Could AI help us see beyond our anthropocentrism?
Artificial intelligence is increasingly being used to interpret the communication of the other species we share our planet with. Project CETI, for example, is developing AI to decode sperm whale communication. According to the researchers of the project, sperm whales are the ideal starting point for decoding non-human animal communication because of their sophisticated cognitive abilities, complex social structures, and click-based communication system. Artificial intelligence enables the analysis of large volumes of data to detect meaningful patterns, which is the first step towards translating the communication system of whales [1]. Convolutional neural networks have already been able to identify the vocal clan to which certain whale codas belong [2], potentially suggesting dialectal differences between whale groups.
While there is a growing body of research on the ethical implications of this technology, that is not what I want to focus on in this blogpost. Instead, I want to reflect on the anthropocentric biases involved in the development of artificial intelligence, a topic richer than I initially expected.
Anthropocentric algorithmic bias
Even with deliberate efforts to prevent it, anthropocentric algorithmic biases are nearly impossible to avoid. An algorithmic bias is caused by the structure and decisions made within an algorithm, which may unintentionally prioritize particular variables, producing unequal results [3]. As a human-made system, AI for non-human animal communication inherently carries an anthropocentric bias by reflecting assumptions about what language should be. It may misinterpret whale communication by attempting to understand it through a human linguistic framework. There is also an underlying assumption that communication is only meaningful if it can be understood by humans. The deeper question is whether AI would even be able to convey the message to us, and whether we would be open to the idea that meaning may exist independently of our ability to understand it.
Human exceptionalist confirmation bias
We might not understand which parts of whale language are essential and which are not. This could to (unintended) human exceptionalist confirmation bias. Confirmation bias occurs when AI systems are used to justify the beliefs and biases of their operators [3]. Project CETI demonstrates a sincere commitment to understanding whale communication and discovering human biases within the development of the technology. Still, there remains a risk that others may interpret their findings in ways that reinforce human exceptionalism. Even though such interpretations might just be a sign that we are not (yet) equipped to recognize what counts as meaningful communication in other species.
The limits of human understanding
Our understanding of the world is confined to the limits of human perception and cognition, which means it excludes forms of knowledge and intelligence that exist beyond the reach of human concepts. Is it even possible to take our own limitations of perception and cognition into account when designing artificial intelligence intended to interpret non-human animal communication? Or could AI become a tool to uncover our human biases in animal communication and intelligence research?
AI as a tool for understanding human biases?
A study published in Nature Communications by Project CETI reported that whale codas exhibit both contextual and combinatorial structures [4]. Initially, researchers approached whale clicks as a kind of Morse code. That hypothesis would make the most sense when you compare it to how humans use click-based communication. However, unsupervised learning algorithms identified additional features as important, such as the spectral properties of codas [5]. According to Beguš, a researcher at Project CETI, the research conducted using AI revealed their human biases and enabled them to think beyond their assumptions [5]. Project CETI now proposes that whale codas may function in similar ways as human vowels and diphthongs [6]. As tempting as it is to explore this discovery further, for now, I want to focus on the fact that these insights were discovered through AI. Through its ability to detect patterns in large amounts of data, it recognized potentially meaningful properties that humans did not (and could not) notice. The way these findings are discovered raises interesting questions. Could artificial intelligence help reveal human biases embedded in the traditional approaches to decoding non-human animal communication? Or is that impossible, simply because artificial intelligence is created by humans?
Conceptually engineering the anthropocentrism out of the notions of language and intelligence
The quest to understand anthropocentric biases in AI should go hand in hand with critically reflecting on the anthropocentrism within our definitions of language and intelligence. I believe that in our preoccupation with proving human uniqueness, we have perhaps blinded ourselves to other forms of intelligence and communication (or language?). Conceptual engineering refers to the process of proposing additions, revisions, or eliminations to concepts, meanings, or norms [7]. As we work to understand whale communication, I suggest we simultaneously revisit our definitions of language and intelligence to ensure these concepts are not just used to justify human exceptionalism and, by extension, to rationalize the way we currently treat non-human animals. Rather than insisting on language as an exclusively human trait, we might open ourselves to the possibility that the complex communication systems of whales are a form of language in their own right, even if they follow different rules or do not meet every human-defined criterion. I believe these discussions should evolve alongside the development of the technology and the future discoveries of projects like Project CETI.
So what now?
The questions I raised throughout this text are not easy to answer, because the truth is that we are only at the very beginning of understanding other species. What I am ultimately calling for is humility, especially regarding the qualities we tend to celebrate as uniquely human. We are not the owners of all the knowledge or intelligence that exists in this world. I believe we should remain humble and open-minded to the possibility that there are layers of intelligence, communication, and understanding that lie beyond the human scope. Realizing this gives us yet another reason to prioritize the conservation of animals. Conservation is, in a sense, an act of knowledge preservation. When species go extinct, entire forms of intelligence and worlds of knowledge will vanish with them.
References
[1] Andreas, Jacob, Gašper Beguš, Michael M. Bronstein, et al. 2021. "Cetacean Translation Initiative: A Roadmap to Deciphering the Communication of Sperm Whales." ArXiv. https://doi.org/10.48550/arXiv.2104.08614.
[2] Bermant, Peter C., Michael M. Bronstein, Robert J. Wood. 2019. "Deep Machine Learning Techniques for the Detection and Classification of Sperm Whale Bioacoustics." Scientific Reports 9: 12588. https://doi.org/10.1038/s41598-019-48909-4.
[3] Ferrara, Emilio. 2023. "Fairness and Bias in Artificial Intelligence: A Brief Survey of Sources, Impacts, and Mitigation Strategies." arXiv. http://arxiv.org/abs/2304.07683.
[4] Sharma, Pratyusha, Shane Gero, Roger Payne, et al. "Contextual and combinatorial structure in sperm whale vocalisations." Nature Communications 15: 3617. https://doi.org/10.1038/s41467-024-47221-8.
[5] Beguš, Gásper. 2025. "New Patterns in Animal Communication with AI: Ethical and Legal Implications." Interspecies Internet, July 21. Youtube video, 1:49:06. https://www.youtube.com/watch?v=mzdeoQ6z2dU.
[6] Beguš, Gašper, Ronald L. Sprouse, Andrej Leban, Miles Silva, and Shane Gero. 2025. "Vowel- and Diphthong-Like Spectral Patterns in Sperm Whale Codas." Open Mind: Discoveries in Cognitive Science 9: 1849-1874. https://doi.org/10.1162/OPMI.a.252.
[7] Löhr, Guido. 2023. "If Conceptual Engineering Is a New Method in the Ethics of AI, What Method Is It Exactly?" AI and Ethics 4(2): 575-585. https://doi.org/10.1007/s43681-023-00295-4.