4 tips for spotting deepfakes and other AI-generated images : Life Kit : NPR

New tool explains how AI sees images and why it might mistake an astronaut for a shovel Brown University

Some data is held out from the training data to be used as evaluation data, which tests how accurate the machine learning model is when it is shown new data. The result is a model that can be used in the future with different sets of data. The goal of AI is to create computer models that exhibit “intelligent behaviors” like humans, according to Boris Katz, a principal research scientist and head of the InfoLab Group at CSAIL. This means machines that can recognize a visual scene, understand a text written in natural language, or perform an action in the physical world. This pervasive and powerful form of artificial intelligence is changing every industry.

While they won’t necessarily tell you if the image is fake or not, you’ll be able to see if it’s widely available online and in what context. These are sometimes so powerful that it is hard to tell AI-generated images from actual pictures, such as the ones taken with some of the best camera phones. There are some clues you can look for to identify these and potentially avoid being tricked into thinking you’re looking at a real picture. The current wave of fake images isn’t perfect, however, especially when it comes to depicting people. Generators can struggle with creating realistic hands, teeth and accessories like glasses and jewelry. If an image includes multiple people, there may be even more irregularities.

Media Download

At the end of the day, using a combination of these methods is the best way to work out if you’re looking at an AI-generated image. But it also produced plenty of wrong analysis, making it not much better than a guess. Extra fingers are a sure giveaway, but there’s also something else going on. It could be the angle of the hands or the way the hand is interacting with subjects in the image, but it clearly looks unnatural and not human-like at all. While these anomalies might go away as AI systems improve, we can all still laugh at why the best AI art generators struggle with hands. Take a quick look at how poorly AI renders the human hand, and it’s not hard to see why.

While initially available to select Google Cloud customers, this technology represents a step toward identifying AI-generated content. In addition to SynthID, Google also announced Tuesday the launch of additional AI tools designed for businesses and structural improvements to its computing systems. Those systems are used to produce AI tools, also known as large language models. Last month, Google’s parent Alphabet joined other major technology companies in agreeing to establish watermark tools to help make AI technology safer.

Software like Adobe’s Photoshop and Lightroom, two of the most widely used image editing apps in the photography industry, can automatically embed this data in the form of C2PA-supported Content Credentials, which note how and when an image has been altered. That includes any use of generative AI tools, which could help to identify images that have been falsely doctored. The Coalition for Content Provenance and Authenticity (C2PA) is one of the largest groups trying to address this chaos, alongside the Content Authenticity Initiative (CAI) that Adobe kicked off in 2019. The technical standard they’ve developed uses cryptographic digital signatures to verify the authenticity of digital media, and it’s already been established. But this progress is still frustratingly inaccessible to the everyday folks who stumble across questionable images online. For example, if someone consistently appears with a flat expression in a dimly lit room for an extended period, the AI model might infer that person is experiencing the onset of depression.

A portable light system that can digitize everyday objects

But there are steps you can take to evaluate images and increase the likelihood that you won’t be fooled by a robot. Specifically, it will include information like when the images and similar images were first indexed by Google, where the image may have first appeared online, and where else the image has been seen online. The latter could include things like news media websites or fact-checking sites, which could potentially direct web searchers to learn more about the image in question — including how it may have been used in misinformation campaigns. MIT researchers have developed a new machine-learning technique that can identify which pixels in an image represent the same material, which could help with robotic scene understanding, reports Kyle Wiggers for TechCrunch. “Since an object can be multiple materials as well as colors and other visual aspects, this is a pretty subtle distinction but also an intuitive one,” writes Wiggers. Before the researchers could develop an AI method to learn how to select similar materials, they had to overcome a few hurdles.

But upon further inspection, you can see the contorted sugar jar, warped knuckles, and skin that’s a little too smooth. My title is Senior Features Writer, which is a license to write about absolutely anything if I can connect it to technology (I can). I’ve been at PCMag since 2011 and have covered the surveillance state, vaccination cards, ghost guns, voting, ISIS, art, fashion, film, design, gender bias, and more. You might have seen me on TV talking about these topics or heard me on your commute home on the radio or a podcast. We’ll send you one email a week with content you actually want to read, curated by the Insight team. If everything you know about Taylor Swift suggests she would not endorse Donald Trump for president, then you probably weren’t persuaded by a recent AI-generated image of Swift dressed as Uncle Sam and encouraging voters to support Trump.

From a distance, the image above shows several dogs sitting around a dinner table, but on closer inspection, you realize that some of the dog’s eyes are missing, and other faces simply look like a smudge of paint. Another good place to look is in the comments section, where the author might have mentioned it. In the images above, for example, the complete prompt used to generate the artwork was posted, which proves useful for anyone wanting to experiment with different AI art prompt ideas. Not everyone agrees that you need to disclose the use of AI when posting images, but for those who do choose to, that information will either be in the title or description section of a post. I have 25 years hands-on experience in SEO, evolving along with the search engines by keeping up with the latest …

Deep learning algorithms are helping computers beat humans in other visual formats. Last year, a team of researchers at Queen Mary University London developed a program called Sketch-a-Net, which identifies objects in sketches. The program correctly identified 74.9 percent of the sketches it analyzed, while the humans participating in the study only correctly identified objects in sketches 73.1 percent of the time. While companies are starting to include signals in their image generators, they haven’t started including them in AI tools that generate audio and video at the same scale, so we can’t yet detect those signals and label this content from other companies. While the industry works towards this capability, we’re adding a feature for people to disclose when they share AI-generated video or audio so we can add a label to it.

“We test our own models and try to break them by identifying weaknesses,” Manyika said. “Building AI responsibility means both addressing the risks and maximizing the benefits of people and society.” “SynthID for text watermarking works best when a language model generates longer responses, and in diverse ways — like when it’s prompted to generate an essay, a theater script or variations on an email,” Google wrote in a blog post. No system is perfect, though, and even more robust options like the C2PA standard can only do so much. Image metadata can be easily stripped simply by taking a screenshot, for example — for which there is currently no solution — and its effectiveness is otherwise dictated by how many platforms and products support it.

How to tell if an image is AI-generated

First, no existing dataset contained materials that were labeled finely enough to train their machine-learning model. The researchers rendered their own synthetic dataset of indoor scenes, which included 50,000 images and more than 16,000 materials randomly applied to each object. The definition holds true, according toMikey Shulman, a lecturer at MIT Sloan and head of machine learning at Kensho, which specializes in artificial intelligence for the finance and U.S. intelligence communities. He compared the traditional way of programming computers, or “software 1.0,” to baking, where a recipe calls for precise amounts of ingredients and tells the baker to mix for an exact amount of time.

Digital signatures added to metadata can then show if an image has been changed. SynthID is being released to a limited number of Vertex AI customers using Imagen, one of our latest text-to-image models that uses input text to create photorealistic images. Unfortunately, simply reading and displaying the information in these tags won’t do much to protect people from disinformation. There’s no guarantee that any particular AI software will use them, and even then, metadata tags can be easily removed or edited after the image has been created. Fast forward to the present, and the team has taken their research a step further with MVT.

A new state of the art for unsupervised computer vision

First, check the lighting and the shadows, as AI often struggles with accurately representing these elements. Shadows should align with the light sources and match the shape of the objects casting them. Artificial intelligence is almost everywhere these days, helping people get work done and also helping them write letters, create content, learn new things, and more.

“We had to model the physics of ultrasound and acoustic wave propagation well enough in order to get believable simulated images,” Bell said. “Then we had to take it a step further to train our computer models to use these simulated data to reliably interpret real scans from patients with affected lungs.” Ever since the public release of tools like Dall-E and Midjourney in the past couple of years, the A.I.-generated images they’ve produced have stoked confusion about breaking news, fashion trends and Taylor Swift. See if you can identify which of these images are real people and which are A.I.-generated. Our Community Standards apply to all content posted on our platforms regardless of how it is created. When it comes to harmful content, the most important thing is that we are able to catch it and take action regardless of whether or not it has been generated using AI.

“Even the smartest machines are still blind,” said computer vision expert Fei-Fei Li at a 2015 TED Talk on image recognition. Computers struggle when, say, only part of an object is in the picture – a scenario known as occlusion – and may have trouble telling the difference between an elephant’s head and trunk and a teapot. Similarly, they stumble when distinguishing between a statue of a man on a horse and a real man on a horse, or mistake a toothbrush being held by a baby for a baseball bat. And let’s not forget, we’re just talking about identification of basic everyday objects – cats, dogs, and so on — in images. SynthID uses two deep learning models — for watermarking and identifying — that have been trained together on a diverse set of images.

Spotting AI imagery based on a picture’s image content rather than its accompanying metadata is significantly more difficult and would typically require the use of more AI. This particular report does not indicate whether Google intends to implement such a feature in Google Photos. The project identified interesting trends in model performance — particularly in relation to scaling. Larger models showed considerable improvement on simpler images but made less progress on more challenging images.

Google Search and ads are adopting the C2PA’s authentication standard to flag an image’s origins.

“The user just clicks one pixel and then the model will automatically select all regions that have the same material,” he says. Images for download on the MIT News office website are made available to non-commercial entities, press and the general public under a

Creative Commons Attribution Non-Commercial No Derivatives license. A credit line must be used when reproducing images; if one is not provided

below, credit the images to “MIT.”

Starling Lab verifies “sensitive digital records, such as the documentation of human rights violations, war crimes, and testimony of genocide,” and securely stores verified digital images in decentralized networks so they can’t be tampered with. The lab’s work isn’t user-facing, but its library of projects are a good resource for someone looking to authenticate images of, say, the war in Ukraine, or the presidential transition from Donald Trump to Joe Biden. These tools use computer vision to examine pixel patterns and determine the likelihood of an image being AI-generated. That means, AI detectors aren’t completely foolproof, but it’s a good way for the average person to determine whether an image merits some scrutiny — especially when it’s not immediately obvious. A reverse image search uncovers the truth, but even then, you need to dig deeper. A quick glance seems to confirm that the event is real, but one click reveals that Midjourney “borrowed” the work of a photojournalist to create something similar.

• It also notes that a third of most people’s galleries are made up of similar photos, so this will result in a significant reduction in clutter.
• But if they leave the feature enabled, Google Photos will automatically organize your gallery for you so that multiple photos of the same moment will be hidden behind the top pick of the “stack,” making things tidier.
• Unlike visible watermarks commonly used today, SynthID’s digital watermark is woven directly into the pixel data.
• However, metadata can be manually removed or even lost when files are edited.

If things seem too perfect to be real in an image, there’s a chance they aren’t real. In a filtered online world, it’s hard to discern, but still this Stable Diffusion-created selfie of a fashion influencer gives itself away with skin that puts Facetune to shame. We tend to believe that computers have almost magical powers, that they can figure out the solution to any problem and, with enough data, eventually solve it better than humans can. So investors, customers, and the public can be tricked by outrageous claims and some digital sleight of hand by companies that aspire to do something great but aren’t quite there yet. Although two objects may look similar, they can have different material properties.

Dartmouth researchers report they have developed the first smartphone application that uses artificial intelligence paired with facial-image processing software to reliably detect the onset of depression before the user even knows something is wrong. SynthID contributes to the broad suite of approaches for identifying digital content. One of the most widely used methods of identifying content is through metadata, which provides information such as who created it and when.

So how can skeptical viewers spot images that may have been generated by an artificial intelligence system such as DALL-E, Midjourney or Stable Diffusion? Each AI image generator—and each image from any given generator—varies in how convincing it may be and in what telltale signs might give its algorithm away. For instance, AI systems have historically struggled to mimic human hands and have produced mangled appendages with too many digits. As the technology improves, however, systems such as Midjourney V5 seem to have cracked the problem—at least in some examples. Across the board, experts say that the best images from the best generators are difficult, if not impossible, to distinguish from real images. Chatbots trained on how people converse on Twitter can pick up on offensive and racist language, for example.

The IPTC metadata will allow Google Photos to easily find out if an image is made using an AI generator. That said, soon it will be very easy to identify AI-created images using the Google Photos app. “Unfortunately, for the human eye — and there are studies — it’s about a fifty-fifty chance that a person gets it,” said Anatoly Kvitnitsky, CEO of AI image detection platform AI or Not.

The ol’ reverse image search

This process is repeated throughout the generated text, so a single sentence might contain ten or more adjusted probability scores, and a page could contain hundreds. The final pattern of scores for both the model’s word choices combined with the adjusted probability scores are considered the watermark. And as the text increases in length, SynthID’s robustness and accuracy increases. To create a sequence of coherent ChatGPT text, the model predicts the next most likely token to generate. These predictions are based on the preceding words and the probability scores assigned to each potential token. What remains to be seen is how well it will work at a time when it’s easier than ever to make and distribute AI-generated imagery that can cause harm — from election misinformation to nonconsensual fake nudes of celebrities.

And like the human brain, little is known about the precise nature of those processes. A team at Google Deep Mind developed the tool, called SynthID, ChatGPT App in partnership with Google Research. SynthID can also scan a single image, or the individual frames of a video to detect digital watermarking.

And the company looks forward to adding the system to other Google products and making it available to more individuals and organizations. Watermarks have long been used with paper documents and money as a way to mark them as being real, or authentic. With this method, paper can be held up to a light to see if a watermark exists and the document is authentic.

It’s not bad advice and takes just a moment to disclose in the title or description of a post. The AI or Not web tool lets you drop in an image and quickly check if it was generated using AI. It claims to be able to detect images from the biggest AI art generators; Midjourney, DALL-E, and Stable Diffusion. The problem is, it’s really easy to download the same image without a watermark if you know how to do it, and doing so isn’t against OpenAI’s policy.

The Midjourney-generated images consisted of photorealistic images, paintings and drawings. Midjourney was programmed to recreate some of the paintings used in the real images dataset. Earlier this year, the New York Times tested five tools designed to detect these AI-generated images. The tools analyse the data contained within images—sometimes millions of pixels—and can ai identify pictures search for clues and patterns that can determine their authenticity. The exercise showed positive progress, but also found shortcomings—two tools, for example, thought a fake photo of Elon Musk kissing an android robot was real. You can foun additiona information about ai customer service and artificial intelligence and NLP. Image recognition algorithms compare three-dimensional models and appearances from various perspectives using edge detection.

• Training image recognition systems can be performed in one of three ways — supervised learning, unsupervised learning or self-supervised learning.
• They utilized the prior knowledge of that model by leveraging the visual features it had already learned.
• In July of this year, Meta was forced to change the labeling of AI content on its Facebook and Instagram platforms after a backlash from users who felt the company had incorrectly identified their pictures as using generative AI.
• We’ll continue to learn from how people use our tools in order to improve them.
• While it has many upsides, the consequences of inaccurate, incorrect, and outright fake information floating around on the Internet are becoming more and more dangerous.

Some photos were snapped in cities, but a few were taken in places nowhere near roads or other easily recognizable landmarks. Meta is also working with other companies to develop common standards for identifying AI-generated images through forums like the Partnership on AI (PAI), Clegg added. This year will also see Meta learning more about how users are creating, and sharing AI-generated content and what kind of transparency netizens are finding valuable, the Clegg said. “While ultra-realistic AI images are highly beneficial in fields like advertising, they could lead to chaos if not accurately disclosed in media. That’s why it’s crucial to implement laws ensuring transparency about the origins of such images to maintain public trust and prevent misinformation,” he adds.

To do this, search for the image in the highest-possible resolution and then zoom in on the details. Other images are more difficult, such as those in which the people in the picture are not so well-known, AI expert Henry Ajder told DW. Pictures showing the arrest of politicians like Putin or former US President Donald Trump can be verified fairly quickly by users if they check reputable media sources.