Theories abound on why this has happened. Was it a case of overzealous moderation? Was it damaging to controversially sack a moderator from the wider Stack Exchange network in 2019? People have been turning to AI with their questions, but would anything like as many have done so if Stack Overflow had been healthier? Or was the management by a for-profit company detrimental?
There are of course other crowd-sourced knowledge bases out there such as Wikipedia and of course OpenStreetMap. Have we got weaknesses or have we avoided them?
What is your concern exactly? That OpenStreetMap maybe get less attention due to AI usage regarding asking questions? That will never affect OSM because it’s a constantly changing environment, tags get created and deprecated all the time, and not all of them documented into the wiki. And sometimes in forum you would find explanation about tags better than in the wiki itself. All this can’t occur without people asking and answering questions.
StackOverflow almost usually answered questions about old stuff that just weren’t easy or impossible to find.
Wikipedia is a bit different story, but I doubt it will end similar to SO due to the AI usage. It’s not just aggregation of information, but a neutral presentation of that info (typically) which LLM tools don’t usually do.
It is possible that AI leads to less activity on the forum though. Just yesterday on the field I was looking for how to tag a specific shop, failed to find the information on the wiki, and ended up asking lumo (ProtonMail’s chatbot) to successfully find the correct tags. Had this not been available I would probably have come back home after and either forgot about the shop or asked on the forum for the proper tags.
I doubt OSM itself will see a decline for the same reasons as StackOverflow, except if we fail to face slop contributions that look and may be intended to be helpful but end up being significantly wrong.
AI often misunderstands OSM tagging formats and fabricates nonsensical tags, even if similar tags exist already. The input of experienced human contributors will always be more valuable than a chatbot.
Of course, ‘I need help’ or ‘How to do this’ questions are the kind of thing that people may turn to chatbots for help - but the core of the forums is discussing tagging in-depth. AI can’t replace those discussions as it’s never mapped anything or seen things with its own eyes.
Yes AI can get it wrong, I’ll annotate my comment by saying that it gave me a couple of tags, I checked them in the wiki and was able to find one that matched what I wanted. I didn’t trust the AI blindly, it’s just a tool that help search, I never trust it without being able to find another trustworthy source.
For better or worse, humans have discovered that it’s more fun to argue with another human than with a chatbot. Therefore forums have a bright future indeed.
On a more serious note, there is a growing concern among publishers of mass media and user-generated content that they’re being “disintermediated” by the RAG “grounding” services that power LLM chatbots. By this they mean that these systems will scrape their sites for content and deliver its essence to users, sometimes for a fee, with little or no attribution. The lack of meaningful attribution results in significantly less traffic that sustains a traditional commercial publishing model and less participation that sustains user-generated content communities – all while the scrapers eat up more time and resources.
These effects are severe for sites that previously relied on search traffic, whereas Google search results for individual elements aren’t a major entry point for OSM contributions. Still, this community is no stranger to the struggle for adequate attribution. Beyond the compulsory fine print, we benefit from developer goodwill for the visibility that brings in a healthy flow of newcomers. When developers bury their OSM acknowledgments or present us as merely a static data source, end users are disempowered, deprived of an opportunity to participate more meaningfully on the Internet.
In other words, these challenges are not exactly new to us. Some community members have taken it upon themselves to be vigilantes for OSM attribution. Meanwhile, if you’ve ever met a GIS professional who has only ever consumed GIS platforms, you know the conversation that goes something like: Oh of course I know OSM! Your company has great data! I didn’t realize you have actual mappers.
We aren’t totally insulated from changes to Internet usage habits. For several years, startups have been touting natural language geoparsing as part of GIS manipulation services or navigation platforms. If this interaction model eventually becomes a basic user expectation, we’d want the viable options to include something open-source and community-developed that’s capable of funneling users into the OSM community, in much the way that some conventional geocoders do today, whether or not it’s powered by an LLM.
As for this forum, the immediate risk is spammers using LLMs to hawk unrelated services after posing as authentic users. Otherwise, the forum is only a collaborative tool for us, not the end goal. If chatbots somehow obviate the need for a forum, or a wiki for that matter, maybe we would spend more time touching grass on field surveys or getting to know each other at mapathons!
This may be a bit besides the point, but the French outdoors community has seen its first (totally irrelevant) contribution made by an AI agent operating on the behalf of an end user who wanted to share their hike.
AI could solve hundreds of mis/incorrect/typo tagging and make it a streetcomplete task for human control.
Could generate highway:area from ways, make em relate each other and not a separate entities and say when differnt areas actually merge or when there is a barrier or height difference.
There is so much to be gained and solve all those unsolvabke issues that haunt us.
I feel OSM logic is stagnated for the sake of ease of mapping.
The search box could finnally start working with typos and other missmatches.
Meantime people still use gmaps as superior tool.
As for forum talk… well speak something new to keep it alive. I often ask AI to tell me what people on the forums think and it looks for all the languages. And really, so much already have been said and people keep arguing the same things and rarelly anyone ever tries to solve the issues with a consensus. People keep thinking their own and dont know how to find a common ground.
So yeah AI is there, use it to make our work more productive and think how OSM can gain from it.
If microsoft’s ai buildings are anything to go by… no thanks. If you want a map filled with hallucinated landuse and incorrect pois, please switch to Google Maps.
meaning you’d rather have no data at all rather than poor resoluttion based mapping?
Along the lines: landuse mapped 12 years ago (ranging from rivers through forests to anything else) is just like a hallucination, coz these days we have got much more high rez photomaps and so all the previous mapping is all crap and so is damn annoying to fix it all up to reality?
Why have height calculated from oblique imagery aided by lidar while you can have nothing and worry not for mistakes as obviously there are no mistakes made from data that is not there as well.
Gues just 1 every 6 people see it through and suprise suprise, AI gues those reactions so as well. Meaning i’d know before i tried to pass that info to the community.
What i was more interested with is inspiring people to look for development. But yeah, can’t hope for much if so few people are interested.
