Editors adding buildings through AI tools?

Oh, Ah, Ooh, revelation day, did a test with the plugin on a dozen buildings, took a few attempts and fortunately there’s restores build in and then try again. As for the history, I’ve removed the MwAI source tag since me doesn’t think to give credit for something of low value… map filler.

(did not find link on the edit menu, used cntrl+space and type conflate to get access, to then find the plugin’s panel icon left bottom of screen to open it)

image

Definitely does not pay on a dozen to go thru the procedure… has to be a hefty number.

I’d say it’s fine as long as the building are correct. The way a mapper would do it, not missing anything.

1 Like

Yes - I think that most or all of the complaints are of poor quality work, not the fact that the building shape was detected algorithmically.

Obligatory XKCD

7 Likes

Are you sure they are people?
Does OSM have ‘liveness detection’ on its account creation?

If someone wanted to tackle “AI buildings”…

Trailer Parks would be a fantastic subset, because the homes:

  • Mostly come in “3 basic shapes”.
    • All rectangular.
  • Are extremely dense.
    • Packed very tightly together.
  • Typically come with 1 shed as well.
    • A tiny square/rectangle nearby.

These all combine to create “really boring work” for a human to do, but a specialized algorithm could definitely create short, highly-accurate work of this.


Perhaps a potential workflow might be…

An OSM user draws:

The “trailer park AI” can then come swooping in, looking for “all the small, rectangular buildings” inside.

Then the human can just verify most buildings + say OK!


Note: Since last year, I’ve added more than 3000 trailer homes! :slight_smile:

With the Rapid/MapWithAI stuff, I was able to zip through ~1500 “with just an OK”…

But I still think a specialized tool could make it even better/faster.


Trailers + Trailer Parks (+ AI!!!)

The 3 Basic Shapes

  1. A single-wide trailer
    • Looks like a “straight line” or “thin pipe |” or “cigarette”.
  2. A single-wide trailer with an awning/expansion.
    • Looks like a “thin line with 1 or 2 little rectangles sticking out”.
    • Looks like a “tetris piece” or “toothbrush”.
      • ‘L’- or ‘T’-shaped.
  3. A double-wide trailer
    • Looks like a “fat rectangle”.

Here’s a photo of each kind:

  • 1st is marked.
  • 2nd is a zoomed-in area, where you can more easily see the “separate sheds”.

Personally, when I was manually drawing ALL the buildings… I would then reach an area like this:

  • 1st image has ~70 trailers… times 2 for sheds!
  • 2nd image has ~150 trailers… ~70 sheds!

After doing 5 or 6 entire trailer parks manually, it just took absolutely forever.

(In the time it took to do 1 “small trailer park”, I could finish an entire big map tile of rural/suburban houses… which was much more satisfying!)

… And then I reached this “giant trailer park”… and just gave up:

  • There was 560 trailers + ~500 sheds here!

Later, after I began dabbling in the “AI” tools, I came swooping back in and couldn’t believe how much faster/easier it was. :slight_smile:

Yes, there was still a lot of elbow grease needed, but at least:

  • 50% was offloaded to a “single button press”.
  • 25% was minor adjustments.
    • Correcting awnings + “splitting off” the sheds.

And now, my “manual drawing” time was spent fixing up the remaining 25%! :slight_smile:


Extra Resources

You may be interested in:

I believe ASU also specialized in some “trailer park” mapping projects too. (I don’t have links on hand, but I do remember running across them within the past year.)

4 Likes

if you see bot edits adding bad geometries, they also should be reported to DWG to be blocked

(and in such case it would be still people running them anyway)

2 Likes

Not necessarily - bots can create accounts in the first place.

OSM has seen people trying to automate the process of creating OSM accounts and there are technical things in place to make that more difficult, but in all cases it is people doing this, even if the use automation (a “bot”).

3 Likes

Well, yeah. But even when bots scan the 'net and create accounts (or break into servers or whatever) autonomously, it’s still “people” that have instructed them to do so[1], so I don’t get it why is there a need for that to be explicitly stated?

It might only change with the emergence of AGI, but even then one* could argue that it is the people’s fault to have coded / trained / upbrought Terminator progenitors in such a way…

Or am I missing some nuance?


* that is, provided that any humans survive the resulting robot uprising, and they still wish to argue to point


  1. as no alternative exist currently in this reality ↩︎

Personally, I found the reminder helpful. It’s easy to forget that behind every bot there is a human operator.

I would find that reminder useful if it was easy to find the person responsible for that automated job in order to make them comply with the community guidelines and procedures.

Perhaps I missed something, but do you have any example of an automated editor whose operator is not easy to reach?

I run a bot myself. I ensured that any comment on any of the bot’s changeset will reach my email inbox. I also deliberately did not automated the bot - that is, the job only runs when I run it manually. I have two main reasons for this:

  1. There is some benefit to the manual oversight. I use achavi or OSMCha to review each edit after it’s done - sometimes I don’t examine it as thoroughly as other times, but I still always give a cursory look. If there is some catastrophic failure, I’m likely to catch it. There are also some log messages that require human attention - if the bot ran automatically, I might miss those messages.
  2. If other editors were unable to reach me (e.g. if I am hospitalized or something happens to me), it would be Really Bad™ if the bot kept making edits with no one able to stop it. If it became problematic for whatever reason, the only way to stop it would be to ban the account.

(There is a third reason - I don’t have an always-on machine to run the automation - but that is relatively simple to solve)

I have plans to MAYBE create a solution for the second problem - if anyone comments “STOP” on the bot’s changesets, then it would not make new changesets until I manually instruct it to resume. (The idea is that the same comment would explain the issue with the bot’s edits, and I will respond). But this is not implemented, and would not address point 1 about the benefit of oversight, so I’m in no rush to implement it.

I know that probably most bot operators wouldn’t have the firesight to implement such rules, but I ask again - do you have an example of a bot whose operator cannot be reached? I’m sure there are examples, but are they common or are they the rare exception?

I guess search like this should find plenty of examples – Not everybody who edits the map wants to be found :frowning_face:

Vandalism is an entirely separate issue. This thread is about use of AI to make automated or semi-automated edits - cases where you can assume good faith.

I agree with the fist part, but am I lost why you seem to assume that second parts of the sentence follows from the first?

I for one do not subscribe to the idea that just because someone makes (semi-)automated edits (with AI or without) that it is necessarily done in good faith.
On one extreme of that bell curve some might indeed be fully informed and in good faith, and on other extreme some are obvious vandalism; but most are harder to classify: ranging from laziness, misguided ideas, not checking the quality of the work (or insufficiently checking), not being informed about automated edits and import guidelines policies (like e.g. recent Mozilla AI “helping OSM”), over gamification-gone-wrong (laziness / hunting for HDYC score, etc.) all the way to the more annoying “I am aware the community dislikes it, but I know wrote this really neat piece of AI software which will help OSM and I know better then them unreasonable AI-hater plebs, so I’ll just do it anyway”.

3 Likes

“Assume good faith” is a core principle of collaborative projects like OpenStreetMap and Wikipedia.

Handling bad imports - OpenStreetMap Wiki :

The procedure would be something similar to handling Vandalism. This includes “assume good faith”. In the first instance we must assume that everyone is at least trying to make the data better. As set out in procedure for Vandalism, you should start off with calm discussion rather than angry accusations.

There is a much more detailed article about this principle on Wikipedia: Wikipedia:Assume good faith - Wikipedia

I don’t think I can describe it any better than that, so I won’t attempt to. Hopefully it’s clear why this principle applies here. Yes, even in all the cases you described.

1 Like

Yes, and I absolutely agree with that principle. Did I give impression that I did not? (If anything, I think I more often than not end up being the one still assuming innocence and giving benefit of the doubt, even when most of the crowd is already convinced of the “crime”)

All I am saying is just because we should assume by default that something was done in good faith, does not somehow imply that 100% of them were actually done in good faith (and while certain percentage will remain unknown, some percentage will often proven beyond reasonable doubt that it was not done in good faith – i.e. just look at existing DWG / wiki / discourse blocks)

Yes. What did you mean by this then?

What I wrote was in reply to that. As you seem to agree, we should assume by default that even automated edits are trying to make the data better, even if they fail to do so for whatever reason.

Obviously there would be edits that were not done in good faith, and that falls under vandalism. But all of the examples you gave for “harder to classify” cases are good faith attempts, any way I slice it:

They are all attempting to make the data better, even though some are spectacularly failing at it. In all of these cases, “assuming good faith” would be a correct assumption. Relevant: What good faith is not:

Many people misunderstand Wikipedia’s “assume good faith” policy as meaning “assume another editor performed due diligence” or “assume blind faith” regarding a reference, editor, or content. However, the actual intention is closer to “presume good intent”, […]

That is not to say that these are all acceptable behaviors, especially when repeated after being confronted by other members of the community.

So “assume good faith” is just a starting point.

Did I imply that they were? Or did anyone in this thread?

TL;DR: probably just a misunderstanding. Let’s move this to chat if you still think further discussion is needed after my clarification.


To simplify, to me your statements seemed to convey the idea that everyone tries to improve the map, but they are not always successful”. My take was more like “some people try to improve the map, and others to damage it (or promote their own agenda without regard how it affects the map). We should be nice to the people by default and presume the good faith, and thus not accuse people of wrongdoing unless/until there is enough evidence to prove it, at which time we should drop that assumption of good faith and react to repair and protect the map”.

We are talking of orthogonal ideas, even if both are stemming from the same phrase “assume good faith”.

I disagree here. People intentionally ignoring the rules and going against community consensus to push their own agenda (e.g. that last example of mine) is IMHO providing enough evidence that is not done in good faith, even it they sincerely believe it will “improve the map” in the end. (Yeah, technically good faith would’ve been assumed at the beginning of their hypothetical post, but that assumption would’ve been disproved by the end of it).

e.g. people repeatedly edit-warring by changing the name from Ukrainian to Russian and removing name:uk in Ukraine might think they’re “doing the good thing” and “improving the map” (“Hey we’ve declared that region annexed as part of the Russia now, so now we decide what language may be spoken there, and if you dislike that, speak to this AK47”); yet I’d argue they would not be not acting in good faith in that case. YMMV.

Agreed. I guess it might be explained by that phrase using less then ideal words.

If it were “presume good faith” (i.e. “take for granted as being true in the absence of proof to the contrary”) instead of “assume good faith” (i.e. “To take for granted” - without further implied qualifiers) there would be much less potential for confusion / incorrect understanding. Wikipedia and OSM wiki work around that problem by explicitly noting in their longish articles under which conditions such assumptions may take place (and in which they may not) and thus clarity what they mean exactly, but those conditions/clarifications were missing from your post, so it got different meaning to me.

Looking your further clarifications from you it seems now that you meant to add those qualifiers i.e. parts in the parenthesis: “Assume good faith (but only until the evidence to the contrary becomes evident)”, but forgot to do it (or assumed everyone would also assume existence of some unknown set of unstated conditions as you did).

Yes, in that quote which it replied to seemed to me to imply so - I’ve understood you meant that “you can assume good faith always when AI is used to make automated or semi-automated edits” and not “you can assume good faith only until such time that assumption becomes suspect - which might be quite often” (which is what I meant).

The meaning of those two sentences is (hopefully evidently) quite different.

1 Like

Agreed, seems like we’re on the same page in all ways that matter.

I will just clarify, when I used the phrase “assume good faith” it was with the assumption (ha) that other participants would be familiar with the term in OSM/Wikipedia and understand it’s not unconditional. When you challenged the term, I assumed (ha) you might not be aware of it, so I linked to the relevant articles - in my view, making those caveats clear and making it clear that I knew of them. After that point I don’t really understand your response, but whatever, that’s water under the bridge. Plus it’s quite late in my timezone, perhaps I’ll wake up tomorrow and re-read your comments and they would make perfect sense then.

I’m not sure I agree with your interpretation of “assume” vs “presume”, but that is definitely not a conversation for this thread! I’m deliberately not opening a dictionary before posting this comment because I don’t want to waste both of our energy on such a pointless argument.

1 Like