{"id":602215,"date":"2023-01-27T06:16:00","date_gmt":"2023-01-27T12:16:00","guid":{"rendered":"https:\/\/news.sellorbuyhomefast.com\/index.php\/2023\/01\/27\/a-watermark-for-chatbots-can-expose-text-written-by-an-ai\/"},"modified":"2023-01-27T06:16:00","modified_gmt":"2023-01-27T12:16:00","slug":"a-watermark-for-chatbots-can-expose-text-written-by-an-ai","status":"publish","type":"post","link":"https:\/\/newsycanuse.com\/index.php\/2023\/01\/27\/a-watermark-for-chatbots-can-expose-text-written-by-an-ai\/","title":{"rendered":"A watermark for chatbots can expose text written by an AI"},"content":{"rendered":"
\n
\n
\n
\n
\n

The tool could let teachers spot plagiarism or help social media platforms fight disinformation bots.<\/p>\n<\/div>\n

\n
<\/p>\n

\"Close<\/p>\n

<\/span>

Close up of woman’s hand whiting to do list<\/span><\/figcaption><\/figure>\n<\/div>\n<\/div>\n<\/header>\n<\/div>\n
\n
\n

Hidden patterns purposely buried in AI-generated texts could help identify them as such, allowing us to tell whether the words we\u2019re reading are written by a human or not.<\/p>\n

These \u201cwatermarks\u201d are invisible to the human eye but let computers detect that the text probably comes from an AI system. If embedded in large language models, they could help prevent some of the problems that these models have already caused.<\/p>\n<\/p><\/div>\n

\n

For example, since OpenAI\u2019s chatbot ChatGPT was launched in November, students have already started cheating by using it to write essays for them. News website CNET has used ChatGPT to write articles, only to have to issue corrections<\/a> amid accusations<\/a> of plagiarism. Building the watermarking approach into such systems before they\u2019re released could help address such problems.\u00a0<\/p>\n

In studies, these watermarks have already been used to identify AI-generated text with near certainty. Researchers at the University of Maryland, for example, were able to spot text created by Meta\u2019s open-source language model, OPT-6.7B, using a detection algorithm they built. The work is described in a paper <\/a>that\u2019s yet to be peer-reviewed, and the code will be available<\/a> for free around February 15.\u00a0<\/p>\n

AI language models work by predicting and generating one word at a time. After each word, the watermarking algorithm randomly divides the language model\u2019s vocabulary into words on a \u201cgreenlist\u201d and a \u201credlist\u201d and then prompts the model to choose words on the greenlist.\u00a0<\/p>\n

The more greenlisted words in a passage, the more likely it is that the text was generated by a machine. Text written by a person tends to contain a more random mix of words. For example, for the word \u201cbeautiful,\u201d the watermarking algorithm could classify the word \u201cflower\u201d as green and \u201corchid\u201d as red. The AI model with the watermarking algorithm would be more likely to use the word \u201cflower\u201d than \u201corchid,\u201d explains Tom Goldstein, an assistant professor at the University of Maryland, who was involved in the research.\u00a0<\/p>\n<\/div>\n

\n

ChatGPT is one of a new breed of large language models that generate text so fluent it could be mistaken for human writing. These AI models regurgitate facts confidently but are notorious for spewing falsehoods and biases. To the untrained eye, it can be almost impossible to distinguish a passage written by an AI model from one written by a human. The breathtaking speed of AI development means that new, more powerful models quickly make our existing tool kit for detecting synthetic text <\/a> less effective. It\u2019s a constant race between AI developers to build new safety tools that can match the latest generation of AI models.<\/p>\n

\u201cRight now, it\u2019s the Wild West,\u201d says John Kirchenbauer, a researcher at the University of Maryland, who was involved in the watermarking work. He hopes watermarking tools might give AI-detection efforts the edge. The tool his team has developed could be adjusted to work with any AI language model that predicts the next word, he says.<\/p>\n

The findings are both promising and timely, says Irene Solaiman, policy director at AI startup Hugging Face, who worked on studying AI output detection<\/a> in her previous role as an AI researcher at OpenAI, but was not involved in this research.\u00a0<\/p>\n

\u201cAs models are being deployed at scale, more people outside the AI community, likely without computer science training, will need to access detection methods,\u201d says Solaiman.\u00a0<\/p>\n<\/p><\/div>\n

\n

There are limitations to this new method, however. Watermarking only works if it is embedded in the large language model by its creators right from the beginning. Although OpenAI is reputedly working on methods to detect AI-generated text, including watermarks, the research remains highly secretive. The company doesn\u2019t tend to give external parties much information about how ChatGPT works or was trained, much less access to tinker with it. OpenAI didn\u2019t immediately respond to our request for comment.\u00a0<\/p>\n

It\u2019s also unclear how the new work will apply to other models besides Meta\u2019s, such as ChatGPT, Solaiman says. The AI model the watermark was tested on is also smaller than popular models like ChatGPT.\u00a0<\/p>\n

More testing is needed to explore different ways someone might try to fight back against watermarking methods, but the researchers say that attackers\u2019 options are limited. \u201cYou\u2019d have to change about half the words in a passage of text before the watermark could be removed,\u201d says Goldstein. \u00a0<\/p>\n

\u201cIt\u2019s dangerous to underestimate high schoolers, so I won\u2019t do that,\u201d Solaiman says. \u201cBut generally the average person will likely be unable to tamper with this kind of watermark.\u201d \u00a0<\/svg> <\/p>\n<\/div>\n<\/div>\n<\/div>\n

Read More<\/a>
\n Melissa Heikkil\u00e4<\/p>\n","protected":false},"excerpt":{"rendered":"

The tool could let teachers spot plagiarism or help social media platforms fight disinformation bots. Close up of woman’s hand whiting to do list Hidden patterns purposely buried in AI-generated texts could help identify them as such, allowing us to tell whether the words we\u2019re reading are written by a human or not. These \u201cwatermarks\u201d […]<\/p>\n","protected":false},"author":1,"featured_media":602216,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[42004,534,97776],"tags":[],"class_list":{"0":"post-602215","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-chatbots","8":"category-financial","9":"category-watermark"},"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/posts\/602215","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/comments?post=602215"}],"version-history":[{"count":0,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/posts\/602215\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/media\/602216"}],"wp:attachment":[{"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/media?parent=602215"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/categories?post=602215"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/tags?post=602215"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}