▲ 198 points
back
71 comments
I sometimes wonder if the true "digital divide" comes down to those who were able to develop critical thinking skills prior to these last few years.
If you had previously developed these skills through widish reading and patient consideration, then tools like LLMs are like somebody handing you the keys to a wave-runner in a choppy sea of information.
But for those now having to learn to critically think, I cannot see how they would endure through the struggle of contemplation without habitually reaching for an LLM. The act of holding ambiguity within yourself, where information is often encoded into knowledge, becomes instantly relieve here.
While I feel lucky to have acquired critically skills prior to 2023, tools like LLMs being unconditionally handed to young people for learning fill me with a kind of dread.
I think a lot of doomerism over AI making everyone stupid is really overestimating how many people have good critical thinking skills today. I've worked in high skill engineering domains for nearly twenty years now and even well-accomplished people are often pretty unimpressive when it comes to really understanding complex concepts.
I anticipate smart people will not suffer some handicap simply because easy answers are now readily available for the vast bulk of people who were never going to think critically in the first place.
I think I am relatively smart and I can feel myself getting dumber when I over-rely on AI for certain things. I've had to carefully observe the effect and choose what I "allow" myself to use it for.
Example: when I use it for prose for fiction or my personal blog, I can feel my own imagination for constructing good sentences and my own vocabulary deteriorate much faster than I would've expected. But when I used it purely for brainstorming plot points _or_ doing a quick proofreading pass on a set of notes I already took myself that I want to post, I feel no sudden negative effects.
Feeling that sudden difference in competency is jarring - in a good way. It's very useful information. I wonder whether people growing up with these tools from the beginning won't get that benefit - they won't have a "before AI" and "after AI" brain to "jar" them into awareness and adjustment. Then again... maybe the skills I am trying to preserve will be seen as entirely irrelevant to them regardless.
Ding ding ding.
There is this overly optimistic view that average person is intelligent, which is definitely not the case. AI doesn't make people stupid, it just exposes the existing stupidity.
I agree with you because young me would have learned nothing for the sake of short term fun
Idiocracy is coming.
I'm both making a joke in response, but also believe my response. Mostly viewing life through another highschool teacher (in Australia).
Idiocracy is accelerating*
All connection technology is a force for homogeneity. Television was the death of the regional accent, for example.
Through unlimited amusement, entertainment, and connection we are creating a sad, boring, lonely world.
Curious where you get the idea that regional accents are gone. If you travel around much in the US you'll hear many different regional accents. I have relatives from the west coast, mid-west, south and east coast (we're spread around) and each region has an easily recognizable accent. Some more pronounced than others, but still very much alive.
I've noticed this more in Belgium Flanders.
A region of a postage stamp country compared to the US where accents/dialects often weren't even fully mutually intelligible over distances that weren't so great.
This homogenization was first pushed trough education and such but it remained among the more rural folk (in as far as rural exists in such a densely populated region) but now media and migration are rapidly killing it off.
In my experience I don't notice any difference in accent between the east coast and west coast. The only regional accent I notice in many native english speakers is southern. All other accents seem to be cultural (AAVE, ESL) or dying (older generations have it, younger ones don't).
Social media has already homogenized our thoughts so much. So many facts and perspectives are presented that it's impossible to construct our own opinions on it all without taking inspiration from others, and the upvote button provides a convenient consensus.
What can we do? Would serious efforts to create offline clubs [0] serve as an antidote? This made the rounds on HN [1] recently.
Technology for touching grass.
Yep, connecting with actual human beings is what life's all about.
Geography still dictates some things in the diversity of the experiences it imparts, though admittedly much of our technology exists to insulate us from that stuff.
The problem in the case of AI is who is curating that homogeneity, and to what end. Dynamic systems like IRC and messengers let folks connect and gravitate more “naturally”, while AI - being a walled garden curated by for-profit entities funded by billionaire Capitalists - naturally have a vested interest in forcing a sort of homogeneity that benefits their bottom line and minimizes risk to their business model.
That’s the real threat: reality authoring.
Not sure about that. Billionare Capitalists live in this world too. They might cause harm, sure, but that harm generally takes predictable form and is of finite magnitude.
AI behavior on the other hand can cause under-informed users to do crazy things that no one would ever want. The form of the harm is less predictable, the magnitude isn't limited by anything except the user's ability and skepticism.
Imagine whatever US president you think is least competent talking to ChatGPT. If their conversation ventures into discussion of a Big Red Switch That Ends The World, it's going to eventually advise on all the reasons the button should be pushed, because that's exactly what would happen in the mountains of narrative material the LLM has been trained on.
Hopefully there is no end the world button and even the worst US president isn't going to push it because ChatGPT said it was a good idea. ... But you get the idea, and there absolutely are people leaving their families and doing all manner of crazy stuff because they accidentally prompted the AI into writing fiction starting them and the AI is advising them to live the life of a fictional character.
I think AI doomers have it all wrong, AI risk to the extent it exists isn't from any kind of super-intelligence, it's significantly from super-insanity. The AI doesn't need any kind of super-human persuasion, turns out vastly _sub_-human persuasion is more than enough for many.
Wealthy people abusing a new communications channel to influence the public isn't a new risk, it's a risk as old as time. It's not irrelevant, by any means, but we do have a long history of dealing with it.
It’s only homogenizing your thoughts if you don’t think for yourself.
(I realize this might be a weak point for many people.)
The “do your own research” types end up with some of the biggest groupthink I’ve ever seen though.
Sure, they were probably picking from the top 10 results on Google. Now with AI, we've got one very effective "I'm feeling lucky" button.
Differing individuals with similarly shaped axiomatic structures will discover similar theorems. Some people that are members of ideologies believe they are thinking just for themselves.
It's strong for members of a community to think alike. On the other hand, some people like to search in todash meme space for a useful idea or strategy in the rough. Problem is this treasure hunter strategy is only available to those with resources to try lots of untested and potentially quite harmful ideas.
> It’s only homogenizing your thoughts if you don’t think for yourself.
Uh oh...
More seriously, if you have non-techie (or less techie) friends or family using ChatGPT please ask to see their conversations.
You're likely to be shocked by at a least few of them... many people really don't understand what these tools are and are using them in crazy and damaging ways.
For example, one friends brother in law has his ChatGPT telling him about varrious penny stocks and obscure cryptocurrencies and promising him 10000x returns, which he absolutely believes and is making investments based on.
Other people are allowing ChatGPT to convince them that God has chosen to speak to them through ChatGPT and is commanding them to do all sorts of nonsense.
The commercial LLMs work well enough that people who don't know how they work are frequently bamboozled.
Consider how much your own skepticism of the output comes from cases where it was confidently but objectively wrong and what happens to someone who never uses it on something where objective correctness can be easily judged.
> ... more than fifty students from universities around Boston were split into three groups... According to Nataliya Kosmyna, a research scientist at M.I.T. Media Lab and one of the co-authors of a new working paper documenting the experiment, the results from the analysis showed a dramatic discrepancy: subjects who used ChatGPT demonstrated less brain activity than either of the other groups. The analysis of the L.L.M. users showed fewer widespread connections between different parts of their brains; less alpha connectivity, which is associated with creativity; and less theta connectivity, which is associated with working memory.
Are we still treating low n-size fMRI studies as anything more than a clout-seeking exercise in confirmation bias?
All these hyped up doom titles are homogenising our thoughts. There's some truth to this but it's much more nuanced than presented here.
See... Television? Social Media? Printing Press?
https://slate.com/technology/2010/02/a-history-of-media-tech...
Mass media homogenizes our input (which influences our output). If we want to think about how AI might be different we should consider how it might directly homogenize our output.
You don't think that didn't do it before? If we don't have printing press, the rest of the world is having 8 billions unique conversations. With printing press, we are surely having less? The tiny social groups of 1-3 persons are coalescing forming bigger and bigger groups
https://web.archive.org/web/20121008025245/http://squid314.l...
But he got it wrong-- for most it doesn't need to be better than what they'd do themselves, it doesn't even need to be particularly good.
Plenty of people would prefer to put out AI copy even when they suspect it's worse than what they'd write themselves because they take less personal injury when it turns out to be flawed.
So is social media, TV, Hollywood, and pop culture in general.
Central limit theorem is unrelenting
To be fair, any time vim comes up in a discussion among programmers, without fail someone always reaches for that one exact joke idea about not being able to quit.
Even for supposed critical thinkers, on average we’re not all that original.
I don't think reaching for that joke is a sign of un-originality, it's a signal to that group that you fit in.
The process they described is not much different from gluing together ideas from studies in university (when I was studying for an arts degree).
I think its on the person to realize whether A.I. is becoming a crutch.
We will just have to stand on the shoulders of homogenous giants.
I've felt the same way about our searches and Google Autocomplete. I find myself only search for stuff that Google Autocomplete will recognize.
This headline is infuriating and the content is just a report on that one study that's been making the rounds on Hacker News all week.
AI chat bots enable passive consumption. Passive consumption homogenizes thought. It's not the only technology to do this.
I suspect that The New Yorker, and similar outlets, will stop caring when it becomes financially and socially advantageous to do so.
A culture that is ambivalent or disinterested in providing practical solutions to this problem is the greater issue.
It's just the MIT study again, isn't it? Cool study.
We should wait for the peer reviews before digging too much into it though. These are, after all, preliminary results.
Also bracket peer review. Peer review also has a homogenizing effect on the content in journals. It's not magic.
We are the peer reviews. It's us.
I mean, when I was young I would often think in terms of whatever philosopher I was currently reading. Brains are plastic AF and adopt very quickly. Just keep your inputs more diverse than literally just chatgpt and you will be fine.
In some narrow sense, this headline reminds me of this Alex Murrell blog post from 2023 called "The Age of Average": https://www.alexmurrell.co.uk/articles/the-age-of-average (HN discussion from that time: https://news.ycombinator.com/item?id=35355703). Covers the strange phenomenon of cultural homogenization, with all designs (cars, buildings, brand logos, websites, etc) reverting to a mean. In some sense, I'm not surprised now that AI (or 'thought' more generally) is doing the same.
This makes it sound like people are original and profound. No. 99% of day to day is just repetition. Critical thinking is innate and rare. This is why religion is such a successful social form. This is also why governments like North Korea exist. 99% live brain-dead lives doing uninspiring work…and they are happy to do it. Might also be the way we are designed and wired for survival.
the comments on this post are pretty diverse.
In other news, you can go outside. Roll around in some grass. Explore. Sniff something. Who cares. There’s an infinity of places and beings and things and they’re all unique, and it’s good.
"There were no pessimistic predictions, no scenarios in which A.I. failed or caused harm."
Can "AI" fail. If yes, then what is next for Silicon Valley.
Censorship was doing that already.