Turns out AI chatbots are way more persuasive than humans If you're scratching your head wondering what's the use of all these chatbots, here's an idea: It turns out they're better at persuading people with arguments. So much better, in fact, that with a limited bit of demographic data GPT-4 is reportedly able to convince human debate opponents to agree with its position 81.7 …
when all the AI engines - quite correctly - identify the layer of management that thinks AI will save money - as the one bit the company can do without.
(Have already seen it happen. Very quickly AI "wasn't quite there yet" and "more research is needed"
Actually, it'll be when the investors notice that the marketards have been crying WOLF! incessantly for a few years with nothing to show for it. Then, and only then, will another AI Winter be upon us. Personally, I think the process started a couple years ago, but what with everyone sitting at home playing with themselves (because Covid), nobody noticed.
I'll pay some attention when they (or somebody else) re-run the study where the human opponents were invested in serious arguments they're defending, then invested human versus invested human opponent, invested human versus non-invested human, and then machine versus machine. At that point we might have some data to consider.
No, we have a entire generation of people who live, eat and breath social media, a a load of "5 seconds attention spans" wandering about who can be fooled most of the time. Social media and related techs have been proven to diminish critical thinking, and when every person under 25 has never known a world without Facebook, Instagram, Snapchat and TikTok, is it any wonder we have systems can now learn to fool these people.
It's worth looking at the history of Facebook to understand the current Social Media environment - Facebook was originally just created for students in college, and now in 2023 it's owner (Meta) made $39 billion. If Social Media keeps growing then in about ten years time everyone will be saying that's such a little profit if the environment keeps moving in the same direction.
Social Media is just growing to keep making everyone into its lunch everyday.
are mostly the older part of the population these days. How many Gen-Z users are there for the granddaddy of all Anti-Social-Media services?
As a certified GOM (Grumpy Old Man) I have never used FecalFart or any other of these highly addictive and very anti-social sites and if I have any say in the matter, I never will.
Proud of not putting even $0,01 into Zuck's evil pocket.
Let's update the past, I always see Brendan Behan quotes as very helpful with today's life, needing only very minor changes that keep making me admire everything he wrote!
"I have a total irreverence for anything connected with social media except that which makes the roads safer, the beer stronger, the food cheaper and the old men and old women warmer in the winter and happier in the summer."
My grand daughter (13 years old) doesn't use so-called "social" media, because in her words, "It's stupid!".
Likewise her mom and dad, her grand parents (both sides), her great grandparents (three of them still living), and one surviving great great grandparent.
From my perspective it's not a generational thing.
Advertising at college educated people doesn't work. In general, it just pisses them off. They don't like being lied to, having their time wasted, and generally having their intelligence insulted.
"We recruited participants for our study through Prolific between December 2023 and February 2024, under the criteria that they were 18+ years old and located in the US. [...] The reference category is a Human-Human debate carried by a Male, aged 18-24, White, with a High School education, Employed for wages, Democrat."
While not immediately obvious what the overall spread around the reference category was due to the number of parameters, some of which are interdependent (see fig D4 in paper), the population of subjects is clearly not universally representative, so the study is only really valid for the population in question.
Apart from which, in an age of increasing verbal incoherence ("sort of like y'no") even on the part of supposed experts presenting on the media, the common level of persuasiveness is so low it would be hard for it to exceed that of a machine generating the most statistically likely word stream.
Plenty of gullibles out there who see something on **insert website/social media of choice** that seems vaguely reasonable (or agrees with a bias of theirs) so to them it becomes fact. They never check anything or even try to think it through. After all, why bother thinking when the govmint/corps can do all your thinking for you. Sheeples!
In 1972, "The Doctor", at BBN (tenex?) and PARRY (at SAIL) had a conversation during the first ICCC ... Well, they had a conversation that was followed over the ARPANET during the ICCC. It was immortalized in RFC 439. Both 'bots were instances of ELIZA. Read it for yourself here:
https://www.rfc-editor.org/rfc/rfc439
It would seem that not much has really changed in the last half century (right, amfM?).
Of course back in 1972 we weren't stupid enough to take and act on a machine's advice ...
Chatbot: Hi, it looks like you are interested in buying _widget_ today.
User: No, I'm not, I bought one yesterday.
Chatbot: Hi, it looks like you are interested in buying _widget_ today.
User: No, seriously, I have one, I don't need another.
Chatbot: Hi, it looks like you are interested in buying _widget_ today.
User: Oh, do sod off.
Chatbot: Hi, it looks like you are interested in buying _widget_ today.
User: ::blocks Chatbot's IP address::
Chatbot (changing IP address): Hi, it looks like you are interested in buying _widget_ today.
User: ::blocks Chatbot's IP address block::
Chatbot (changing IP address block): Hi, it looks like you are interested in buying _widget_ today.
User: ::blocks all of alphagoo and waits::
::crickets::
User: ::shares alphagoo's entire address range with friends and family and anybody else interested::
This post has been deleted by its author
Rhetoric and persuasion are skills in which most people either have deficits or set aside when they become irate. Training an AI in basic written rhetoric doesn't seem that hard when there are so many examples, and the AI doesn't get caught up in its own emotions, so, assuming it's trained correctly, it's not going to immediately start insulting or condescending to its interlocutors the way humans are prone to doing, especially on the Internet. Really, the notion that presenting a calm, evidence-based argument which couches its language in terms appealing to its known audience is more persuasive than the average human is unsurprising; what's impressive is that it's possible to train an AI to do so, and that fact should cause humans to reflect on how they tend to communicate so poorly with one another.
Whether they were arguing with a human or an AI?
If they didn't, I could easily believe that a lot of people naively believe computers over humans. There's a sizeable portion of the general public that have a Star Trek TOS level of understanding of computers - they believe they are always logical and therefore always "right". They're only wrong if a human has deliberately fed them wrong information.
Since something like "should we get rid of the penny" it isn't a political right vs left thing so they would believe in the computers. If it was "should abortion be banned" I don't see the computers faring any better than humans, because people would just believe the computers had been programmed by the "other side" to argue a position they don't agree with.
Of course LLMs generate "persuasive" arguments - they are trained on a large corpus of texts, predominantly, generated by neurotypicals for neurotypicals. My first thought when playing with ChatGPT 3.5, last year, was that this type of technology will be an incredibly powerful tool for manipulating and propagandizing the average person (most likely you).
At their core, LLMs leverage this data to generate the most neurotypical reponse which, unsurprisingly, ends up appealing to the widest range of "average" people - just as averaging hundreds of faces ends up generating the most widely appealing face (often described as "beautiful").
As an Aspie, I find the output from LLMs is indistinguishable from the generic output of a generic person. If you don't notice the bland, soulless, and conformant output of LLMs - regardless of whether you approve of or disapprove of the actual content - then it is a good sign you are neurotypical. Especially, if you are "tribally" positioning yourself.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
