Tweet, tremolo, growl, how, hoarse. Animals intersect in various ways, but humans only catch the surface of how they communicate with each other and other living worlds. Our species have trained some animals, and if you ask cats, the animals trained us, but we haven’t really really cracked the code in interspecies communication.
Animal researchers are increasingly deploying artificial intelligence to accelerate our investigation of animal communication, including within species and between branches of the tree of life. As scientists challenge animals’ complex communication systems, they get closer to understanding what organisms are saying and maybe even how they can reply. But as we try to bridge the linguistic gap between humans and animals, some experts are raising effective concerns about whether this ability is appropriate or whether we should try to communicate with animals.
Use AI to unlock animal language
At the front of the package (or should I say Pod?) is CETI, which uses machine learning to analyze structured click patterns for more than 8,000 whale-verbal “Codas”. Researchers found context and combinatorial structures on the whale’s click point, naming functions such as “rubato” and “decoration” to describe how whales skillfully adjust their voice in conversation. These patterns helped the team create a phonetic letter for the animals, a structured system of expression that may not be the language we know, but reveals a level of complexity that researchers didn’t know before. The CETI project is also developing ethical codes for the technology, a key goal given the risk of using AI to “talk” to animals.
Meanwhile, Google and Wild Dolphin Project recently launched Dolphingemma, a large language model (LLM), which has received 40 years of training in dolphin vocalization. Just as Chatgpt is a human-invested LLM, it involves visual information such as research papers and images, and responds to related queries – polyadrenaline to dolphin sound data and predicts what will be vocal next. Dolphingemma can even produce dolphin-like audio, the researchers’ prototype bidirectional system, Cetacean hearing-enhanced telemetry (properly CHAT), using a smartphone-based interface, Dolphins uses smartphones to demand angles between scarves or hagrids such as scarves or hagrids such as future events.
Denise Herzing, founder and director of the Wild Dolphin Project, said: “Dolphingemma was used in the fields this season to improve real-time sound recognition in our chat system. “This fall, we will spend time ingesting known dolphin vocalizations and let Gemma show us any repetitive patterns they find, such as those used in courtship and mother discipline.
In this way, AI applications are two aspects: Researchers can use it to explore the natural sounds of dolphins and better understand the animal’s response to human mimicking dolphin sounds, which are synthesized by AI chat systems.
Extended Animal AI Toolkit
Outside the ocean, researchers found that human speech models could also be reused to decode terrestrial animal signals. The University of Michigan-led team used WAV2VEC2, a speech recognition model trained in human voices, to identify dogs’ emotions, gender, breeds, and even bark-based personal identities. Pre-trained human models are better than versions trained based on dog data only, suggesting that human language model architectures may be very effective in decoding animal communication.
Of course, we need to consider the different levels of goals of these AI models. Determine whether the dog’s bark is aggressive or playful, or whether it is male or female – it is understandable that the model is easier to determine the model than the subtle meaning encoded in the speech whale pronunciation. However, each study brings scientists closer to understanding how AI tools are best applied to such a wide range of fields and gives AI the opportunity to train itself as a more useful part of the researchers’ toolkit.
Even cats (often considered detached) show more communicative than they allow. In a 2022 study at the Paris Nanterre University in Paris, cats showed obvious signs that their voices recognize their owner’s voice, but beyond that, the cats responded more intensely when talking directly to them in “cat talk.” This shows that cats should not only pay attention to what we say, but also what we say, especially when it comes from someone they know.
Earlier this month, a pair of Murphyfish researchers discovered that the animals had four “waves” or physical gestures, which created four “waves” or physical gestures between each other, and human playback of the ink bubble waves. The team plans to use algorithms to classify wave types, automatically track organisms’ movements, and understand animals express their environment faster.
Private companies such as Google are also joining the bill. Last week, Baidu, China’s largest search engine, proposed a patent to the country’s IP administration, suggesting that animal (especially cats) voices be transformed into human language. The quick and dirty thing about this technology is that it will take a lot of data from your kitten and then use AI models to analyze the data, determine the emotional state of the animals, and output obvious human language information that the pet is trying to convey.
Common translation of animals?
Together, these studies represent a major shift in how scientists approach an animal communication. Instead of starting from scratch, the research team builds tools and models designed for humans and makes progress, which will take longer. The ultimate goal can be (read: Can) Rosetta Stone, the animal kingdom powered by artificial intelligence.
“Over the past five years, we have been very good at analyzing human languages, and we have begun to refine this model trained in a dataset and apply it to new data,” Sara Keen, a behavioral ecologist and electrical engineer at the GeoSpecies Project, conducted in a video with Gizmodo.
Earth Species Project plans to launch its flagship audio language model for animal sounds, Naturelm, this year, and Naturelm-Audio’s demonstration is already there. With input data from the Tree of Life, as well as human speech, ambient sounds and even music detection, the model is designed to be a converter for human speech to animal analogues. The project “shows a promising area shift from human speech to animal communication”, “supporting our hypothesis that shared representations in AI can help decode animal language.”
“A big part of our work is actually trying to change people’s perception of our place in the world,” Keen added. “Our discovery of animal communication is cool, but ultimately we find other species as complex and subtle as we do. This revelation is very exciting.”
Moral dilemma
In fact, researchers often agree to AI-based tools to improve the collection and interpretation of animal communication data. However, some people believe there is a classification between academic familiarity and public perceptions of how these tools are applied.
“I think there are a lot of misunderstandings in the coverage of this topic at the moment – in some way, machine learning can make knowledge of this context ubiquitous. As long as you have thousands of hours of recording, there are some magical machine learning black boxes that can squeeze meaning out of it.” “This won’t happen.”
“The meaning comes through contextual annotations, which I think is a period of excitement and enthusiasm for the entire field, which is really important for the entire field, and don’t forget that this annotation comes from basic behavioral ecology and natural history expertise.” In other words, let’s not put horses on the cart, especially because in this case (in this case) what powers the horse.
But there is a powerful force…you know the cliché. In essence, how can humans develop and apply these technologies in a way that scientifically illuminates and minimizes harm or damage to animal subjects? Experts have proposed ethical standards and guardrails to use in the technology we are getting closer and closer, regardless of the technology’s development,
As AI develops, dialogue about animal rights will have to develop. In the future, animals may become more active in these conversations, an idea that legal experts are exploring as a practice of thought, but may one day become a reality.
“What we urgently need in advancing machine learning is to build these meaningful collaborations between machine learning experts and animal behavior researchers,” Rutz said.
There is no shortage of communication data, unable to enter the AI model that longs for data, from the squeaking grassland dogs to the slimy trails of snails (yes, really). But, to be precise, how we use the information we gather from these new methods requires thorough consideration of the ethics involved in “talking” to animals.
A recent paper on the ethical issues of using AI to communicate with whales outlines six major problem areas. These include privacy rights, cultural and emotional harm to whales, anthropomorphism, technological solutionism (over-reliance on problem-solving), gender bias, and limited effectiveness of actual whale protection. The last question is especially urgent given how many whale populations are already under serious threats.
It seems that we are increasingly on the verge of learning more about how animals interact with each other – in fact, withdrawing their communication curtains backwards can also give them insight into the ways they learn, socialize, and act in the environment. But there are still major challenges, such as asking yourself how to use the powerful technology currently being developed.
