Niantic, the company behind the extremely popular augmented reality mobile games Pokémon Go and Ingress, announced that it is using data collected by its millions of players to create an AI model that can navigate the physical world.
In a blog post published last week, first spotted by Garbage Day, Niantic says it is building a “Large Geospatial Model.” This name, the company explains, is a direct reference to Large Language Models (LLMs) Like OpenAI’s GPT, which are trained on vast quantities of text scraped from the internet in order to process and produce natural language. Niantic explains that a Large Geospatial Model, or LGM, aims to do the same for the physical world, a technology it says “will enable computers not only to perceive and understand physical spaces, but also to interact with them in new ways, forming a critical component of AR glasses and fields beyond, including robotics, content creation and autonomous systems. As we move from phones to wearable technology linked to the real world, spatial intelligence will become the world’s future operating system.”
By training an AI model on millions of geolocated images from around the world, the model will be able to predict its immediate environment in the same way an LLM is able to produce coherent and convincing sentences by statistically determining what word is likely to follow another.
I’ll copypaste an interesting comment here:
[Stephen Smith] This article is a great example of a trend I don’t think companies realize they’ve started yet: They have killed the golden goose of user-generated content for short-term profit. // Who would willingly contribute to a modern-day YouTube, Reddit, StackOverflow, or Twitter knowing that they are just feeding the robots that will one day replace them?
You don’t even need robots replacing humans, or people believing so. All you need is people feeling that you’re profiting at their expense.
Also obligatory “If you’re not paying for the product, then you are the product”.
I’ve found myself thinking “well, you just helped teach the AI about that one…” various times when reading content online.
It’s a strange thing to know that a form of the basilisk is real. Things posted will help AI get better, if only my teeny tiny increments each time.
AI learning isn’t the issue, its not something we will be able to put a lid on either way. Either it destroys or saves the world. It doesn’t need to learn much to do so besides evolving actual self-agency and sovereign thought.
What is a huge issue is the secretive non-consentual mining of peoples identity and expressions.
And then acting all normal about It.
Thing is, consider Google maps. It’s been harvesting data secretly and openly for a long time. I vaguely remember a time when Street View cars were found to be harvesting WiFi information in Australia and their response was, “oops, our engineers made a mistake.” Yeah, right.
But, Google maps is an amazing tool. All that traffic info? All those time estimates? Maybe it’s worth it. Maybe if people knew what they were providing, and the result they’d get, they’d still be happy to give all that “free” data to Google.
Putting aside the ethics of a company taking (stealing? or shall we call it, pirating?) all the ownership of that knowledge asset, if they make a really useful tool from it perhaps Pokémon players will be glad to have been part of such an epic achievement.
The only down side, IMO, is that the models are proprietary and closed.
Also Obligatory: “You can pay for something and still be the product”
I think people will still “contribute” because they also don’t care that their use of certain platforms leaks data used to target ads at them.
In the same vein though, once AI essentially destroys a site like Stack Overflow, where will AI companies source new training data with updated information? Also, we are likely to see something like 50% of content being AI generated. Are AI models then going to train on the content they themselves created? What is the impact of that? What is the use?
Are AI models then going to train on the content they themselves created? What is the impact of that?
It leads to model collapse. The second AI starts to focuses on certain patterns in the output of the first AI instead of the actual content and you get degraded output. They are pattern matching machines after all. Repeat the cycle a few times and all output becomes gibberish. Think of it as data incest.
So the AI companies are pretty desperate for more fresh user data. More data is the only way they have currently to push through the diminishing returns.
What are platforms going to do about it? Start to demonetize AI generated videos and ban AI written fan fiction?
Thanks for sharing the Stephen Smith quote, I had not made that connection yet.
This Pokémon Go player has unwittingly poisoned an AI dataset by spoofing across bodies of water for years.
Gunna have cars attempt to fly!
Ever wonder why websites that use Captchas prefer pictures of cars, busses, crosswalks, stop signs, bicycles, motorcycles and stairs?
They’re using YOU to train their AI models.
No you pretty much have this wrong in fundamental way.
These websites are not using you to train their models. Google is using these websites to use you to train Google’s models.
same way an LLM is able to produce coherent and convincing sentences by statistically determining what word is likely to follow another
To me this implies that the navigation AI is going to hallucinate parts of its model of the world, because it’s basing that model on what’s statically the most likely to be there as opposed to what’s actually there. What could go wrong?
AI: Dave, turn right and walk across the bridge.
Dave : But AI, there is no bridge
AI: I am 99% sure based on 99 billion images that there should be a bridge
Dave: ok , you’re the smart one
Dave: aaaargh . . . .
SPLAT
Reminds me of Morrowind’s directions, with the frequent east-west mixups, and sporadic north-south mixups.
Fun fact, I worked with several other people on a localization patch for polish version of Morrowind, and we had so many of those east-west mixups fixed. Of course the publisher just translated strings and didn’t QA anything.
I presume the idea is to generate a base idea with ai then correct it with real time data.
Like the way go AI has one part to make a ‘policy’ of moves and a second part to simulate (‘read’) the results of those moves many steps ahead.
It’s only going to hallucinate until it gets new input from reality. Not nearly as precarious as generative models.
Its 2030. your state of the art AR glasses have a bug. mr. mime is lurking behind every corner… always watching
It’s 2024 how do I get rid of Mr. Mime in my peripheral vision when I’m not wearing glasses?
From all the apps invading your privacy and abusing your data, I didn’t suspect Pokemon GO to be one of them.
This should be so extremely illegal that it should bring criminal charges to all the members of their board.
From all the apps invading your privacy and abusing your data, I didn’t suspect Pokemon GO to be one of them.
I can’t tell if you’re being sarcastic.
Even if you never played the game, it’s fairly common knowledge that it uses GPS data to place in-game elements and to track where players are.
The game also uses real-world locations as in-game “treasure chests”, which people were theorizing all the way back in 2016 would eventually become open to “sponsored” locations. (Every McDonalds where I am is now a PokeStop)
And if you’ve played the game, you’ve likely seen all the invitations to turn on your camera and submit photos (which are tied to your GPS), move to specific locations, walk (or create) walking routes, take short videos of landmarks, etcetcetc.
I’ve been playing on and off since 2016, and I’ve known I’ve been trading data in exchange for a low-cost game this whole time.
Hu? Really? I thought that was known even when pkm go wasn’t released and only ingress existed.
Niantic is a google split up after all, if they were not collecting data, I would have been very surprised
It’s almost like listening to my crazy rants predicts the future.
Hope you guys don’t have those loyalty rewards cards to grocery stores or pharmacies. Oh, who am I kidding? All of you do.
Hope you guys don’t have those loyalty rewards cards to grocery stores or pharmacies. Oh, who am I kidding? All of you do.
Does it count if they’re all just copies of someone else’s cards?
I mean, good luck shopping without them. All shops artificially inflate the prices without them and then act like you’re getting a huge discount. For example, Tesco, as much as 100% price increase without their loyalty card, and most products have some. At least a 25% price increase.Jenny’s number: (area code) 867-5309
Of course it probably doesn’t matter if you also use a credit card to make the purchase - every single purchase is fed into your personal consumer profile.
And this software will probably be able to route soumeone from one special Pokemon point to the other. Wow. There are three of them in our town. It will be very smart in speedrunning that triangle.
They have added tasks that make you photograph your surroundings or Objects and give them real world lidarr data linked with geo data for some in-game benefits, last I checked.
OK, that is actually something usable. So far what they could learn from here is how to take a shortcut through the fields ;-)
Now let’s wait and see how google trains Earth 2 AI with their streetview data. We will be able to hallucinate places too just like that AI Minecraft project.
Lol, Niantic coding anything that actually works well is hilarious.
I’d also argue that Ingress players likely gave them way more useful data than PoGo
I considered trying to make a mini version of this to auto-contribute to OSM. Street view image shows a compacted dirt road? Submit to OSM. Two lanes with lines? Submit to OSM.
So slavery, then. No surprise.
Bro…
