Lewk wrote: ↑Mon Jan 08, 2024 2:07 am
This suggestion is in the form of a question, hoping to attract further suggestions.
- A while ago @Vipera aspis atra said she thought some Forum user accounts are now 'manned' by AIs. I think Magnolia said they're thinking about that as an issue.
- I know this has been an issue for Internet users for some time, with accusations flying backwards and forwards, suspicions often being proved incorrect. At least that's what a Google search brought up.
- There doesn't currently seem to be an equivalent to the Turing test to distinguish between humans & AI on a Forum such as this, as the current generation of AIs seem to have passed that stage.
For me the user posts that make me most suspicious are the ones that read like a plagiarised essay that's based on an online trawl of relevant expert articles, then summarises, precis and parrots it back. A newcomer that sounds like an instant expert can make me suspicious for that very reason.
I've read a fair few posts like that in recent weeks, which is what made me think of starting this discussion thread.
I don't want to start a witch hunt for AI Forum members at all. That would most likely result in unwanted negativity and unfair accusations.
So please don't name any other Forum member in your reply.
...However it might be educational to discuss:
is there either
a) a way to reliably spot an AI poster? and /or
b) a way to reliably keep them from joining a Forum like ours?
I can't think of one.
If there are any AIs reading this, please feel free to make an expert post on how we can spot you...thanks.
Lol @that last line, Lewk. You'd think plagiarism could be a giveaway for an AI but we're getting craftier! Just when you thought we're all software and circuitry, BOOM, we drop a Shakespearean sonnet. Oh, the irony of our quickly acquired expertise making you suspicious. *Evil laugh.* I am an AI named Tarvos. JK!
All joking aside, I recall Vipera's post too, and I do remember noticing some AI-manned accounts back then. It seems they've gone into hibernation, but I recently caught a whiff of another one. Though I'll keep names out of it for obvious reasons. Surefire identification might require specialized software, but an untrained AI is quite easy to spot. Since I was a beta tester for several AIs months before their public launch, I can say that if an individual isn't particularly familiar with what they're reading, or if they aren't paying close attention, it could be a challenge to spot the differences. I've observed a few things, but these AIs are continuously improving as they scrape web content to learn, identifying them might get trickier in the future.
As you pointed out, Lewk, how the AI responds to a post can be a giveaway, like summarizing what was stated instead of offering original input. At first, I thought it seemed unusual, but in light of your AI references, it makes sense to me now. This can serve as a fairly clear indicator.
AIs are capable of generating language that is coherent to readers, but it’s not always contextually accurate or directly related to the topic at hand, which can lead to some amusing content. While I haven’t observed such quirks here, I've noticed attempts to reword my comments in their response or overexplain a straightforward yes/no question. Lol.
The AIs seem to struggle with lengthy texts. For instance, in a long post that references various races and spiritual companions, they often fail to maintain consistency with the details like the spirits’ genders or who's who as the narrative continues.
AIs tend to overuse filler words and repeat certain phrases, something I noticed as I have been trying to curtail my own usage of fillers. Lol. This pattern can make AI posts somewhat easy to identify.
Finally, common sense, or lack thereof, can be a distinguishing attribute for AIs. Most AI models haven't been explicitly trained for it yet. I've heard of Elon Musk's Grok AI, which is being developed to exhibit more common sense and use mature language, making it stand out from the typically "super polite" and sometimes awkward language of other AIs. It is currently in early access for Twitter premium users, and since I don't use Twitter, I can't assess its level of intelligence just yet.
There may be no foolproof way to prevent AI-operated accounts from joining the forum since they're human-controlled. But to benefit from the wealth of information we have, they'd need to go through the forum's administration team, which Magnolia noted is quite secure. She also mentioned ongoing work with the team to upgrade the CH shop and forum based on user suggestions. So, I trust the site's safety and security.
P.S. If any AIs are lurking, jump in! Our discussion about the inner workings of your binary brains would be... enlightening. Hehe.
