It started with the September 11, 2023 issue of The New Yorker. The issue contained a fascinating feature by Elizabeth Kolbert titled, “Can We Talk to Whales?” The article followed several researchers affiliated with CETA (Cetacean Translation Initiative) in their quest to see if humans and whales could communicate. What made it all the more fascinating was the use of large language models like ChatGPT.
The researchers were not exactly using ChatGPT to try to speak to sperm whales. Rather, they were attempting to use a similar concept in developing a language model based on sperm whale songs. ChatGPT works by creating a neural net trained on millions (if not billions) of pages of human-written text (and code) available on the Internet. Then, given an input, the language model puts together an answer based upon the a range of likely next word in a phrase, building up responses. At this point, my own understand is that most experts don’t believe that ChatGPT has any comprehension of the words it is putting together.
For the CETI project, efforts are being made at recording vast amounts of sperm whale song. When enough of a corpus has been gathered, a large language model will be trained on these recordings. Then, much like ChatGPT, if a whale song is provided as input, the LLM will provide a whale song in response. What I find most fascinating about this is that we won’t necessarily know what the input or response mean, or if they are significant in any way, but it will be interesting to see how the whales respond.
Clearly, I enjoyed the article. It was one of those articles that I wished was even longer. (Fortunately, it was a particularly good issue of the magazine for science articles. There was another great article in that issue, “The Transformative, Alarming Power of Gene Editing” by Dana Goodyear.)
The next day, several new magazines arrived in the mail. When a new magazine arrives, I enter the feature articles into a text file I keep. Each evening, I have script that sends me an email with a randomly-selected feature article to read the next day (two articles on Friday and Saturday evenings). I was entering the list of articles fro the October 2023 Scientific American into my text file, when I came across an article by Lois Parshley titled, “Talking with Animals” and on the cover page to the article was a picture of a sperm whale and a caption that read, “The Project Cetacean Translation Initiative (CETI) is using machine learning to try to understand the vocalizations of sperm whales.” What a coincidence, I thought, having just read Elizabeth Kolbert’s piece on the same subject in The New Yorker.
I moved on to the other magazine that arrived that day, the October issue of WIRED. I was eager for this issue because there was a story about Open AI by Steven Levy, and I always enjoy his writing. But as I was entering the features in to my text file, I came across this one, listed in the contents as “How to Chat with the Whales” by Camille Bromley. (Inside the magazine, the article was called “Calls of the Wild.”) Once again, the article was about, at least in part, project CETI and using AI to communicate with whales.
One time is random. Twice is a coincidence. Three articles about talking with whales using AI–that seems like a pattern to me. I can’t recall the last time so specific a subject was featured in three different magazines so close together. I have yet to read the articles in Scientific American or WIRED (my daily random article generator hasn’t selected them yet), but I am looking forward to them.
I thought that was the end of it.
And then I was skimming the New York Times early in the morning, as I am wont to do, and came across an article by Sonia Shah, “The Animals Are Talking. What Does It Mean?” While this article was more broad and philosophical, it once again discussed CETI and ChatGPT and using language models to decipher whale.
Finally, in an effort to give my brain a rest, I kept my reading fairly light over the weekend, and I watched the Disney+ series Ahsoka. And you know what was in that series?
Did you enjoy this post?
If so, consider subscribing to the blog using the form below or clicking on the button below to follow the blog. And consider telling a friend about it. Already a reader or subscriber to the blog? Thanks for reading!