An attempt to realize ``communication with animals using AI'' in cooperation with Google is underway



In recent years, image generation and high-performance chat using AI have attracted attention, but AI has the potential to be useful not only for art but also for various scientific research. Newly, Google's cloud service Google Cloud introduced a blog about the efforts of the non-profit organization

Earth Species Project, which is trying to realize communication with animals using machine learning.

Can generative AI lead people to understand animals? | Google Cloud Blog
https://cloud.google.com/blog/transform/can-generative-ai-help-humans-understand-animals-earth-species-project-conservation



In recent years, research that uses AI to analyze animal sounds has attracted attention. In addition to being advanced by people, there is also an application that records bird calls and identifies the type of bird with AI .

The Earth Species Project is a non-profit organization established in 2018 with the aim of better understanding and communicating with the various animals that inhabit the earth using machine learning. “As humans, our understanding is limited by our ability to perceive things,” said Aza Raskin, co-founder of the Earth Species Project. That's it,' he said.

You may be wondering, 'Do animals other than humans really communicate verbally?' In addition to being known to be transmitted and popular , research results have been reported that tits manipulate languages with words and grammar . A study that analyzed a huge amount of chimpanzee calls also showed that 12 different calls were combined to create 390 different constructs.

Recordings of 5000 barks suggest that chimpanzees are talking using 390 sentences - GIGAZINE



Raskin believes that whales and dolphins, which communicate by barking, may have transmitted culture and songs for 34 million years, and humans may learn from them. In order to realize communication with animals, the Earth Species Project is promoting efforts to use large-scale language models such as LaMDA for research.

Even if the same translation system is built, there is a big difference between human language and animal language. In the case of humans, people who speak different languages can build a common understanding of 'what this word means' and proceed with translation, but in the case of animal languages, humans and animals can understand each other. It is difficult to build a common understanding on both sides. However, by using research results on animal cries that have been accumulated for decades and AI's excellent pattern recognition ability, it will be possible to decipher animal cries.

The Earth Species Project is a natural language model that is also the basis of generative AI. Machine learning can convert the semantic relationships of words into geometric relationships, and can define words as relationships between concepts. I am inspired by the idea that it is possible.

For example, the word 'DOG' is associated with words such as 'Man', 'Cat', 'Wolf', 'Fur', and 'Yelp'. You can show the relationships between them as a visual map.



This relational pattern between words has similarity between different languages, and whether it is spanish, english or japanese, the word 'dog' is almost They appear in the same place. Raskin said, ``If there is only one thing to keep in mind about how AI works, it is that it turns semantic relationships into geometric relationships. This is the core concept. It is.'



At the time of writing, the Earth Species Project is working with scientists and researchers around the world to create benchmarks and basic models that include data for many animal species, including whales, seals, dolphins, and crows. In addition, we are also conducting communication experiments with early animals in cooperation with our partner animal research institute. etc., are being developed.

Katie Zacarian, CEO of the Earth Species Project, said, ``Partnerships are essential to advance this research field.We work with more than 40 biologists and research institutions to collect data. Turning the data into something useful for our partners and us, and adding it to our benchmarks, allows us and others to know if our algorithms are getting better.' 'Decades of research. The contribution of partners built on top of that is crucial.” Google Cloud supports the Earth Species Project by providing computing resources used to enhance machine learning, and Google's speech language modeling team also interacts with the Earth Species Project.

Raskin speculates that within 12 to 36 months it will be possible to communicate with animals using voice generated by AI. Since AI can respond to animal cries by pattern recognition, there is a possibility that communication between animals and AI will be established in a state where humans cannot understand what they are saying.

On the other hand, the Earth Species Project also recognizes that AI-animal communication raises ethical concerns, and is discussing responsible AI use with various researchers. For example, there is a risk that AI will transmit harmful messages to animals, which will adversely affect their culture and lives. As such, much of the Earth Species Project's efforts have been directed toward data collection and building underlying models, Google Cloud said.

Regarding the significance of realizing communication with animals, Mr. Raskin said, ``I think there will be moments when you change your perspective and see yourself in a new way, and everything changes - your relationship with others and the world around you. That moment creates a movement.'



in Software,   Web Service,   Science,   Creature, Posted by log1h_ik