This sentence has been written by a human or by an artificial intelligence? It is a question that, until recently, we would not have even imagined being able to ask.
Things have changed in recent months, especially since December when OpenAI debuted the now famous ChatGPT.
CES 2023
Increased or annihilated? What if artificial intelligence invades Hollywood
by our correspondent Emanuele Capone
The end of written assignments?
The chatbot that uses the GPT-3.5 linguistic model has been the subject of discussion for some weeks now constant in the tech world but not only: it is also being talked about a lot in the education sector, above all by virtue of the ability of OpenAI’s artificial intelligence to generate credible texts, which may seem written by students, with zero effort. This is despite the accuracy of the information generated by ChatGPT is not always guaranteed, as we also verified in our test. Despite this, Corry Wang, writer and market strategist at Google, wrote on Twitter that “we are seeing the death of real-time written tasks”.
The prediction Wang’s extreme reflects very present concerns in the world of university education, for now especially in the United States. Nature, one of the most famous and respected scientific journals in the world, in early December he launched a survey to ask university professors if their students are using artificial intelligence to write new texts.
But the theme also concerns primary education: the New York Times ran an experiment asking ChatGPT to write essays like a fourth grader. He then submitted the papers to a group of experts, including educators and teachers. And the latter were not always able to identify the text generated by the artificial intelligence. The New York Department of Education it even restricted access to ChatGPT for all students and teachers, across school devices and networks. Underlying, they explained, are concerns about “the impact on learning and the accuracy and security of information.”
The challenge: locate AI-generated text
Understanding whether or not a text has been generated by an artificial intelligence is one important challenge, not only for the world of education. It’s pretty easy to predict that the availability of systems like ChatGPT will also exponentially increase the amount of AI-generated content on the web. As Melissa Heikkilä writes in the MIT Technology Review, “in an already polarized and politically tense online world, these AI tools could further distort the information we consume. Implemented in the real world, in real products, the consequences could be devastating.”
In short, the spread of these tools brings with it the need for ways to understand whether or not a text was generated by an artificial intelligence. A solution could come from OpenAI itself which, like confirmed in a talk by Scott Aaronson in Texas, is working on adding some sort of invisible watermark to generated content: “We want it to be much more difficult to take GPT output and pass it as human. This could be useful to prevent academic plagiarism, but also the mass generation of propaganda, or for impersonate someone’s writing style to indict him. Those are all things you might want to make harder, right?”
Search engine
Microsoft challenges Google: ChatGPT will be integrated into Bing
by Francesco Marino
Now how do you do it?
While waiting for an official solution, there are a number of tools you can use to try to understand if a text is generated by an AI. They are free and work in a relatively simple way: just enter a text to analyze it and understand the probabilities that it was created by a system like ChatGPT.
The latest arrival is GPTZero, created by Edward Tiana Princeton University student: analyzes the text starting from two main characteristics: perplexity (perplexity) and vehemence (burstiness). These indicators measure the level of complexity and randomness of what is written, submitting it to the analysis of an artificial intelligence (GPT-2 in this case): the less the system appreciates the text, the higher the possibility that it was generated by a human being. Open AI Detector also works in a similar waydeveloped and hosted by Hugging Face and also based on GPT-2, which can be considered a sort of ancestor of ChatGPT.
It is a bit different though Detect GPT, which is a Google Chrome extension which, once installed, indicates whether or not the text of the page we are visiting has been generated by artificial intelligence.