Field Notes: Defining Decode

By Sara Keen, Senior Research Scientist, Behavioral Ecology and AI

Last month, in the foothills of New Mexico’s Sangre de Cristo Mountains, I joined a group of 50 researchers, writers, and artists who had gathered together to chart the future course for decoding animal communication. The event, "Animals in Translation: Creating Criteria and Frameworks for Decoding Communication in Other Species," took place at the Santa Fe Institute, a celebrated hub for complex science and interdisciplinary collaboration. 

The event was organized by Interspecies Internet and Santa Fe Institute in partnership with XPRIZE Conservation and Biodiversity domain. It was inspired by the recent surge of research at the intersection of machine learning and non-human communication. The goal of the meeting was to develop a comprehensive set of criteria to define evidence for the successful decoding of another species' communication systems. In other words, how will we know when we’ve successfully “translated” non-human communication? 

The workshop was not just an opportunity to connect with animal communication experts, it also dovetailed perfectly with Earth Species Project’s discussions about the outstanding challenges in our field – from defining “decode” to outlining what success will look like to building ethical guardrails. 

As ESP continues to refine our own theory of change and roadmap, engaging in these broader conversations is crucial for ensuring that our work addresses the most pressing questions that researchers face today.

Why now?

Researchers have explored the intricacies of animal signals for decades, uncovering phenomena such as seismic signaling in elephants, dolphins' use of individualized whistles, and deceptive signaling in birds (O’Connell-Rodwell 2007, Janik et al. 2016, Igic et al 2015). 

Now, with the availability of powerful AI models, increased compute, and enormous datasets, the research community is on the brink of unprecedented breakthroughs in understanding animal communication.  

This surge of interest in the research community and in the media has led to new research incentives. For example, XPRIZE is considering a large global prize for decoding animal communication. The Dr. Dolittle Challenge recently launched by the Coller Foundation is another example. 

While this is all incredibly exciting and energizing, it's also important to consider how research in this space will impact the natural world and how we can avoid any unintended consequences. 

Together, we took the first steps in defining research goals, identifying specific areas of interest, and proposing ethical guidelines that will shape practices, norms, and research outcomes as the field expands. 

Thinking across disciplines

To create the conditions necessary for paradigm-shifting discoveries, we need diverse voices and perspectives. Alongside academics studying animal communication, cognition, AI, philosophy, and linguistics, we heard from documentary makers, alien intelligence researchers from  SETI Institute, internet pioneer Vint Cent, musician Peter Gabriel, and sci-fi writer Ted Chiang to name a few. 

Thoughtful, interdisciplinary meetings like this one will be key to the responsible growth of AI-driven animal communication research. 

AI, ethics, and decoding

Several research groups, including ESP, have recently released ML models for analyzing non-human signals. Many researchers at “Animals in Translation” are hopeful about the advanced analyses these new models could enable. They also cautioned against the unintentional use of biased models, such as those that inadvertently favor particular species. 

At ESP, we have developed an ethical framework for avoiding bias in model development and application, as well as recommendations for best practices for researchers. We plan to make these materials publicly available to set an example of the appropriate use of AI in studying non-human signals. 

We are also considering how to catalyze a broader conversation around the need for ethical guardrails that will ensure the models we develop won’t be used to harm animals or ecosystems, or accidentally disrupt animals’ natural behaviors.

The workshop also included discussions about training animals to communicate with humans.  While human-taught systems, such as keyboard interfaces for animals, have given us great insight, there was consensus that decoding efforts should focus on the ways animals communicate with each other in nature, encouraging the use of minimally invasive observational methods. This aligns well with ESP’s approach. Ultimately, we want to help humans listen more effectively to non-human species, and  we are developing technologies to make this possible.

Looking forward

Several key questions also emerged around how far decoding is even possible, and what frameworks should be used:

• How much emphasis should be placed on knowing what animals think versus knowing what they “talk” about?
• Should frameworks from human speech, such as Hockett’s design features, be applied to animal communication?
• Can we decode animal communication without comprehending their social landscape and societal structures? 

The discussion was provocative and productive. During the workshop, the group compiled a draft list of criteria. Over the coming months, Interspecies Internet will lead a process to collaboratively iterate and revise the criteria, towards a publication to be released later this year.

More than anything, the workshop underscored that we are on the precipice of something big: more research focus, data, and technology are being directed towards animal communication than ever before. As we start to uncover more about our planet and its inhabitants, our understanding of other species will fundamentally change, enabling humans to reimagine our relationships with the natural world.