AI-generated replies

user1977

I've noticed quite a few forum members simply running a question through their generative AI machine of choice and copying and pasting the result (sometimes openly stating where the answer came from, sometimes not, but still fairly obvious - this being an example). Some are obviously post-building spammers but others aren't.

I think the forum ought to have some form of stance on this, as the answers aren't necessarily being sense-checked by the human who is posting it (I appreciate many human-generated answers aren't helpful or accurate either!). Worried we may end up with a bunch of chatbots arguing with each other!

Any thoughts?

Baldytyke88

I do sometimes ask AI questions, it can be very informative.

People ask questions on here, which a simple Google would resolve, some are not skilled at using online tools such as Google and AI, so they are being helped.

Which is better, a persons response or AI, depends on the person?

eskbanker

user1977 said:

I've noticed quite a few forum members simply running a question through their generative AI machine of choice and copying and pasting the result (sometimes openly stating where the answer came from, sometimes not, but still fairly obvious - this being an example). Some are obviously post-building spammers but others aren't.

I think the forum ought to have some form of stance on this, as the answers aren't necessarily being sense-checked by the human who is posting it (I appreciate many human-generated answers aren't helpful or accurate either!). Worried we may end up with a bunch of chatbots arguing with each other!

Any thoughts?

I agree that they grate, especially where unattributed, but how would AI-generated responses be identified?  Currently the process for identifying misleading or unhelpful posts is completely dependent on other members spotting the flaws and reporting rule breaches, and I'm struggling to think how the process for AI responses could be any different, and arguably there shouldn't even be any need for it to be different, i.e. a post should be judged on its accuracy or helpfulness regardless of source.

Pollycat

I think this has been raised before.
I'll see if I can find the thread.

I really dislike AI generated descriptions on eBay.
They are meaningless, uninformative tosh.

Pollycat

https://forums.moneysavingexpert.com/discussion/6524101/ai-posters#latest

I'm not sure this is the one I meant...

Pollycat

And this one
https://forums.moneysavingexpert.com/discussion/6479154/chatbots-on-the-forum#latest

Brie

Maybe someone could explain to me how AI or a chatbot works on this site?  I mean - I know in general what people are talking about when they refer to AI and that they may say to a programme "give me an answer about tulips as Brie might write" or similar.  But a chatbot?  how does that differ from someone who uses a read/write programme?

Sorry if I'm asking about things that are stupidly obvious to others.

user1977

Pollycat said:

And this one
https://forums.moneysavingexpert.com/discussion/6479154/chatbots-on-the-forum#latest

Ah, admittedly I hadn't checked whether it already was mentioned in the rules. I presume that includes AI-generated "answers" and not merely the original posts?

It's relatively benign if a forum member openly says "I don't have a clue but I put it into ChatGPT and this is what it says" but I think more of a concern where they haven't cited the source, and might give the impression that it's been sense-checked by a human. AI engines can provide plausibly fluent answers which are factually nonsense, whereas with human responses there can be more clues about the extent of knowledge or uncertainty behind them.

user1977

Brie said:

Maybe someone could explain to me how AI or a chatbot works on this site?  I mean - I know in general what people are talking about when they refer to AI and that they may say to a programme "give me an answer about tulips as Brie might write" or similar.  But a chatbot?  how does that differ from someone who uses a read/write programme?

Not sure what you mean by a "read/write programme"?

AI doesn't particularly "work on this site", we're talking about forum members copying and pasting things from AI sources.

Brie

user1977 said:

Brie said:

Maybe someone could explain to me how AI or a chatbot works on this site?  I mean - I know in general what people are talking about when they refer to AI and that they may say to a programme "give me an answer about tulips as Brie might write" or similar.  But a chatbot?  how does that differ from someone who uses a read/write programme?

Not sure what you mean by a "read/write programme"?

AI doesn't particularly "work on this site", we're talking about forum members copying and pasting things from AI sources.

apologies for the confusion.  a read/write programme is used by people who may be blind or dyslexic or unable to type.  They talk to the computer and it types the information out.  It will also read out things that have been typed.  So the programme reads things out and writes things down - hence read/write.  

So is that different to a chatbot? 

And what is an AI source?  Is that different than me doing a google search and copying and pasting from that.

DullGreyGuy

Brie said:

user1977 said:

Brie said:

Maybe someone could explain to me how AI or a chatbot works on this site?  I mean - I know in general what people are talking about when they refer to AI and that they may say to a programme "give me an answer about tulips as Brie might write" or similar.  But a chatbot?  how does that differ from someone who uses a read/write programme?

Not sure what you mean by a "read/write programme"?

AI doesn't particularly "work on this site", we're talking about forum members copying and pasting things from AI sources.

apologies for the confusion.  a read/write programme is used by people who may be blind or dyslexic or unable to type.  They talk to the computer and it types the information out.  It will also read out things that have been typed.  So the programme reads things out and writes things down - hence read/write.  

So is that different to a chatbot? 

And what is an AI source?  Is that different than me doing a google search and copying and pasting from that.

You're talking about a Text -> Speech -> Text application or accessibility feature however in that scenario it's still a human that is deciding what is being written. 

AI is different, it reads the question thats being asked and comes up with its own reply rather than taking a dictation. It will be influenced by a number of different sources, prompts in the question etc but in principle the text is uniquely generated. 

Google Search, if we ignore Google Gemini, simply presents back to the user a series of links it thinks is relevant. Someone would then need to select one of those and copy/paste what's on that website which is most likely been written by a person again. 

Google is now including Gemini responses in some searches which is its form of AI, so it presents a summary of responses from the top X results from the search, it will condense down the various responses from multiple sources into a single summary. For somethings it will give you more than one summary as it realises there can be multiple meanings -v- differing opinions. 

They are certainly getting better but as humans have mannerisms and biases such that people can guess that anonymous or differently accredited text comes from the same person so do the AI scripts and so people can differentiate them from a natural person and potentially say which AI was used to generate it. 


Chatbots tend to be less to do with coming up with unique answers but more about understanding natural language so it knows if you ask your banking chatbot "what's my account balance?", "how much dough do I have?" "Can I afford drinks tonight" that these should all be responded to with the account balance but it's done through training rather than having a list of 100,000 different ways to ask the same question. 

Some do have full AI on generating their own responses but there are also groups of people that can go round and exploit these things. Some are just funny and get the bot to slag off their own company, some have been fairly serious, MS's Azure Health Chatbot was found to be able to enable cross tenant breaches before the security flaw was breached.