Watermarks for Chatbots

ACM NEWS

A Watermark for Chatbots can Expose Text Written by an AI, By MIT Technology Review

February 6, 2023

“Right now, it’s the Wild West,” says John Kirchenbauer, a researcher at the University of Maryland, who was involved in the watermarking work.

Hidden patterns purposely buried in AI-generated texts could help identify them as such, allowing us to tell whether the words we're reading are written by a human or not.

These "watermarks" 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.

For example, since OpenAI's 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 amid accusations of plagiarism. Building the watermarking approach into such systems before they're released could help address such problems. 

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's open-source language model, OPT-6.7B, using a detection algorithm they built. The work is described in a paper that's yet to be peer-reviewed, and the code will be available for free around February 15.

From MIT Technology Review  

Enhanced Packaging via Google Lens

Enhancing packaging with scans?  I thought this had been done already with QR codes, I still see it on packages.   Google lens can be found on IOS and Android.   You point it at a package, I assume anywhere on the package,  and you will be linked to their information.    We tested related watermarking ideas.  This also assumes you know about the linkage, probably from an ad.  But you could have done the same with a QR code. Difference appears to be that you don't have to find a QR to scan.

Uncle Ben's Adds Google Lens to Packaging      By Jacqueline Barba - 09/18/2019

Mars Inc.’s Uncle Ben’s has teamed with connected food platform Innit to launch an artificial intelligence-driven initiative that leverages the Google Lens visual search technology to connect digital information to physical products and unlock meal solutions.

The partnership makes Uncle Ben’s the first food brand to adopt Google Lens, an image recognition technology designed to let users access relevant information related to objects the device identifies using visual analysis and "search what they see," according to a media release from Innit.

In this case, Uncle Ben’s is using the technology to help consumers “cut through the clutter” of meal planning by instantly delivering recommendations, information and inspiration about physical products in stores and homes.

To activate, users open either the Google mobile app on iOS devices or the Google Lens app on Android devices, and point their device at Uncle Ben’s Ready Rice packages and retail displays. Innit will then instantly suggest meals that can be built around the product, along with ingredient lists, nutrition advice and step-by-step cooking videos.  .... "

Accoustical Watermarking and the Second Screen

And yet more on context switching for the voice assistant.    Here work by Amazon, to be presented at an upcoming conference.   Originating from work to ignore 'wake words' in 'second screens' from other media.  Which seems to work quite well now.   But immediately made me think of:  Why not include more data in the watermark to identify it further, transmit information it learns about a context.  Nice direction.

Audio Watermarking Algorithm Is First to Solve "Second-Screen Problem" in Real Time   By Yuan-yen Tai

Audio watermarking is the process of adding a distinctive sound pattern — undetectable to the human ear — to an audio signal to make it identifiable to a computer. It’s one of the ways that video sites recognize copyrighted recordings that have been posted illegally.

To identify a watermark, a computer usually converts a digital file into an audio signal, which it processes internally. If the watermark were embedded in the digital file, rather than in the signal itself, then re-encoding the audio in a different file format would eliminate the watermark.

Watermarking schemes designed for on-device processing tend to break down, however, when a signal is broadcast over a loudspeaker, captured by a microphone, and only then inspected for watermarks. In what is referred to as the second-screen problem, noise and interference distort the watermark, and delays from acoustic transmission make it difficult to synchronize the detector with the signal. 

At this year’s International Conference on Acoustics, Speech, and Signal Processing, in May, Amazon senior research scientist Mohamed Mansour and I will present a new audio-watermarking algorithm that effectively solves the second-screen problem in real time for the first time in the watermarking literature.   .... "

Digimarc at the NRF

I had mentioned Digimarc and their advanced product marking technologies, recently in a post.   They will be present at the upcoming 2017 NRF show on January 15-17 in NYC.   Here is an invite and more information about their company.  See the tag below for more up to date information about them in this blog.

Digital Watermarks

A method we experimented with:

Tech Partnership Looks Beyond the Bar Code With Digital Watermarks
Backers say watermark technology on product packaging can speed up supply chains and retail transactions   ... " 

Stolen Data Phones Home

In CSOonline:  " ... Digital ‘watermarking’ won’t prevent your data from being stolen, but it can let you find out about it much more quickly than the average of seven months it takes organizations to discover a breach  ... " 

QR Codes and Counterfeiting

In Retailwire: Using QR codes to eliminate counterfeit goods.  We examined a number of watermarking methods.

Shopping by Scanning Images

In MarketWatch:   We experimented with approaches that used watermarks embedded in packaging.   Includes a short video demo to show it in operation.

" .... You can now shop on Amazon by taking photos
App’s new “Flow” feature allows for point-and-shoot shopping

 " ...  Amazon app's image recognition
Amazon is integrating product recognition software, called “Flow,” into its main shopping app. Point your phone’s camera at a labeled product, and if Amazon sells it, you’ll see it pop up.

Flow—as its name suggests—aims to make it as seamless as possible to shop. MarketWatch carried out its own “showrooming” with the app. Scanning a three-bottle package of the hair growth serum Rogaine, Flow immediately found the item on Amazon for $43.85, 30% cheaper than the $62.99 price in a Duane Reade store. .... "  

Digimarc's Watermarking

We examined Digimarc's watermarking technology in the innovation center as an alternative to RFID.   " ... Digimarc has created an invisible watermark that allows bar-code information to be placed all over a product, making scanning easier and checkout faster. A company team used the technology to set a Guinness World Record for bagging groceries. Digimarc says the technology increases scanning capability by 50%  ... "   Not sure how widely this has been implemented.