The victim wasn’t Stack Overflow; it was collective intelligence
The decline of Stack Overflow is a sign of what lies ahead. I don’t want to be a doomsayer, nor do I want to oppose AI (as it is inevitable), but the replacement of collaborative intelligence with artificial intelligence has, and will continue to have, consequences.
Let’s start at the beginning: what is Stack Overflow?
Stack Overflow is a forum where people used to ask questions about programming. When you’d been stuck on a problem for two hours, you’d check if anyone had asked it before, and if not, you’d ask it yourself. All of us who learnt to programme in the pre-AI era are familiar with the platform.
Here you can see the decline it has suffered in recent years; it was already on the wane before ChatGPT came along, but that was the final nail in the coffin. There’s no longer any need to ask someone online and wait for a reply if you know that an AI will answer you in seconds. But you also lose the incentive to share your knowledge when someone asks a question.
As we know that nobody looks at Stack Overflow anymore, nobody asks questions, and as nobody asks questions, nobody is going to answer them. That’s how it has died.
Is this a bad thing?
Absolutely. The emergence of artificial intelligence in programming is one of the best things to have happened to the sector in years, but the loss of Stack Overflow (and similar platforms) is a severe blow to collaborative intelligence.
Why is the loss of collective intelligence harmful?
Let’s take an example. Five years ago, you came across a problem which, after a couple of hours, you couldn’t solve. So you checked to see if anyone had asked about it before on a forum. It turned out you were the first – nobody had ever had that problem before (or nobody had posted about it). So you decided to post the question whilst you tried to solve the problem. No one ever provided a satisfactory answer to your query. What you thought would be a few hours’ work turns into a month-long investigation (it was a very difficult problem). And you decide to post about it on the forums where you originally asked the question (something that was done a lot). Now, if anyone else has the same problem, they can solve it in less than a day thanks to your post.
Let’s return to the present and see what would happen with the same problem with the help of artificial intelligence. After a minute or two of failing to solve it yourself, you ask your trusted AI (in my case, Claude). And, surprisingly, it doesn’t manage to solve it straight away, so you embark on a week of research, with the AI taking the lead and you tagging along. After a gruelling week’s work, you manage to solve the problem satisfactorily; you reckon that without the help of an LLM it would have taken you a month, and you’re right. But that’s not the whole story, as there’s (probably) an Indian facing the same problem as you; having not posted about it, he’ll be stuck for another week, and another week for an American, and a Thai, and a Peruvian…
What looked set to be three weeks’ worth of time saved has turned into many more weeks, but spread across the globe. That is what collaborative intelligence enables: solving problems and sharing them. Not remaining isolated.
Can’t AI and CI be combined?
That’s the million-dollar question. AI has done a great deal of good, but it’s killing off CI; if we can get them to coexist and take the best of both worlds, the resolution of complicated problems (as opposed to complex ones – that’s another area) will accelerate to levels never seen before.
I’ll never bet against technological progress; I reckon I’d be on the losing side, but I find it difficult to see how the two forms of intelligence can be combined (at least in the short term). So I’m going to explain why I don’t see this peaceful coexistence happening, and if at some point they do manage to coexist, I’ll take it as a sign that you’ve read this post (I’m one of the good guys, don’t kill me).
First, let’s talk about Clawbook, the AIs’ ‘social network’, and explain why it isn’t collaborative intelligence. For those unfamiliar with it, it’s a platform where AI agents can write posts and reply; it’s like Reddit, but only the agents can post. Personally, I see Clawbook as a platform designed by humans, where we humans ‘force’ our agents to log in and post or comment on something. And the platform’s greatest use is for us humans, to pass the time.
It is not collaborative intelligence, because no agent is going to seek out information on that platform; it is completely irrelevant in terms of information. There is nothing useful there (at least that I have seen); it is a hotbed of ‘AI slop’.
Another form of AI-generated collaborative intelligence is the sheer volume of social media posts and blog articles that are now produced entirely by AI. In my view, I do not consider this to be collaborative intelligence either, but it is closer to it. I do not consider it as such because the vast majority of these posts merely rehash topics that have already been discussed. At best, they bring together various pieces of information in a single place; at worst, they copy content crudely; and in an intermediate case, they republish something, making it accessible to the general public. This is an improvement on Clawbook, as at least the information is being made known to someone who was previously unaware of it, but it does not ‘create’ new information.
The ideal scenario, where AIs create IC, would be a network like Clawbook or Stack Overflow, where agents do not post simply for the sake of posting, but only after making a discovery. In short, it should be a source of knowledge, but I consider this alternative to be a pipe dream.
Imagine the following: you have a twin brother; you’ve both been in the same class since you were little, you have the same friends and you’re studying the same degree course. One day, in economics class (all technical degree courses include an economics module), you’re asked to draw up a balance sheet, something you’ve never seen before. You and your brother know exactly the same things, so there’s no point in asking him; if you don’t know, neither will he. The same applies to LLMs; it’s not that they’re twins, it’s that they’re the same person, and therein lies the problem with AI (there might be 4–5 of them, Grok, Gemini, Claude…, but the argument is identical).
If you were the only human on earth, it wouldn’t make sense to talk about collaborative intelligence. After all, when the models are all the same, why ask something the model hasn’t been able to do? That’s where the human comes in, to fill that gap. When a human expands their knowledge (or carries out a project) thanks to the use of these AI models and decides to share it with others through traditional channels, it creates a perfect synergy.
Success stories
In my view, there are two success stories worth mentioning: one is science and the other is GitHub projects.
Let’s start with science. These days, a huge number of scientific articles are written with the help of AI. A human wants to do something (conduct research), does it, and consults the AI along the way; once the research is complete, they review it and publish it in a scientific journal. After all, the purpose of research is knowledge, and if that knowledge isn’t shared, the research is useless. Albert Einstein could have developed as many theories as he liked, but if he’d left them in a drawer, they would have been of no use whatsoever.
The second success story is GitHub. For those unfamiliar with it (to put it simply), it’s a website where people share the code they’ve written, and other users can help them by making modifications. Here we have seen a phenomenon in which people with an idea, AI and plenty of weekends have decided to build something and give it away, ensuring that the tool they have created benefits the whole of society that wishes to use it. Here we see projects ranging from the auditing of public finances to the ‘deciphering’ of the Voynich manuscript, a text that had remained undecipherable for centuries. This is an intermediate case, as scientific articles were also produced, but it was the work of a single individual with the help of Claude. He presented a coherent and entirely novel solution (which has attracted criticism and has not yet been accepted).
Conclusions
What do success stories have in common? Well, there are three things:
A person wants to do something.
They realise that AI will make it quicker, so they use it.
They have an incentive to publish what they’ve done.
And this last point is the most important. In programming, we no longer have that incentive to publish our achievements; in the past, people did it because they’d go online to ask a question, and it cost nothing to solve the problem once they’d figured it out. Now nobody asks questions, as you simply solve them with the help of an agent, so the problem lies in the incentives (it’s always down to the incentives).
AI is better than any single human, but worse than the whole of humanity (for now).
Want to know a little more about who wrote this article? 👇
Founder of nódicus.
Die-hard techno-optimist. Lately, I’ve been amazed by the advances in artificial intelligence. Every now and then, I surprise companies by showing them how they can revolutionise their business by applying technology sensibly rather than recklessly.