Software to unmask the text robot – ChatGPT inventors are testing new text recognition

OpenAI

The San Francisco company developed ChatGPT.

(Photo: AP)

Dusseldorf, New York human or machine? The Open AI development laboratory has released a new program that is designed to detect whether a text was written by a computer. It is designed for products from different manufacturers, it said on Tuesday. The company itself has released ChatGPT, a dialog system that mimics human speech.

It is impossible to reliably recognize all content written by artificial intelligence, Open AI explained in a blog entry. But good classification software can, for example, help identify fraud in schools and universities and automated disinformation campaigns.

However, the detection currently does not work reliably, as the company explained. In test runs, the software correctly identified texts that were written by a computer in 26 percent of the cases. However, nine percent of the texts formulated by humans were incorrectly assigned to a machine.

The company therefore recommends that you do not rely on the assessment of the classification software when rating content. In any case, this is currently only designed for English texts.

Open AI is hoping for feedback on whether such tools can be used usefully. Work on the recognition of computer-generated text will continue – “and we hope that we can provide improved methods in the future”.

Where plagiarism scanners fail

ChatGPT is software based on artificial intelligence that can generate texts on command, from tweets to marketing campaigns to essays for the university. Since its release in November, the system has made a lot of headlines for its capabilities. Microsoft has therefore invested billions in Open AI.

>> Read here: How Open AI’s chatbot works

However, the technology can be abused. In the education sector, for example, there is a widespread concern that the software is being used to cheat. Some schools and universities have therefore already banned the technology.

Plagiarism scanners, such as those used by universities today, only help to a limited extent: The programs only check whether the text or parts of it already appear in other sources. ChatGPT, on the other hand, formulates new texts based on various sources.

Open AI is also talking about using a kind of digital watermark for ChatGPT that would be invisible to the human eye. A special verification software would then signal whether it is an AI text or not.

Programs like ChatGPT could also be used for criminal activities. The company Newsguard, which produces software for detecting disinformation, has had the system produce texts with false reports. And according to IT security researcher With Secure, AI could also compose persuasive spam and phishing messages.

Google is working on its own chatbot

Open AI’s classification software is not the only project of this kind. The company itself had already presented a first, albeit much less powerful, detector in 2019.

A computer science student at Princeton University recently published a program called GPT Zero, which is primarily intended for use in education. This should also be able to identify individual sentences that come from an artificial intelligence. An evaluation is still pending.

>> Read here: What Microsoft plans with ChatGPT

There is a similar project at Stanford University called DetectGPT. According to the research team, it has succeeded in identifying texts from generators with AI in 95 percent of the cases.

The need for such software is likely to continue to grow. In addition to Open AI, other companies are also working on text generators with artificial intelligence, above all Alphabet with its subsidiary Google, which fears a threat to its own business model.

The company has been developing software that can write and speak like a human for years, but has so far refrained from releasing it. Now the Internet company is letting employees test a chatbot that works similarly to ChatGPT, the broadcaster CNBC reported on Wednesday night.

An internal email states that a response to ChatGPT has priority. Google is also experimenting with a question-and-answer version of its Internet search engine.

More: Baidu is developing a competitor for ChatGPT

source site-17