Reliable detection of AI-generated text is impossible, a new study says

[Alfonso Maruccia]

What just happened? The suffocating hype around generative algorithms and their unhinged proliferation have pushed many people to try and find a reliable solution to the AI-text identification problem. According to a recently published study, said problem is destined to be left unsolved.

While Silicon Valley corporations are tweaking business models around new, ubiquitous buzzwords such as machine learning, ChatpGPT, generative AIs and large language models (LLM), someone is trying to avoid a future where no one will be able to recognize statistically composed texts from those put together by actual human intelligence.

According to a study by five computer scientists from the University of Maryland, however, the future could already be here. The scientists asked themselves: "Can AI-Generated Text be Reliably Detected?" The answer they landed on is that text generated by LLMs cannot be reliably detected in practical scenarios, both from a theoretical and practical standpoint.

The unregulated use of LLMs can lead to "malicious consequences" such as plagiarism, fake news, spamming, etc., the scientists warn, therefore reliable detection of AI-based text would be a critical element to ensure the responsible use of services like ChatGPT and Google's Bard.

The study looked at state-of-the-art LLM detection methods already on the market, showing that a simple "paraphrasing attack" is enough to fool them all. By employing a light word rearrangement of the originally generated text, a smart (or even a malicious) LLM service can "break a whole range of detectors."

Even using watermarking schemes, or neural-network based scanners, it's "empirically" impossible to reliably detect LLM-based text. Worst-case scenario, paraphrasing can bring the accuracy of LLM detection down from a baseline of 97 percent to 57 percent. This means a detector would do no better than a "random classifier" or a coin toss, the scientists noted.

Watermarking algorithms, which put an undetectable signature over the AI-generated text, are completely erased by paraphrasing and they even come with an additional security risk. A malicious (human) actor could "infer hidden watermarking signatures and add them to their generated text," the researchers say, so that the malicious / spam / fake text would be detected as text generated by the LLM.

According to Soheil Feizi, one of the study's authors, we just need to learn to live with the fact that "we may never be able to reliably say if a text is written by a human or an AI."

A possible solution to this fake text-generation mess would be an increased effort in verifying the source of text information. The scientist mentions how social platforms have started to widely verify accounts, which could make spreading AI-based misinformation more difficult.

Permalink to story.

https://www.techspot.com/news/98031-reliable-detection-ai-generated-text-impossible-new-study.html

 

[tellmewhy]

The main difference between humans and AI is that humans can feel pain. The basic motivation of humans is to avoid some form of pain in the future. For example, they build water supply networks to avoid the pain of thirst, they farm and raise livestock to avoid the pain of hunger, they make drugs to avoid the biological and psychological pain of death, they build houses to avoid the pain of external conditions, they make toys, cars, tools, etc., all either directly or indirectly stem from an attempt to avoid biological or psychological pain.

Artificial intelligence, on the other hand, does not have this, which is why it is not self-aware, nor can it understand data outside of the associations that humans have given to that data (through language), precisely because everyone feels the same about pain, and through language they cooperate against the common enemy. AI has no intrinsic motivation to be one way or the other, it simply "swims" quickly (faster than humans) through the borrowed knowledge that humans have structured within language.

My point is that AI will not take the initiative to start producing harmful responses. It must find a human with its own particular motivations and specific intention to produce something negative, which is nothing new. Artificial intelligence, precisely because it has no intrinsic understanding of concepts, only the correlations between them, cannot direct the user, it must be directed by the user (via prompts). AI responses are a bit like quantum mechanics, they only make sense when read by a human, they mean nothing meaningful to the machine precisely because it cannot feel pain.

So there's no need to apply internal filters and detection algorithms, because it's always only about a human who has well known and established freedoms. Just as we don't burn books for what they contain, we shouldn't filter what artificial intelligence says when it's asked.

 

[maxxcool7421]

The main difference between humans and AI is that humans can feel pain. The basic motivation of humans is to avoid some form of pain in the future. For example, they build water supply networks to avoid the pain of thirst, they farm and raise livestock to avoid the pain of hunger, they make drugs to avoid the biological and psychological pain of death, they build houses to avoid the pain of external conditions, they make toys, cars, tools, etc., all either directly or indirectly stem from an attempt to avoid biological or psychological pain.

Artificial intelligence, on the other hand, does not have this, which is why it is not self-aware, nor can it understand data outside of the associations that humans have given to that data (through language), precisely because everyone feels the same about pain, and through language they cooperate against the common enemy. AI has no intrinsic motivation to be one way or the other, it simply "swims" quickly (faster than humans) through the borrowed knowledge that humans have structured within language.

My point is that AI will not take the initiative to start producing harmful responses. It must find a human with its own particular motivations and specific intention to produce something negative, which is nothing new. Artificial intelligence, precisely because it has no intrinsic understanding of concepts, only the correlations between them, cannot direct the user, it must be directed by the user (via prompts). AI responses are a bit like quantum mechanics, they only make sense when read by a human, they mean nothing meaningful to the machine precisely because it cannot feel pain.

So there's no need to apply internal filters and detection algorithms, because it's always only about a human who has well known and established freedoms. Just as we don't burn books for what they contain, we shouldn't filter what artificial intelligence says when it's asked.

""It must find a human with its own particular motivations and specific intention to produce something negative""

we are all going to die .. and the next generation will have absolutely no idea what is real or fake ..

 

[yRaz]

""It must find a human with its own particular motivations and specific intention to produce something negative""

we are all going to die .. and the next generation will have absolutely no idea what is real or fake ..

The world is already fake, is it really such a bad thing that the next generation doesn't need to be aware of that?

 

[kiwigraeme]

The main difference between humans and AI is that humans can feel pain. The basic motivation of humans is to avoid some form of pain in the future. For example, they build water supply networks to avoid the pain of thirst, they farm and raise livestock to avoid the pain of hunger, they make drugs to avoid the biological and psychological pain of death, they build houses to avoid the pain of external conditions, they make toys, cars, tools, etc., all either directly or indirectly stem from an attempt to avoid biological or psychological pain.

Artificial intelligence, on the other hand, does not have this, which is why it is not self-aware, nor can it understand data outside of the associations that humans have given to that data (through language), precisely because everyone feels the same about pain, and through language they cooperate against the common enemy. AI has no intrinsic motivation to be one way or the other, it simply "swims" quickly (faster than humans) through the borrowed knowledge that humans have structured within language.

My point is that AI will not take the initiative to start producing harmful responses. It must find a human with its own particular motivations and specific intention to produce something negative, which is nothing new. Artificial intelligence, precisely because it has no intrinsic understanding of concepts, only the correlations between them, cannot direct the user, it must be directed by the user (via prompts). AI responses are a bit like quantum mechanics, they only make sense when read by a human, they mean nothing meaningful to the machine precisely because it cannot feel pain.

So there's no need to apply internal filters and detection algorithms, because it's always only about a human who has well known and established freedoms. Just as we don't burn books for what they contain, we shouldn't filter what artificial intelligence says when it's asked.

I agree with a lot of what you say - many people studying the brain now consider the whole of the body , senses and nervous system.
The question is if we give AI brains a body with parts and senses - what will happen?
Brains are also strange as they Need to process information - if you take away stimuli and info - the brain will create it's own - hallucinations/dreaming etc.

Going the other way the idea of a brain controlling a spaceship is quite old :- Solar Plexus by James Blish 1941

 

[daffy duck]

I read an excellent article about the current state of AI yesterday. It was very well written and the real schocker was that the entire article was written by ChatGPT after the AI read the transcripts of a video about AI.

I can see how schools and universities are going to have a huge battle with plagiarism going forward. Even if you don't submit the AI written article, if you edit it and add you own touches all the heavy lifting was done for you. Also, you may be struggling to even come with ideas about an essay and will be able to get the AI to at least prod you in the right direction.

 

[ET3D]

It's nice to see that we're passed the "Turing test" moment. The future is exciting.