You are a biochemistry student. The professor gave you a simple homework assignment to talk about five pros and cons of biotechnology. Normally, you have to spend two hours to complete such a small paper. By chance, you learned that artificial intelligence (AI) takes only twenty minutes to generate a paper that can be turned in. What would you do? To use it or not, that's the question.
Reddit user "innovate_rye" chose to use AI, and the professor gave him an "A." He is not alone, there are people who are bigger and stronger, not only using AI to solve their homework, but also earning $100 by helping classmates write homework.
How well are AI papers written?
AI can write short stories. Artist Miao Ying once wrote an animated script in GPT-3: a cockroach fell in love with the AI responsible for monitoring it, and the AI with the appearance of a movie star stole the power stone in the village, so the cockroach began to dig bitcoin to save him. AI is good at writing stories like this, and it's a little funny in its absurdity. Miao Ying said in an interview with Wired: "The part I like about Bitcoin is written by AI, I think because it (Bitcoin) was very popular last year. ”
She doesn't alter AI-generated text. GPT-3 first generates a short story, then she breaks it down into parts and feeds it back to GPT-3, generating more content. She selected the most reasonable content from each variant to compose the script.
It's like making a collage, where AI generates materials and humans are responsible for putting together materials into a meaningful whole. If we use AI to write papers, will it really be as reliable as the rumors?
The same process begins with "What is consciousness?" Is AI that can create art conscious? The Paper reporter used the AI writing tool Rytr to generate a short paper of more than a thousand words. Its structure looks like a look, divided into five parts: "What is consciousness", "Consciousness in philosophy", "Consciousness in science", "What is the AI machine consciousness we have been hearing about?" And "Will machines have consciousness?" What does this mean for humanity? ”。
When we put on our glasses and carefully examine logic and check the facts like a strict teacher, we can still find something wrong with this article.
Like many AI writing tools, Rytr uses GPT-3 and matures on the vast amount of text on the open web, "because of its lack of long-term memory, it cuts and pastes together large amounts of previously read content in a stream-of-consciousness free-association fashion," commented Alley Wurds, author of a book co-authored with GPT-3.
Think of GPT-3 as an old professor with erudite but confused memories, who may have written several articles about "Does AI have consciousness?" In 1950, Turing published the article "Can Machines Think?" , proposed the famous "Turing test", in simple terms, if the machine can disguise itself as a human in conversation, confusing the person it is talking to, then it can be judged that the machine has intelligence; Philosopher John Searle proposed the "Chinese Room" thought experiment in 1980 to refute the idea of strong artificial intelligence represented by the "Turing test".
GPT-3's memory is like a soft dough, facts, words are kneaded together, and it is impossible to distinguish between the top and bottom, so it writes in the blank space on the paper: "Machine consciousness was first proposed by philosopher John Thearl in 1990, when he published the paper Can Machines Think? 》。 ”
Because these words are primitive enough — like the average person conceived to write down a string of text — plagiarism checking tools are often helpless about AI-written articles.
In CNKI, which makes graduates "scared", the short paper we concocted with Rytr was judged to be purely original, with a repetition rate of 0%. For most plagiarism checking tools, it is qualified as a paper. The repetition rate given by the "early detection network" is very high, but the repeated paragraphs marked by it, from a human point of view, are really far-fetched and do not constitute plagiarism.
We also used Rytr to generate a short article entitled "Sexual Harassment in the Metaverse", which ended with four actionable suggestions, such as "creating a safe space for people who are harassed or bullied" and "creating an online community", and it is almost impossible to distinguish whether it is written by a person or an AI. It seems that AI has a way to deal with those broad topics that people talk about a lot.
A moral issue? Ask the big guys
AI writing essays involves a moral question, is it a student who does this to cheat? At the same time, there is also an educational question, can students who rely on AI to write papers learn something?
With these two questions in mind, we interviewed the great educator Confucius, Musk, and one of the most outstanding writers, Shakespeare, through the chatbot website character.ai.
"Shakespeare" said: "The life of a poet is a difficult road, he must have passion, love, experience... But does AI have that? "Learning is a difficult path, it has to experience frustration, patience, before there is a flash of light, students who write papers with AI, will they have these?"