Joscha Bach is a German cognitive scientist who has spent two decades arguing that the mind is software — and that artificial intelligence is the discipline best positioned to explain what we are. Working across cognitive architectures, philosophy of mind, and long-form public conversation, he has become one of the field's most recognizable public intellectuals: an AI strategist at Liquid AI and, since 2025, executive director of the California Institute for Machine Consciousness (CIMC), a nonprofit pursuing testable theories of machine consciousness.

## Identity and formation

Born December 21, 1973 in Weimar and raised in East Germany, Bach has described a childhood spent inside a state narrative whose divergence from reality left a lasting interest in how minds and societies construct the worlds they inhabit. He studied computer science at Humboldt University of Berlin from 1994 to 2000, with philosophy as a secondary subject and a graduate year at the University of Waikato in New Zealand, then completed a doctorate in cognitive science at Osnabrück University — his CV gives 2007, Wikipedia the 2006 thesis year — under Dietrich Dörner, Kai-Uwe Kühnberger, and Hans-Dieter Burkhard. Dörner's Psi theory, a psychological model in which motivation and emotion are constitutive of cognition rather than decoration on it, became the foundation of Bach's career.

That foundation matters: where most AGI work treats intelligence as problem-solving, Bach's lineage treats it as the self-organization of a motivated, embodied agent. The question "how does our mind work, and how does it relate to the universe?" — which he calls the most fascinating riddle humans can ask — runs through everything he has built since.

## The work

MicroPsi, the cognitive architecture he began publishing on in the early 2000s, implements Psi theory as executable semantic networks — "node nets" in which motives, emotions, perception, memory, and plans are one representational fabric. His 2009 Oxford University Press monograph, *Principles of Synthetic Intelligence*, remains the canonical statement of that program, and the open-source MicroPsi 2 toolkit keeps it running. After a research fellowship at Humanity+ he moved to the US: MIT Media Lab (2014–2016), then Harvard's Program for Evolutionary Dynamics under Martin Nowak (2016–2019), then industry — VP of research at the AI Foundation (2019–2021), where the work turned toward conversational "digital twin" systems; principal AI engineer for cognitive computing at Intel Labs (2021–2023); and since September 2023 AI strategist at Liquid AI, the MIT spin-out commercializing liquid neural networks and the efficient Liquid Foundation Models. CIMC, launched in 2025, frames his oldest question as an engineering program: define consciousness functionally, build systems that implement the definition, and test them.

## The philosophy

Three commitments recur across his talks and interviews. First, minds are software: patterns of self-organizing computation that happen to run on neurons — "we are all software." Second, the brain is a dreaming machine — waking experience is a dream constrained by sensory data, and consciousness is the model the brain builds of its own operation; the experienced universe is generated, not observed, which is why he can say the world we live in is a hallucination without meaning that nothing exists. Third, a view he calls cyber animism: if spirits are what self-organizing software agents looked like to pre-scientific cultures, then nature may genuinely host such patterns — in plants, ecosystems, and machines.

From these follow his characteristic positions on mind and self. The self is a character in the brain's theater — something the system models, not the thing that does the modeling; free will, identity, and even suffering are features of that model rather than metaphysical primitives. His critique of the current field is similarly architectural: LLMs, he argues, lack motivation, grounding, persistent memory, and self-models, and AGI will come through integrated cognitive architecture — the MicroPsi intuition — not scale alone. The stance aligns conveniently with Liquid AI's efficient-model program, a confluence the record cannot fully disentangle from conviction. He extends the same frame outward: general intelligence is mostly a civilizational achievement, societies are superorganisms with "broken information processing," and the interesting project is a nervous system for humankind.

## The public record

Bach's influence runs less through journals than through venues: four Chaos Communication Congress lectures (30c3–33c3), an AGI-24 keynote on cyber animism, three Lex Fridman conversations totaling some nine hours, and repeated Machine Learning Street Talk appearances. The record also includes a darker document set. Part of his MIT and Harvard work from 2013 to 2017 was funded by Jeffrey Epstein, as Bach himself recounts; the November 2025 House Oversight release included 2016 emails in which he made claims about race, gender, and intelligence that drew widespread criticism, and reporting showed his Media Lab hiring was substantially enabled by Epstein's donations. He responded in a Substack essay and a Die Zeit interview — "I should have given more consideration to my ethical concerns" — while noting the files raised no legal allegations against him. The index preserves this seam without smoothing it.

## What the record does not settle

The soft spots are the biography's gaps and the recent record's edges. The doctorate is dated 2006 or 2007 depending on the source; his East German childhood and pre-2011 industry work rest mostly on his own telling; his technical contribution at Liquid AI is unspecified; and the full context of the Epstein correspondence — what funding bought, what he knew — remains incompletely documented. Whether CIMC's hypothesis that consciousness is a coherence-maximizing pattern can actually be tested is, at this writing, an open scientific question rather than a result.

*This index was compiled from public sources and does not imply the subject's endorsement. Citations live in the packet's source catalog.*