{"id":541,"date":"2024-09-07T11:00:00","date_gmt":"2024-09-07T11:00:00","guid":{"rendered":"http:\/\/washnow.me\/?p=541"},"modified":"2024-09-13T14:21:28","modified_gmt":"2024-09-13T14:21:28","slug":"theres-a-fix-for-ai-generated-essays-why-arent-we-using-it","status":"publish","type":"post","link":"http:\/\/washnow.me\/index.php\/2024\/09\/07\/theres-a-fix-for-ai-generated-essays-why-arent-we-using-it\/","title":{"rendered":"There\u2019s a fix for AI-generated essays. Why aren\u2019t we using it?"},"content":{"rendered":"
\n

<img alt="A computer screen full of lines of code." data-caption="The language program developed by the US company OpenAI uses artificial intelligence to write a random binary code consisting of zeros and ones. | Frank Rumpenhorst\/picture alliance via Getty Images<\/span>” data-portal-copyright=”Frank Rumpenhorst\/picture alliance via Getty Images<\/span>” data-has-syndication-rights=”1″ src=”https:\/\/platform.vox.com\/wp-content\/uploads\/sites\/2\/2024\/09\/gettyimages-2165204307_8b9ee8.jpg?quality=90&strip=all&crop=0,0,100,100″ \/>

The language program developed by the US company OpenAI uses artificial intelligence to write a random binary code consisting of zeros and ones. | Frank Rumpenhorst\/picture alliance via Getty Images<\/span><\/figcaption><\/figure>\n

It\u2019s the start of the school year, and thus the start of a fresh round<\/a> of discourse<\/a> on generative AI\u2019s new role in schools. In the space of about three years, essays have gone from a mainstay of classroom education everywhere to a much less useful tool, for one reason: ChatGPT. Estimates of how many students use ChatGPT for essays vary<\/a>, but it\u2019s commonplace enough to force teachers to adapt<\/a>.<\/p>\n

While generative AI has many limitations, student essays fall into the category of services that they\u2019re very good at: There are lots of examples of essays on the assigned topics in their training data, there\u2019s demand for an enormous volume of such essays, and the standards for prose quality and original research in student essays are not all that high.<\/p>\n

\n

This story was first featured in the Future Perfect newsletter<\/a>.<\/h2>\n

Sign up here<\/a> to explore the big, complicated problems the world faces and the most efficient ways to solve them. Sent twice a week.<\/p>\n<\/div>\n

Right now, cheating on essays via the use of AI tools is hard to catch. A number of tools advertise they can verify that text is AI-generated, but they\u2019re not very reliable<\/a>. Since falsely accusing students of plagiarism<\/a> is a big deal, these tools would have to be extremely accurate to work at all \u2014 and they simply aren\u2019t.\u00a0<\/p>\n

AI fingerprinting with technology<\/strong><\/h2>\n

But there is a technical solution here. Back in 2022, a team at OpenAI, led by quantum computing researcher Scott Aaronson<\/a>, developed a \u201cwatermarking\u201d solution that makes AI text virtually unmistakable \u2014 even if the end user changes a few words here and there or rearranges text. The solution is a bit technically complicated, but bear with me, because it\u2019s also very interesting.\u00a0<\/p>\n

At its core, the way that AI text generation works is that the AI \u201cguesses\u201d a bunch of possible next tokens given what appears in a text so far. In order not to be overly predictable and produce the same repetitive output constantly, AI models don\u2019t just guess the most probable token \u2014 instead, they include an element of randomization, favoring \u201cmore likely\u201d completions but sometimes selecting a less likely one.\u00a0<\/p>\n

The watermarking works at this stage. Instead of having the AI generate the next token according to random selection, it has the AI use a nonrandom process: favoring next tokens that get a high score in an internal \u201cscoring\u201d function OpenAI invented. It might, for example, favor words with the letter V just slightly, so that text generated with this scoring rule will have 20 percent more Vs than normal human text (though the actual scoring functions are more complicated than this). Readers wouldn\u2019t normally notice this \u2014 in fact, I edited this newsletter to increase the number of Vs in it, and I doubt this variation in my normal writing stood out.\u00a0<\/p>\n

Similarly, the watermarked text will not, at a glance, be different from normal AI output. But it would be straightforward for OpenAI, which knows the secret scoring rule, to evaluate whether a given body of text gets a much higher score on that hidden scoring rule than human-generated text ever would. If, for example, the scoring rule were my above example about the letter V, you could run this newsletter through a verification program and see that it has about 90 Vs in 1,200 words, more than you\u2019d expect based on how often V is used in English. It\u2019s a clever, technically sophisticated solution to a hard problem, and OpenAI has had a working prototype for two years<\/a>.<\/p>\n

So if we wanted to solve the problem of AI text masquerading as human-written text, it\u2019s very much solvable. But OpenAI hasn\u2019t released their watermarking system, nor has anyone else in the industry. Why not?<\/p>\n

It\u2019s all about competition<\/h2>\n

If OpenAI \u2014 and only OpenAI \u2014 released a watermarking system for ChatGPT, making it easy to tell when generative AI had produced a text, this wouldn\u2019t affect student essay plagiarism in the slightest. Word would get out fast, and everyone would just switch over to one of the many AI options available today: Meta\u2019s Llama, Anthropic\u2019s Claude, Google\u2019s Gemini. Plagiarism would continue unabated, and OpenAI would lose a lot of its user base. So it\u2019s not shocking that they would keep their watermarking system under wraps.\u00a0<\/p>\n

In a situation like this, it might seem appropriate for regulators to step in. If every generative AI system is required to have watermarking, then it\u2019s not a competitive disadvantage. This is the logic behind a bill introduced this year in the California state Assembly, known as the California Digital Content Provenance Standards<\/a>, which would require generative AI providers to make their AI-generated content detectable, along with requiring providers to label generative AI and remove deceptive content. OpenAI is in favor<\/a> of the bill \u2014 not surprisingly, as they\u2019re the only generative AI provider known to have a system that does this. Their rivals are mostly opposed.<\/p>\n

I\u2019m broadly in favor of some kind of watermarking requirements for generative AI content. AI can be incredibly useful<\/a>, but its productive uses don\u2019t require it to pretend to be human-created. And while I don\u2019t think it\u2019s the place of government to ban newspapers from replacing us journalists with AI, I certainly don\u2019t want outlets to misinform readers about whether the content they\u2019re reading was created by real humans<\/a>.\u00a0<\/p>\n

Though I\u2019d like some kind of watermarking obligation, I am not sure it\u2019s possible to implement. The best of the \u201copen\u201d AI models that have been released (like the latest Llama), models that you can run yourself on your own computer, are very high quality \u2014 certainly good enough for student essays. They\u2019re already out there, and there\u2019s no way to go back and add watermarking to them because anyone can run the current versions, whatever updates are applied in future versions. (This is among the many ways I have complicated feelings about open models. They enable an enormous amount of creativity, research, and discovery \u2014 and they also make it impossible to do all kinds of common-sense anti-impersonation or anti-child sexual abuse material<\/a> measures that we otherwise might really like to have.)<\/p>\n

So even though watermarking is possible, I don\u2019t think we can count on it, which means we\u2019ll have to figure out how to address the ubiquity of easy, AI-generated content as a society. Teachers are already switching to in-class essay requirements and other approaches to cut down on student cheating. We\u2019re likely to see a switch away from college admissions essays as well \u2014 and, honestly, it\u2019ll be good riddance, as those were probably never a good way to select students<\/a>.\u00a0<\/p>\n

But while I won\u2019t mourn much over the college admissions essay, and while I think teachers are very much capable of finding better ways to assess students, I do notice some troubling trends in the whole saga. There was a simple way to let us harness the benefits of AI without obvious downsides like impersonation and plagiarism, yet AI development happened so fast that society more or less just let the opportunity pass us by. Individual labs could do it, but they won\u2019t because it\u2019d put them at a competitive disadvantage \u2014 and there isn\u2019t likely to be a good way to make everyone do it.\u00a0<\/p>\n

In the school plagiarism debate, the stakes are low. But the same dynamic reflected in the AI watermarking debate \u2014 where commercial incentives stop companies from self-regulating and the pace of change stops external regulators from stepping in until it\u2019s too late \u2014 seems likely to remain as the stakes get higher.<\/p>\n","protected":false},"excerpt":{"rendered":"

<img alt="A computer screen full of lines of code." data-caption="The language program developed by the US company OpenAI uses artificial intelligence to write a random binary code consisting of zeros and ones. | Frank Rumpenhorst\/picture alliance via Getty Images” data-portal-copyright=”Frank…<\/p>\n","protected":false},"author":1,"featured_media":543,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[9],"tags":[],"_links":{"self":[{"href":"http:\/\/washnow.me\/index.php\/wp-json\/wp\/v2\/posts\/541"}],"collection":[{"href":"http:\/\/washnow.me\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/washnow.me\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/washnow.me\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/washnow.me\/index.php\/wp-json\/wp\/v2\/comments?post=541"}],"version-history":[{"count":1,"href":"http:\/\/washnow.me\/index.php\/wp-json\/wp\/v2\/posts\/541\/revisions"}],"predecessor-version":[{"id":542,"href":"http:\/\/washnow.me\/index.php\/wp-json\/wp\/v2\/posts\/541\/revisions\/542"}],"wp:featuredmedia":[{"embeddable":true,"href":"http:\/\/washnow.me\/index.php\/wp-json\/wp\/v2\/media\/543"}],"wp:attachment":[{"href":"http:\/\/washnow.me\/index.php\/wp-json\/wp\/v2\/media?parent=541"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/washnow.me\/index.php\/wp-json\/wp\/v2\/categories?post=541"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/washnow.me\/index.php\/wp-json\/wp\/v2\/tags?post=541"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}