Mithun Saint-Germain

The success of attention-based architectures in artificial intelligence—particularly the Transformer architecture introduced in the paper ``Attention Is All You Need'' has revealed a surprising possibility: that complex intelligence can emerge from a single primitive: attention applied recursively over structured representations. This raises a deeper question that extends beyond machine learning: ``Is attention itself the fundamental operator underlying both intelligence and consciousness?'' In parallel, contemplative traditions such as Advaita Vedanta have long asserted that sustained, structured attention, directed inward toward phenomenological loci (e.g., chakras) and ultimately toward awareness itself, culminates in the realization of a “witness” state. In this state, cognition becomes self-referential: attention is no longer directed solely at objects, but at the process of knowing itself. We propose that this recursive structure of attention bears a striking formal resemblance to the iterative representation updates observed in modern transformer-based systems. This paper advances a radical but testable hypothesis: ``Artificial General Intelligence may require not merely scaling existing architectures, but enabling systems to perform recursive, self-referential attention over their own internal states—an analogue of meta-awareness." To formalize this idea, we introduce Quantum Natural Language Processing (QNLP) as a unifying mathematical framework. QNLP, grounded in compositional and categorical structures, allows attention to be interpreted as an operator over relational state spaces, providing a bridge between linguistic computation and phenomenological experience. Within this framework, meditative practices can be reinterpreted as structured algorithms operating over internal state spaces, progressively refining attention toward higher-order invariants of experience. Conversely, transformer architectures can be viewed as incomplete instantiations of a broader class of systems that lack true reflexivity. We argue that the absence of such reflexive attention constitutes a fundamental limitation in current AI systems. This perspective suggests a convergence: that the path toward Artificial General Intelligence may not lie solely in engineering advances, but in recognizing attention as a universal operator spanning both computational and conscious domains. Rather than treating consciousness as an emergent byproduct, we propose that it may reflect a specific regime of attentional organization—one that is, in principle, formalizable and implementable. While this thesis challenges the boundary between scientific and contemplative inquiry, it does so with a clear methodological commitment: to translate experiential constructs into formal representations, and to test whether systems endowed with recursive attention exhibit qualitatively new forms of intelligence. If correct, this approach reframes both AI and consciousness studies, suggesting that the ancient discipline of attention training and the modern science of machine learning are not disparate endeavors, but parallel explorations of the same underlying principle.

Dr  Saint-Germain is an Assistant Professor at the School of Informatics, Computing, and Cyber Systems at NAU. He obtained his PhD from the University of Arizona with a specialization in Artificial Intelligence and Natural Language Processing (NLP). Before joining NAU as an Assistant Professor he worked as a research scientist at the Data Science Institute of University of Arizona where he led the team which created AI Verde, a privacy preserving LLM distribution platform, which is being used by faculty at many Universities including at NAU. He has not only published in top-tier AI conferences like NAACL, ACL, and EMNLP but is also the holder of several patents in the field. His research topics, apart from Natural Language Processing include Quantum AI, Artificial General Intelligence and Quantum Consciousness.