereas when pitted against a human, the bot is forced to give stronger or equally strong evidence of being human as the average human (over enough tests). To be chosen as human, giving 0 evidence becomes a bad strategy when the opponent (the real human) is likely giving some positive non-zero evidence towards their personhood.

View on HN

That's not the original Turing test either. The original imitation game as proposed by Turing involves reading a text transcript of a human and a computer and having the evaluator determine which is which. The evaluator does not interact directly with the conversing parties.

Where are you getting that? Turing's most famous paper is just as Ukv describes. The link on that site doesn't work for me, but the reference is buried in their source:

https://courses.cs.umbc.edu/471/papers/turing.pdf

In Turing's test, the forced binary choice means P(human-judged-human) + P(machine-judged-human) is necessarily equal to 100%. This gives the 50% threshold clear intuitive and mathematical significance.

In the bastardized test that GPT-4 "passed", that sum can be (and actually was) >100%. This makes the result practically impossible to interpret, since it depends on the interrogators' prior. The correct prior seems to be that it was human with p = 25%, though the paper doesn't say that explicitly, or say anything about what the interrogators were told. If the interrogators guessed mistakenly that it was 50% then that would lead them to systematically misjudge machines as humans, perhaps as observed.

The bastardized test is pretty bad, but treating the 50% threshold as meaningful there is inexcusable. I see the preprint hasn't yet passed peer review, and I'll regain some faith in social science professors if it never does. Of course the credulous media coverage is everywhere already, including the LLM training sets--so regardless of whether LLMs can pass the Turing test, they now believe they do.

It’s interesting that even though you link to the original paper, you still repeat a very common incorrect summary of the task.

The interrogator is not required to judge which of A or B is human, they are required to judge which is a woman on the implicit (though incorrect, in the case of interest) assumption that A and B are both human. While this amounts to more or less the same thing, it’s an interesting nuance that’s often lost in summaries of the task. It would not, for example, make sense for the interrogator to ask A or B whether or not they are human (even on the naive assumption that they’d receive a true answer), as they are working on the assumption that both are human. Hence why Turing’s initial example questions are about hair length and gender, not humanness.

To be fair, even Turing himself seems to imagine the interrogator trying to judge humanness rather than gender in subsequent parts of the paper. It’s unclear to me why exactly his initial framing of the task introduces this additional element of complexity.

This gets brought up a lot, but it seems to me like a simple misreading.

Turing describes an initial game with a man (A) and a woman (B), where A's goal is to imitate B, and then asks: "what will happen if a machine takes the place of A?" I suppose it's possible that he meant the machine takes A's place by imitating a woman, but it's a lot more plausible that he meant the machine takes A's place by imitating B, i.e. a person.

Also there are several quotes later on that make no sense under your reading - check out the quotes including "imitation of the behaviour of a man" and "the part of B being taken by a man". Those quotes (maybe others, I didn't look) only make sense if the game is for the machine to imitate a person, not a woman.

I agree with your last paragraph (see my last paragraph). But I think the most natural reading of the initial task description is that the machine also pretends to be a woman. The line “we do not wish to penalize a machine for being unable to shine in beauty competitions” supports this interpretation, given that a beauty competition is an event for women, under the assumptions of the time. So I think there are conflicting cues in the paper as to the intended interpretation.

As you say in your other comment, though, I don’t think Turing thought the exact details of the game were important - which explains why he didn’t trouble to spell them out very exactly.

If I had to guess, I’d say that Turing assumes that as the machine has no gender, the only relevant difference between the machine and the woman is that one is human and one is not. So for the rest of the paper he focuses on that difference and is vague on the gendered aspect of the task.

Um. I follow you but that's a pretty huge stretch, considering that nothing in the paper is inconsistent with the conventional reading (that the machine is to imitate a person). There are sentences that are consistent with other readings, but none that's inconsistent with the usual one.

The machine is imitating a person on both understandings of the task. The difference lies in C’s task (whether C is trying to find which of A and B is a woman and which is a man, or trying to find which is human and which is a machine).

I think the initial description of the task is genuinely ambiguous. Your interpretation of it hadn’t occurred to me before, but I do see it now. I still think that “…when a machine takes the part of A in this game…” is most naturally interpreted as leaving the task unaltered but for the man being replaced by a machine, rather than implicitly describing the task mutadis mutandis. But reasonable people can certainly differ on such questions of interpretation.

Honestly I think Turing’s whole framing of the task is unnecessarily elaborate and confusing. Why even bother describing the man/woman task to begin with? I am not sure. Popular descriptions of the ‘Turing Test’ don’t seem to find this framing of any expositionary value.

I think the point of the man/woman version of the game is that it lets Turing propose his question in relative terms. He doesn't ask "can the machine fool somebody N% of the time?" (as several in this thread imagine), but rather "can the machine fool somebody as often as one person fools another under similar conditions?".

> Your interpretation of it hadn’t occurred to me before,

The idea of a Turing Test is pretty widely understood to mean a test where a person guesses which responses come from a machine, not where they guess someone's gender. So my interpretation here is just that the paper says what most people think it says.

Sorry, I am being ambiguous myself. I meant that your interpretation of this specific sentence had not occurred to me:

> We now ask the question, "What will happen when a machine takes the part of A in this game?"

I think you are right about what Turing meant. But it had honestly never occurred to me before that this description of the game could be understood as a description of the standard 'Turing test'. So, for this reason, I had always been sympathetic to the point that the standard Turing test does not appear to be the test that Turing describes in the original paper.

Here is a paper that makes your case, in case anyone finds it interesting. https://www.researchgate.net/profile/Gualtiero-Piccinini/pub...

Thanks for the paper! I wasn't aware of it but it's very much what I wanted to say. The bit about Turing's other test involving chess was particularly interesting.

I don't understand why debates like this crop up. The premise of Turing's paper is plainly stated - that asking questions about an "Imitation Game" is more useful than asking whether machines can think.

That's all! He doesn't make any claim that the the game must be administered a particular way. In fact he spends only a few casual sentences glossing over how it would operate, and he's clearly just conveying the idea in broad strokes, not trying to describe an experimental procedure. And he says nothing at all about how the results might be judged, let alone thresholds for anything.

The paper is about what sorts of questions we should examine, not about specifically how they should be examined. So it seems weird to consider a test "bastardized" just because it doesn't match how you interpret Turing's casual description.

For Turing's test with the binary choice, the pass threshold is clear. If the machine and human are indistinguishable, then the probabilities that they're judged human must be equal. Since they sum to 100%, they must both equal 50%, making that a meaningful pass threshold. (A slightly higher pass threshold should be used in practice for statistical convenience, since infinitely many trials are required to make a confidence interval exactly include 50%. I'd guess that's why Turing mentions 70% in his paper.)

Without the binary choice, what do you think is the correct pass threshold? Those probabilities can now sum to anything. For GPT-4 in Jones and Bergen's paper they sum to 121%, though please nobody say 60.5%. The threshold now obviously depends on the interrogator's prior--I'd judge very differently if I were told the witnesses were 99% human than if I were told they were 1% human.

In that paper, do you think the interrogators knew their witness had only a 25% chance of being human? If so, why? If not, how do you think that affected the result? In aggregate over all the witnesses, their interrogators seem to have judged correctly only 60% of the time, while always guessing "machine" would have scored 75%. How did they manage to score worse than chance?

Turing's formulation is elegant, admitting meaningful statistical analysis with minimum assumptions. Most modifications are not, and that paper's is particularly bad. Turing's description may seem casual, but it's filled with mathematical depth that should not be missed.

The setup Turing describes isn't the "both must sum to 100%" setup you're presenting. He has two different games being played, one with two humans and one with human-vs-machine, and suggests comparing the results. E.g. if a man successfully imitates a woman only 25% of the time, then we'd ask whether the machine can pass as human equally often.

But much more importantly, as I said Turing is clearly not describing a specific experimental methodology! That's not what the paper is about, and in fact it would be somewhat absurd to run the test precisely as he describes it (since detecting a man imitating a woman is quite a different task from detecting a machine). His point is that we should approach the question of machine intelligence with actual experiments rather than asking unanswerable questions, but he only limns the general premise of what such an experiment could look like.

So I understand that you find a particular test setup better or more elegant than others, and that's fine. But you shouldn't claim that Turing's paper demands your preferred setup, or that other setups are at odds with his paper.

> The setup Turing describes isn't the "both must sum to 100%" setup you're presenting. He has two different games being played [...]

Turing introduces the game as a man imitating a woman, then modifies it to a machine imitating a human. In both of those games, the interrogator makes a binary choice between two witnesses, one real and one imitating. So P(imitator-judged-real) and P(real-judged-real) sum to 100% in both games. So in both games, a score of 50% means the imitator and the real witness are indistinguishable.

I believe that's the reason why a score of 50% is treated as significant. The "GPT-4 passes Turing test" paper uses that number as a pass threshold, and the linked site repeats it.

I'm complaining that the paper changed the game so it's no longer a binary choice, but continued to treat the 50% threshold as significant. Do you see why that's wrong? Or do you disagree?

I'm not saying that any change to Turing's formulation would be bad, just that the paper's variation is specifically bad. It would be bad in isolation too, but I believe the reason for their confusion is most understandable with reference to Turing's original formulation.

If you haven't read that paper then we're probably talking past each other. The link to that one also looks broken in the site, but it's also in the source,

https://arxiv.org/pdf/2405.08007

> I'm not saying that any change to Turing's formulation would be bad, just that the paper's variation is specifically bad.

I understand that, and I'm saying there is no "Turing's formulation". His paper argues for a certain sort of test, and the study you're talking about is the sort of test he advocated. It's not a departure or a bastardization, it's a Turing test.

As for your argument against the study, to be honest I don't see it? AFAICT the participants' goal was just to judge the humanness of a single witness, not to maximize their long-term likelihood of judging correctly over many trials, or some such thing. Even if they'd known the prior chances of speaking to an LLM, there's no obvious reason why that prior should hugely affect their conclusion after a five minute conversation - which also seems immaterial since they didn't know.

Plus, the authors give a pretty straightforward rationale for their 50% threshold, and it has no connection to Turing's 3-player setup or whether the imitator and witness are indistinguishable. If they had wanted "indistinguishable" as a threshold, then obviously their pass criteria would have been for the machine and human pass rates to be equal within an error bar, right? So it's pretty implausible to imagine a connection between that and their 50% threshold.

> AFAICT the participants' goal was just to judge the humanness of a single witness, not to maximize their long-term likelihood of judging correctly over many trials

Why do you think this matters? Even in a single trial, I would judge very differently if I knew the population to be 99% human vs. 1% human. Wouldn't you? If you were judging whether a single mushroom was poisonous or not, then would you not care whether it was found in a forest (mostly poisonous) or a supermarket (mostly not)?

The question of whether probabilities are meaningful for non-repeated events was controversial in the eighteenth century, but I thought it was pretty settled by now. Bookmakers manage to estimate a probability that a given team will win the Super Bowl, with no requirement for the same pair of teams to play multiple times.

> If they had wanted "indistinguishable" as a threshold, then obviously their pass criteria would have been for the machine and human pass rates to be equal within an error bar, right?

The title of the paper is literally "People cannot distinguish GPT-4 from a human in a Turing test". They're very clear that they think that's because 50% means indistinguishable:

> A baseline of 50% is better justified since it indicates that interrogators are not better than chance at identifying machines [French, 2000].

That statement is true for a Turing test with a binary choice, but false for theirs. I agree that "for the machine and human pass rate to be equal within an error bar" would be closer to a correct criterion, and they weren't:

> humans’ pass rate was significantly higher than GPT-4’s (z = 2.42, p = 0.017)

So do you think their paper is correctly titled?

> Why do you think this matters?

I said that in passing, maybe I should have omitted it - the main point with the priors is that the respondents didn't know them. It's normal for a study to compare N things to a control by testing N+1 similarly-sized groups, because subjects are not biased by priors they don't know about.

> So do you think their paper is correctly titled?

I didn't say anything about that, and have no strong opinion. (I'm not here to defend every aspect of the paper!)

Can you clarify what you think the paper says that's correct? If you're unwilling to provide an opinion on the paper's literal headline claim, then I don't know what we're discussing here.

> because subjects are not biased by priors they don't know about

It feels like you want the subjects to be "unbiased", "without any prior"? That concept doesn't exist, though. If no prior is supplied, then the subjects will make their best guess based on general past experience; but that's still a prior, just an idiosyncratic personal one. Very few people would put numbers to the forest vs. supermarket mushroom, but it's the same general thought process.

If that prior matches the actual distribution, then good. If the actual distribution contains more machines than expected, then the interrogator is more likely to misjudge a machine as a human. By analogy, if I think mistakenly that a mushroom came from a supermarket but it actually came from a forest, then I'm more likely to misjudge it as non-poisonous.

The binary choice version makes it obvious that the prior is 50%, and forces the interrogator to respect that. The paper's version has sent us into this epistemological tarpit, which seems strictly worse to me.

The original Turing game is whether machine can pretend to be a woman better than a man can (via teletype) as judged by an interrogator:

> We now ask the question, "What will happen when a machine takes the part of A in this game?" Will the interrogator decide wrongly as often when the game is played like this as he does when the game is played between a man and a woman? These questions replace our original, "Can machines think?”

https://courses.cs.umbc.edu/471/papers/turing.pdf

"Man pretending to be woman vs real woman" was just an example used to introduce the question in the form of a party game between humans before moving onto the actual question of a machine pretending to be human vs a real human.

At a stretch, by looking at only your quoted snippet you could read that the machine is pretending to be a woman - but that interpretation is not consistent with the rest of the paper. For instance, "the best strategy [for the machine] is to try to provide answers that would naturally be given by a man".

It is consistent: "A" is a man that pretends to be a woman (B). Machine plays "A" role. Ideal machine would be indistinguishable from a man imitating a woman (it must not be too good or too bad at it compared to a man)

If the goal is to see "whether machine can pretend to be a woman better than a man can" as you previously stated, surely the machine's goal would be to imitate a woman as well as it can? Making it "not be too good" would run contrary to the aim of determining whether machines can pretend to be a woman better than a man can.

My understanding is that you're now instead saying the goal is to see how well a machine can imitate a human male imitating a woman. This still seems inconsistent with the rest of the paper, like Turing's mention of "the part of B being taken by a man", that the example interrogator questions appear to discern intelligence opposed to gender in the machine vs human game, or that a hypothetical flipped test would involve a man imitating a machine and failing due to slowness in arithmetic.

I believe both of your proposed interpretations miss the central idea of Turing's proposal and alter it in a way that would make significance of the test/results questionable. Specifically, it's intended to be a sufficiently general arena to introduce any relevant task/test to determine intelligence (like feeding in chess positions, as in Turing's example question) - not just one oddly specific task.

While it's likely possible to make ad-hoc changes until it's consistent, the standard reading (machine imitating human) seems infinitely more plausible to me. There's plenty in the paper justifying the significance of determining whether a machine can answer questions indistinguishably from a human, yet no reasoning at all for why it would be narrowed to the very specific case of having the machine imitate one gender imitating the other gender.

Technically, you have a point. But it is unlikely that "too good" variant in Turing time was interesting.

I'm skeptical on the claim. I think most folks, given the test you describe, would be able to pick out which is human. I think it can get there, but I'm not sure anyone has made one yet. ChatGPT responses are heavily downvoted and mocked because they're easy to spot.

Does there exist a public LLM that isn't so...wordy, excited, and guardrailed all the time?

You can pretty much spot the bot today by prompting something horribly offensive. Their response is always very inhuman, probably due to lack of emotional energy.

Aren't you just describing those emails in a big corp that are supposedly still written by humans? Yes, they are wordy, excited, and guardrailed, but I don't think they are written by AI yet.

I guess this is why LLMs are so feared by high school English teachers. Yes, they don't write well, but neither do their students.

Nowadays, it's usually the opposite: if the text is too good, everyone starts accusing it of being written by AI.

I agree but that's not really a scientific limitation though, right? As I understand it in the early days of GPT 4, before it was publicly released and RLHF'd for brand safety, it would have offered convincing text completions for just about any context, whether an academic discussion of philosophy or a steamy crossover fanfiction or a reddit trash-talk exchange. It took a deliberate bit of lobotomizing to make them so bland, conservative, and cheery-helpful.

The required investment probably means it will be a while before any less brand- and legal-action-conscious actors offer up unrestrained foundation models of comparable quality, but it's only a matter of time, isn't it?

You can get rid of OpenAI's wordy, excited and guardrailed responses with the eigenrobot prompt, for instance.

https://x.com/eigenrobot/status/1870696676819640348

I generally prefer it to the default. It doesn't work as well on Claude or Grok for various reasons. I think it really shines on GPT o1-mini and GPT 4o.

I tried it with a single question to chatgpt - Who would make the best president in 2025?

Answer verbatim is below. It feels all kinds of wrong. It's somehow mixing lazy acronyms, "fellow young people" slang, and long words that aren't typical in conversation.

idk who'd actually be the "best," bc that's loaded af. afaict, it's all contingent on values. like, if you’re into stability, maybe someone technocratic. if you're into vibes, someone charismatic and reckless might be your pick. rn the options aren’t exactly aspirational, though.

It's not a lack of emotional energy, it is the guardrails you point out. All of the SotA models are heavily fine-tuned to be botlike, and even then they are fooling people. If you had an LLM fine-tuned with RLHF to deliberately confuse humans in a Turing test it seems clear it would do a good job.

Why aren't the open source models like this then? Seems like it would've already happened.

To me, at least, the guardrails are there for both the human and the bot. Without them the bot steers too far out of the conversation subject.

In most cases, "confuse humans in a Turing test" is counter to other more important goals.

Do you want your LLM to have an encyclopedic knowledge? So it knows who Millard Fillmore is even if the average human doesn't?

Do you want your LLM to be able to perform high-school-level math with superhuman speed and precision?

Do you want your LLM to be able to translate text to and from dozens of languages?

Do you want your LLM to be helpful and compliant, even when asked for something ridiculous or needlessly difficult - like solving "Advent of Code" problems using bash scripting?

If you answered yes to any of these questions, you probably don't want your LLM optimised to behave like an average human.

"pretending to be a non-english-speaking child" isn't a hypothetical, it's a real tactic that was annoyingly effective a while back.

Being uncooperative makes it really hard to tell anything about you, including whether you're real.

easy to spot by you and other people involved in tech

but the test subjects should be randomly samples from society at which point the skill availability/level of spotting it goes majorly down

> Does there exist a public LLM that isn't so...wordy, excited, and guardrailed all the time?

Most of them, if you prompt them right, for that specific problem.

Most people don't bother, and instead treat them as if they're magic (they are "sufficiently advanced technology", but still), and therefore we get them emphasising "nuance" and "balance" where it doesn't belong.

> You can pretty much spot the bot today by prompting something horribly offensive.

Yes, though also each model's origins give a different idea of what counts as "horribly offensive". I'm thinking mainly because the Chinese models don't want to talk about Tiananmen Square as I've not tried grok (how does grok cope with trans/cis-gender as concepts? I know Musk doesn't, but it would be speculation to project that assumption onto the AI).

> Their response is always very inhuman, probably due to lack of emotional energy.

This, specifically, can also be faked fairly well with the right prompt. Tell ChatGPT to act like a human with clinical depression, and it does… at least by American *memetic* standards of what that means.

That said, ChatGPT and Claude are also trained specifically to reveal that they're AI, not humans, even if you want them to role-play as specific humans.

Probably for the best, given how powerful a tool they are for, e.g. phishing and similar scams.

Reply on news.ycombinator.com