Enter your email to get Benzinga's ultimate morning update: The PreMarket Activity Newsletter
Artificial intelligence (AI) technology, especially generated text, has become increasingly prevalent in the modern digital landscape. Though AI platforms like ChatGPT have been under development for almost a decade at this point, they have only seen stable releases in the past few months. Despite its recency, ChatGPT already has over 200 million active users, applying its generative AI to simple question and answer prompts to advanced coding solutions. The technology has advanced so far beyond its initial release state that AI-generated text is nearly indistinguishable from human writing, with images, video, and audio following closely behind.
AI-Generated Text and Inauthenticity Online
The rapid propagation of AI-generated text online has revealed the need for a tool capable of identifying authentic or inauthentic content during one's navigation of the digital landscape. Whether creating problems of academic dishonesty, hurting the accuracy of search engines, or potentially spreading misinformation, there are increasing concerns about AI-generated text creating issues of distrust and plagiarism online.
Are AI Detectors a Reliable Solution?
AI no longer exists in a state where its generated content is easily identifiable at a glance, creating the need for a tool which can identify AI-generated content quickly and effectively. AI detectors have emerged as the most reliable tool for detecting AI-generated content in this online era, but you might have seen news questioning their accuracy. Many people wonder: are AI detectors accurate?
How Does an AI Detector Work?
In order to answer that question, it is crucial to first understand how AI detectors work. In truth, AI detection tools rely on the same technology as AI generators, a combination of machine learning (ML) and large language models (LLMs). An AI text generator is trained on existing human writing, enabling the model to accurately predict what word should come after another in a sentence, providing an answer to a given prompt. An AI detector is trained on both human writing and AI-generated text, which the tool uses to identify the slight differences between each type of content.
Fundamentally, an AI detector is looking out for two main factors: perplexity and burstiness. The ultimate goal of the modern AI generator is to provide sentences and paragraphs that communicate information quickly and efficiently, so that its content is uniform and easy to understand. Contrastingly, humans write in a less cohesive manner, making it more difficult to discern what is being said while creating a more individual style. Word choice is tied to perplexity, while sentence length and grammatical construction is tied to burstiness.
False Positives, False Negatives
That said, there are some human writers who might simply aim to communicate information in a straightforward manner that happens to be similar to AI-generated text. Clear communication has its benefits, to say the least, but an AI detector is more likely to pick up on their writing as having the hallmarks of AI-generated text. This is the fundamental shortcoming of AI detectors, based on the way the tools are currently trained.
When an AI detector flags human writing as potentially written by AI, this is called a "false positive." However, this is not necessarily a fundamental failing of the tool. If your writing is consistently identified as AI by an AI detector, it might be a good sign to adjust your writing style. As it stands, AI detectors have become very accurate, but they are in constant competition with advancements in AI generation. The demand for undetectable AI content is on the rise alongside the improvements of AI detectors, further demonstrating the overall reliability of the technology while creating new issues to resolve.
An AI Arms Race
ChatGPT 4o is the most recent iteration of the text generation product, but ChatGPT 5 was recently announced as the next advancement for its rapidly-improving model. The goal of these models is very clearly to increase the perplexity and burstiness of its model, bringing AI more closely into alignment with human writing.
AI detectors do not have the benefit of improving directly alongside AI generators, since they rely on the next generator's output of text in order to train a model, by which time ChatGPT and others will be moving on to the next iteration. When a new version of a product is released, AI detectors will have to play catch up and will temporarily have a harder time distinguishing between human writing and AI-generated text.
An AI Detector Is Still Essential, Despite the Shortcomings
So, while AI detectors are becoming more accurate, improvements to AI generation and the persistence of false positives can detract from the technology's reliability. Though this might be the case, it is still important to add an AI detector to your arsenal of online protections, even if only as a starting point for identifying misinformation.
If an article or other piece of text seems suspicious or too good to be true, it is simple enough to run it by an AI detector and receive the likelihood that something is AI-generated. That said, even if it provides you with a negative or a positive, that result should be treated as a direction forward for identifying the veracity of a given piece of text, rather than the be-all and end-all.
Beyond identifying text you observe online as AI-generated, an AI detection tool can also be used to check human work against receiving a false positive. AI detectors are not infallible, and your writing might happen to err on the side of low perplexity and burstiness. For the student, journalist, or writer, an AI detection tool can be a way to protect against future accusations based on one's writing style by addressing places in their text which was detected as potentially being AI-generated.
Protection Against AI
AI detectors will continue to improve and adapt as AI generators advance forward, ensuring some level of protection against the disinformation created by AI-generated text online. They might not be perfect, but AI detectors are the best way to guard against the unchecked propagation of AI-generated text that exists in the modern day with accuracy and reliability. With experts claiming that nearly 700 million people will be using AI technology by 2030, it is essential to have some means of identifying AI-generated content moving forward.
Image Credit: Pexels
This post was authored by an external contributor and does not represent Benzinga's opinions and has not been edited for content. This content is for informational purposes only and not intended to be investing advice.
Market News and Data brought to you by Benzinga APIs
