- Understanding the Potential Benefits of AI
- Understanding the Potential Risks of AI
- How to Spot AI in the Wild
This chapter is from the book
This chapter is from the book
This chapter is from the book
How to Spot AI in the Wild
I talked earlier in this chapter about how AI can be and is being used today to write text, create images, and more. Unfortunately, that means that AI can be used to write stories that spread lies and propaganda or create images that represent scenes that never took place. The AI-generated content looks so real and reads so naturally that it’s difficult to tell it from the real thing.
How, then, can you keep from being fooled by fake content? How can you identify AI in the wild?
How to Identify AI-Written Text
Many, many companies are using AI to create content for their blogs, social media feeds, and websites. AI content is increasingly popular because it’s free or low cost; companies don’t have to pay a real human writer to write it.
AI-written content can sometimes be good. It can appear well-written. It can be informative.
On the other hand, AI-written content can sometimes be poorly written; it can read just a little “off.” And, worst-case scenario, it can contain incorrect information or even deliberate misinformation.
When examining written content, you find on the web or elsewhere, look for these signs that it may be AI-generated:
Incorrect information: AI systems are often trained on limited data sets that don’t contain precise information. In this scenario, AI might “guess” at an answer and guess incorrectly. In addition, misinformation deliberately spreads falsehoods that are often easily identified.
Controversial opinions presented as facts: While human propagandists can also couch falsehoods as facts, AI systems do it better. Malicious actors are using AI to spread misinformation, so if you see something especially controversial, question its authorship.
Outdated information: AI systems are often trained on data sets that contain older information. References in a given article that are several years old could signal that it may be AI-generated from information that isn’t current.
Repetitive words and phrases: Many AI models have a limited vocabulary and continually reuse those words and phrases they know.
Tone of voice: AI writing tends to be drier and, dare I say, more robotic. Humans tend to write more conversationally and informally.
All that said, today’s most advanced AI models can write text that is difficult to distinguish from human-written text. Because of these advancements, it’s becoming more and more difficult to determine which text is written by a human and which by AI. If you can’t tell the difference between AI and human text, you’re not alone.
Then how can you identify text that was generated by AI? The best piece of advice is to trust your instincts. If a piece of information doesn’t seem right, you should look for another source—ideally one you can trust. Always go with a trusted source over a random article from an unknown website.
How to Identify AI-Generated Photos and Images
AI-generated images are often more dangerous than AI-generated text. It’s the old adage of a picture being worth a thousand words; we tend to believe things we can see with our own eyes—even if those images are manipulated.
Some individuals create AI-generated images for their own amusement. Some create AI-generated images for profit. And some malicious actors create AI-generated images to try to fool people like you and me.
Fake AI-generated images can be used to try to influence public opinion. They can be used to try to influence voters in an election. They can be used just to confuse people and stir up trouble.
Whatever the goal, we collectively need to be aware of these fake images and do our best to identify and avoid them. Given the highly realistic quality of some of these images, however, that can be difficult.
How, then, can you tell an AI-generated image from a real one? Here are some tips:
Too many—or not enough—fingers: For some reason, today’s AI image generators have problems with human hands. Some AI generated images show people with four fingers, some with six, and some, like the one in Figure 2.2, with some bizarre arrangement of fingers that bear no resemblance to reality. Always check the hands and fingers (and other limbs) in a photo; if something’s obviously not right, it’s probably AI generated.
FIGURE 2.2 A not-very-human looking human hand generated by AI. (Image generated by Microsoft Image Creator from the prompt “human hands.”)
Unnatural body proportions and parts: Similar to the hands and fingers thing, AI doesn’t always get body proportions right. Ears might be slightly misplaced, larger, or smaller than they should be. Faces might be oddly asymmetrical. Legs and arms might be thicker or thinner than is physically possible. If it doesn’t look right, it’s probably been poorly constructed or manipulated.
Bad hair: I’m not talking about messy hair or a bad part. AI often has trouble with all the details in human hair. Hair might appear blurred or have unusual changes in texture. It might look wrong on a head, like a bad wig. Even little issues, like looking too thick or too thin, can be a sign of AI manipulation.
Overly rendered appearance: AI will sometimes “over render” the details on a face or other object, making it look unnaturally sharp, especially compared to other elements or the image background. It’s not a natural look.
Weird or missing details: AI doesn’t always get the details right. You can sometimes find surprising errors when you zoom in to a high-resolution image. Look for things in the background that shouldn’t be there, distinct elements blending weirdly into one another, items that don’t quite match up, and other things that don’t make sense. AI’s generated reality doesn’t always reflect our real world.
Unusual backgrounds: AI-generated images sometimes get foreground elements right but do odd things to the background. Look for unusual textures, unnaturally repetitive patterns, or glossy effects. Another telltale sign is a blurry or airbrushed background.
Poor architecture, furniture, and accessories: AI can make little mistakes that really stand out, especially with buildings and rooms. Look for oddly curved walls, sloping ceilings, misaligned steps, and such. A chair might be missing a leg or a coffee table might have too many legs. Consider minor elements in the image, such as coffee mugs, purses, jewelry, and the like; AI often gets the size wrong or has them hanging in mid-air, defying the law of gravity. It’s almost as if these details are an afterthought for the AI models.
Nonsense text: Look for any items with text within the image, such as newspapers, books, and posters. More often than not, AI generates nonsense words and letters for these elements, as shown in Figure 2.3.
FIGURE 2.3 Look closely at the newspaper; those aren’t real words. Also, the woman has three legs and a very odd-looking left hand. (Image generated by PIXLR AI Image Generator from the prompt “woman reading a newspaper”)
)
)
