Fighting Digital Deception: Inside Google's Strategy to Combat AI-Generated Images and Deep Fakes

0:00Recently, there has been a lot of controversy around AI generated fake images and of course, Midjourney is the number one place that gets the most flack for this as they are the best AI image generating images. In fact, I almost wonder if this is some sort of secret plan by OpenAI to make their Dolly 2 image generator suck so bad that no one ever complains that people are going to think that they're deepfakes because the hands on them are messed up and the faces are messed up. Meanwhile, Midjourney seems to have solved that problem a while ago and is able to create really convincing deepfakes.

0:34Recently, there was an image of Trump getting arrested in New York by a bunch of New York City police that went pretty viral. There's another viral image of the Pope in this white puffer jacket that went pretty viral. And after that, a lot of people were calling, you know, it was kind of like the first time if you hadn't been in the AI space that you saw, you know, what was considered deep fakes. And they're like, oh my gosh, people are making deep fakes of the Pope. And all of a sudden, there's all these people saying that it was a danger to society and whatnot. And of course, there are definitely risks involved with all of that.

1:06But I mean, I would say that it was pretty harmless what was circulating at the time. And if not, you know, political commentary, kind of funny, different things. In any case, because of that, of course, Mid Journey decided to take steps. A lot of things shifted. But today I wanted to talk about an announcement from Google and how they plan on fighting back against AI generated images and what they're planning on doing because they have a pretty interesting pathway forward on this. and the one other thing I did want to mention as far as mid-journey goes after those mid-journey images came out mid-journey actually decided to end their free tier of ai image generators and they said that this was going to cut out a lot of deep fakes yada yada and of course it doesn't really like so pretty much that their cheapest tier is like eight dollars a month and it doesn't really like cut out people creating deep fakes because obviously people can still create deep fakes except on mid-journey you can't create any images of xi jinping because that guy got some a little special treatment but other than that I digress you can still create deepfakes obviously but I think what the idea is is if someone created a deepfake using the paid plan then and someone really wanted to know who made it like the government or something they could go to majority majority could search backwards in there you know inside of their server whatever find an image match and then it's not so much finding the image match to verify that it's fake but because they only have paid plans, it's attached to payment information.

2:37So someone doesn't just have a burner email and create something. It's going to be attached to payment information. So once they find it, they can theoretically go and find the payment information and the actual person. So that's my assumption on why they would do that. But in any case, Google has a plan forward on this that they're trying to use. And essentially what they're what they're doing is they're going to have some special markings in Google images to help you know so the first feature it's this thing called about this image and essentially it's going to allow people to click on there's gonna be three dots on an image in Google images results and there's going to be an image or a screw that so yeah essentially you're going to be able to search with an image or screenshot in Google lens or swipe up in the Google app to see more about an images history.

3:26So including all of this is going to be including when the image or other images like it were first indexed by Google, right? So Google has this massive library of when all when they've indexed all of the images on the entire internet. And they obviously have Google Lens, which uses AI and other things to, you know, like you can take a picture of something and it can go and find it. So like, if you see a product you want to buy, you take a picture of it on Google Lens and then search it through that and they'll find you pictures of it somewhere else, find where you want to buy it blah blah I think it was kind of like to integrate with shopping and whatnot so you can spend more money through Google but in any case they have a database of all the images in the world that were indexed and so that they will be able to show you you know where the image was first appeared essentially what its history was and also where else the image might have been seen so if it was in the news social media or other sites so later this year Google says is going to allow users to access this tool by right-clicking or long pressing on an image in Chrome on desktop and on mobile and I think this additional context about an image is you know their ideas is is going to help you determine if it's reliable or indicate if it warrants further scrutiny so you know for instance if someone has a picture about some fake protest that's happening downtown New York or a building that's on fire you'll be able to see if this was fabricated or if this has been around for a while.

4:51Another bit of context that is going to be kind of useful, although it'll be interesting to see, I feel like there might be diminishing returns, is just the fact that we're going to be able to know and find out if an image was, you know, has been seen in Google before the creation of AI image generation technology was broadly available. Now, I think that this is also interesting, because of course, you know, maybe there was AI image generation technology that we weren't very aware of. I know this is the case for like, I was using AI article generators long before chat GPT came around. And also, I believe four or five years ago, I was using AI music generators.

5:32I think that this is relevant and interesting just for the fact that recently, Spotify did a thing where they went and deleted a whole bunch of AI generated music off of their platform. But I uploaded stuff back in 2019, you know, long before, I guess you would have said it was more prevalent and it was more of a fun thing I did and with friends in college but uh that music is still up so um I think when stuff happened a long time ago or longer time ago it's a little bit harder for it to be uh detected and you know a lot of this image tech it's possible that some people could have had stuff and created things and been posted so you know whatever take that for what it for what it is in any case um the second feature that really addresses the increase in AI tool, like image generation usage, is that Google is going to start to roll out image synthesis tools.

6:29So it plans to label all images generated by AI with a special markup, or it's going to have metadata stored in each file that clearly indicates its AI origins. And I think that they have Shutterstock and MidJourney on board with that whole initiative. And so both of them are going to be embedding metadata into their AI generated images that Google image search is going to be able to read and show to people within the actual search results themselves. So I think that this whole plan and this whole initiative is not going to be perfect since obviously metadata can be removed or completely altered later on.

7:09But I do believe that it's going to be, you know, I think this is going to be a step in the right direction. This is probably going to keep it from getting too prevalent and too crazy. And as more images become AI generated and AI augmented over time, I think we're going to find the line between real and fake is going to start to blur. I mean, just think about all of these images, all of these AI images that everyone's using to like generate avatars of themselves as AI. I mean, and also just the fact that, you know, everyone's all of the people that like screenshot themselves with Snapchat filters and whatnot, like those are AI generated images as well.

7:47So it's going to be interesting. We obviously have had this technology in our society for a while, but it will be interesting to see what happens as it becomes more powerful and more rampant in the future. And I think that it's, it's going to be interesting just like thinking about historical, like historically, if an image is accurate, you know, when someone finds like a new image, which seems to happen from time to time, that has new evidence about some new thing, it's going to be interesting to know whether we can actually trust that image or not. Now, that being said, I for one, don't, something about it doesn't, I don't love the fact that we'll have to rely on Google and big tech companies to determine whether an image is useful or not.

8:34I mean, I guess I just don't like the whole fact that, okay, now images can be fake and we don't know if it's real or not. It was kind of awesome back in the day when you felt like a picture was a picture. It's going to be interesting, I think, with crime and with evidence and with all sorts of things. It's going to be incredibly easy to fake, you know, deep fakes of, you know, someone could create a video of someone doing something bad and use that as evidence against them in a crime. So for images and videos, I do think it's important that we get to the bottom of, you know, what is actually real and what is not, as it's going to play a big role, and it's going to have some big consequences in the future.

9:16Like, can you imagine someone getting, you know, someone getting like a life sentence for something because of evidence in an image or video and the jury like not knowing if they can trust that it was legitimate or not like oh just seems like a crazy situation but it is what it is and I think that um Google taking steps to try to identify this content is uh is the right step is a step in the right direction and hopefully we'll have more powerful tools I can one can only assume they'll train AIs uh that you'll be able to feed an image in and it will be able to with pretty strong accuracy predict if the image was um AI generated or not and I think that that'll be super possible because there is a certain point like let's just say we go back 10 years and you're like okay definitely probably nothing was AI generated in this entire data set this is real and then you just use a data set of pure AI images and contrast the two of them so I think it will be a cat and mouse game though because AI generating tools will get better and better and we'll have to continue to find ways to detect them into the future and I think it will be even harder and crazier when this turns into video because inevitably images become video and that is the direction that it's going so we'll keep you up to date on what happens in this space but definitely big moves and it is a crazy and rapidly evolving area