Hi, we’re Advanced Cognitive Systems Lab, a privately owned, not-for-profit AI research company based in Karlsruhe, Germany.
As the world turns into turmoil, the race for some vague ’superintelligences‘ becomes fueled by top-level geopolitical interest. At the same time, it becomes obvious that token-centric system architectures are hitting a plateau.
In that particular constellation of multidimensional interest and multinational competition, AI security and alignment tends to be seen as a burden which, as some people argue, needs to be overcome. We couldn’t disagree more!
In our view, the cognitive performance of an autonomous system alone is never the single success criteria. Without structure, strength becomes brittle. Without reliability, it becomes a risk. Without control, it ceases to serve.
For us, building advanced intelligence (we call it ‚Higher-Order Machine Cognition‘) means building something that stays accountable to us as humans — in every moment, under every condition. Structure and control are essential if AI technology is to serve as a foundation for meaningful and sustainable economic value.
We strive to combine cognitive performance with structure and control. This is our foundational motivation and our most noble guiding principle.
At ACSL, we run foundational research to advance today’s model-centric AI architectures into architecturally grounded Autonomous Systems based on our innovative Leibniz-von-Neumann-Architecture — a specific blueprint for compound system structures that embrace the strengths of language model technology, while overcoming their structural limits around safety, reliability, and human control. Read our initial positioning paper, dated Aug 15th, 2025.
Update October 2025: Toward Cognitive State Machines — Technical Research Reflections #002 — read here.
Update June 2026: A New Scaling Paradigm to Higher-Order Machine Cognition — Technical Research Reflections #003 — read here.
We work closely with our shareholder, embraceable Technology, whose Enterprise AI Tech Stack, including the first real-life implementations of the Leibniz-von-Neumann-Architecture allow us to test and evaluate what safe, controlled, and sovereign AI can become.
We have founded ACSL because we believe humans must remain in control at all times — and because we refuse to settle for less than machine intelligence that likewise earns our admiration and our trust.
Because we are still in control.
Now is the time.
Karlsruhe, August 2025.
The ACSL Team.
Advanced Cognitive Systems Lab (ACSL) is an independent research organization based in Karlsruhe, Germany.
Over the past years, the evolution of GenAI has been shaped by a core assumption: that increasingly capable machine cognition can emerge from increasingly better and bigger models. The resulting progress has been extraordinary. Yet as Autonomous Systems are entrusted with more complex cognitive workloads, a growing number of limitations appear: computational effort explodes, yet reliability remains brittle.
Many of the capabilities required for reliable machine autonomy circle around higher-order cognition – handling conflicting priorities, complex constraints, policy alignment, traceability, reproducibility, and cognitive control. Computing such complex cognitive workloads within their linguistic and semantic representation space can become inefficient and — due to the stochastic nature of the dominating autoregressive prediction mechanisms — not sufficiently stable and reliable (at least not in mission-critical and high-stake environments).
For us, this is not a model crisis per se, but a crisis of an intelligent system‘s representation architectures. Language is a perfect linguistic surface, but not necessarily the best substrate for encoding higher-order structures (such as abstract concepts and their relations).
In other words: Semantic intelligence and higher-order cognition are different classes of cognition – and we do believe that different classes of cognition deserve multiple representation spaces, each of which having its own computing substrate wo that cognitive workloads can be shifted across computational substrates. In multi-space architectures, the actual cognition arises from the directed, stateful interaction of different representation spaces.
We see this as a fundamental transition similar to those that have repeatedly shaped the history of computing: from monolithic structures toward specialized architectures. In that perspective, today’s language models remain an integral part of intelligent systems, but will no longer be the sole computational substrate for higher-order machine cognition.
Our research explores a fundamental question: whether higher-order machine cognition can be scaled by advancing representation architectures – multiple representation spaces with different information density, structure and computational primitives. As intelligent systems move from assisting humans toward autonomous execution of complex real-world processes, questions of representation architecture become increasingly impactful and urgent.
If successful, the future of reliable machine autonomy will no longer be shaped by scaling parameters and compute, but even more so by advancing representation architectures and innovative computing paradigms.
We exist to pave that way.
The ACSL Team.
Imprint
Advanced Cognitive Systems Lab gGmbH i.G.
Gablonzer Str. 21
76185 Karlsruhe
GERMANY
Managing Director: Dr.-Ing. Christian Gilcher