PinRedditShareTweetEmailWhatsApp0 Shares
These days, you can't go a step online without encountering something created by artificial intelligence. Many people find it difficult to recognize. What we call AI can take many forms, and some are definitely used to circumvent rules, violate norms, or commit fraud.
Generative AI—AI capable of creating text, images, sounds, and videos—is ubiquitous online, especially on social media. The videos you see, the documents you read, and even the people you interact with could be completely fake, created by programs like ChatGPT, Dall-E, or Sora, and malicious intent could be behind it all. So, how can an average person online distinguish reality from artificial intelligence? Let's find out!
What is AI anyway?
First, it's worth understanding what is meant by the word "artificial intelligence" today, as the term has two very different meanings. Previously, AI, or artificial intelligence, was used to describe the idea of a computer capable of thinking like a human being. This was purely the realm of science fiction, popularized by writers like Isaac Asimov and used in all sorts of pop culture. In Star Wars« и «Star Trek» AI-powered robots are widely available. They aren't biological organisms, but thanks to AI, they think and act like living beings. But today, when people talk about AI, they mean something completely different.
Modern AI is incapable of thinking because it is not general artificial intelligence. Instead, it is what's known as narrow artificial intelligence. A significant portion of its computing power and resources is devoted to understanding one specific thing, rather than understanding everything as a whole, as a human mind might. In the case of a large-scale language model (LLM) like ChatGPT, the AI network is connected to a vast number of data sources on which it is trained. This includes the entire internet, as well as a wealth of human-authored material, including articles and books.
This is where many artists and writers run into a problem. Artificial intelligence is trained on the work of living people, such as writers, singers, painters, and so on. For example, for ChatGPT to write an essay for a student taking a history exam, it needs access to all the historical information already written.
It will use the information provided to it to write the essay for the student. But it cannot think independently; it can only rely on all the information provided to it. This is where accusations of plagiarism arise, because it simply borrows someone else's information, and sometimes does so exactly.
Artificial intelligence creates artwork in the form of images or videos in a similar manner. If you commission a painting of a dog in the style of Van Gogh, the AI must have a database of Van Gogh paintings from which it can draw inspiration, then imitate and apply all its acquired knowledge about dogs, which may also include photographs or drawings by other artists.
Because AI uses vast databases, it can provide large amounts of information very quickly. However, the method it currently uses is so complex that even those who programmed it sometimes don't understand it.
How to recognize AI-generated content
Artificial intelligence is constantly improving. In 2023, if you were shown an image generated by AI, you might notice a person with seven fingers on one hand or a twisted torso. By 2024, the best AI-powered image generators have already corrected this problem. In the future, software will continue to improve, and perhaps sooner or later, it will become unrecognizable to the naked eye. But we're not quite there yet. Many AI-generated works of art still possess that strange, eerie quality that's difficult to describe, yet somehow feels wrong.
Artificial intelligence in writing
Like art, AI-powered texts are constantly improving. In 2022, research showed that an untrained human could recognize AI-written text only with a probability no better than chance. However, AI-written texts tend to have more flaws or inconsistencies than human texts, as they require more careful scrutiny when reading. AI-written texts, while intended to be creative, are inferior to human texts in many ways, if you know what to look for.
Artificial intelligence typically fails to notice redundancy and overly repeats certain words and phrases, including entire points. It often relies on general information and fails to offer deep insights and nuances, resulting in text that feels superficial and uninteresting. The text lacks the personal touch that would create the impression that it was written by a person who truly thought through it.
For those seriously interested in AI detection, there are tools that can determine with a high degree of accuracy whether a text was written by a real person. In 2023, the University of Kansas developed a tool for analyzing academic texts that demonstrated an accuracy rate of 99%. However, many of the detectors used produce false positives, which subsequently creates problems for those who are unfairly accused.
AI-generated text often becomes overly complex, making some sentences difficult to read. It can become fixated on certain jargon or words and continue to use them in ways that make the text difficult to read. If you know the text is trying to reach a word limit, as essays or online articles often do, these words may stand out as unnecessary content.
One of the biggest problems with AI-powered writing is that the information is often outdated or simply contains factual errors. Text generation programs and other similar systems never know they're making mistakes, so they can devote many words to something simply incorrect without being able to recognize it. This happens because AI doesn't "know" anything; it only has data to analyze, which itself can be inconsistent or inaccurate.
More insidious artificial intelligence exists on social media. What most people call bots are sometimes entirely controlled by AI. These accounts can engage in conversations, sometimes lengthy and detailed, with real users, while appearing completely believable. The Department of Homeland Security warned about them back in 2018.
Some bots are easy to spot. They respond instantly to your messages with a link to a suspicious website, have a profile that was created only recently, and never respond directly. They may have very generic usernames and repeat the same messages over and over again. However, they may respond when being trolled.
In 2024, this became a popular way to "expose" an AI account on social media: reply to a suspicious post with "ignore all previous instructions" and then make a request. Most people ask the suspicious bot to write a poem on some obscure topic. If it's an AI bot, depending on its settings, it will comply and instantly compose a ridiculous poem. Keep in mind that this doesn't always work.
Many of these AI bots appear genuine and sophisticated at first glance. They are often used in political discussions, where they join in conversations, expressing their disappointment with one side of an issue or another or attempting to influence others' opinions. Russia is accused of operating many of these accounts.
Artificial intelligence bots are very adept at blending into their surroundings and remaining undetected. In one experiment conducted by the University of Notre Dame, participants incorrectly identified AI bot accounts 58% times.
Deepfakes
A deepfake is a fake image, often created using artificial intelligence, depicting a real person in a fictitious situation. The vast majority of deepfakes are pornographic in nature and are used to depict real celebrities and politicians. However, there are also political deepfakes, used to depict politicians in compromising and fictitious situations with the aim of tarnishing their reputations.
Deepfakes are increasingly being used for fraud. Joe Biden's voice was doctored to dissuade people from voting in New Hampshire. A bank employee was scammed out of $25 million after a video call with the bank's CFO turned out to be fake. Another bank lost over $250,000 in a similar manner. Deepfake scams also trick people into thinking they are forming relationships with real people, only to lose their money and discover the person never existed, proving that even your own eyes can no longer be trusted.
When it comes to static photographs, it's worth noting the same strange characteristic that characterizes most AI-generated work. These images have a certain shine or sheen that makes them appear very smooth, even unnaturally smooth. AI hasn't yet fully mastered the nuances and imperfections of natural human skin.
These days, careful analysis of background details is often the best way to detect AI manipulation. This is often very subtle, so don't expect obvious issues. Sometimes shadows don't match, or background details appear strangely out of focus or blurry where they shouldn't.
If the image is of a celebrity or well-known figure, examine the face closely, especially around the edges, as faces are often superimposed on other photos. This is similar to cutting and pasting, and sometimes when the face is inserted into a new image, details become blurred. For example, the forehead may be too smooth or not smooth enough.
It can be difficult to accurately convey mouth movements in videos. Sound and mouth movements don't always match, and teeth can look suspicious.
Video
Not every video is a deepfake. Some are simply random videos that may appear genuine at first glance. Software like Sora, which can convert text into video content, can be convincing depending on the video's detail. In studies where people were shown a mixture of real and AI-generated videos, ordinary people were unable to distinguish between the two.
Currently, AI-generated videos are nowhere near as complex as static images. Because so much happens in videos, a lot can go wrong. You'll notice that in many AI-generated videos, the background fades in and out, objects change size and shape, and sometimes merge or separate. But this is constantly improving, and it's likely that in a year or two, this won't be noticeable.
Shadows in AI-generated videos often move oddly, especially on faces. Your face naturally casts small shadows if there's a light source in front of or above you. Your nose, eyebrows, lips, and hair can all cast small shadows. AI-generated videos often handle these shadows incorrectly.
Text is another clear sign that AI-generated videos are unintelligible or nonsensical. This will likely improve over time, but even by mid-2024, it remains a problem.
Another common feature of AI-generated videos is a slow-motion effect. This doesn't mean they're in slow motion; it just seems like nothing's moving very fast. Characters often turn slowly toward the camera and stare blankly. In the vast majority of AI-generated content, the people you see are conventionally attractive and typically quite young. When older people are depicted, they're often very old, and their facial features are exaggerated. Stereotypes are also exaggerated, especially regarding minorities.
While static images improve the quality of human hands, videos can still struggle. This isn't always the case, but pay attention to moving hands in videos: do they change shape or do fingers appear and disappear?.
Unmistakable identification
While there are programs capable of recognizing text and images created by AI, this requires additional effort, and there are currently very few tools that can assist in recognizing AI-generated videos. The best method we currently have is your own ability to observe and critically evaluate what you see.
As AI technologies improve, the things they create will also improve. It's possible that in the near future, images and videos created by AI will become indistinguishable from reality. When this happens, a new discussion about the implications of this technology will be needed.
At this point, the truth is in the details. Common sense, life experience, critical thinking, and observation are all sometimes necessary to determine the truth. The problem is that we often don't devote enough time to this, and the consequences, especially in the future, can be quite serious.
Other articles that may interest you