Skip to Content

Principles & Builders — tokum.ai

About Us


The current generation of AI systems is built on a foundational illusion: that intelligence is an asset to be accumulated — in parameters, in data centers, in compute — and that scaling these resources indefinitely will eventually produce genuine understanding. It will not. Today's large language models are thermodynamically and mathematically constrained systems that compress language into opaque statistical manifolds, entirely disconnected from verifiable reality. Hallucinations are not an engineering bug to be patched with more data or reinforcement learning. They are a structural inevitability — a direct consequence of the underlying geometry. There is no compute fix. The topology is fixed.​

The deeper problem is the question no one is asking: what is intelligence, actually? It is not a property that a sufficiently large model eventually possesses. Intelligence is a verified flow of gap-closure emerging from a network — biological, social, or computational. A peregrine falcon doesn't navigate at 390 km/h because it runs the largest internal model. It does so because verified sensory information flows through a tight, grounded feedback loop at one watt of metabolic power. The race to build bigger black boxes is not a race toward intelligence. It is a race in the wrong direction.​

The Tokum Initiative was founded on a different equation. The Semiotic Web replaces statistical text prediction with Tokums — cryptographically verified, atomic units of meaning, each independently discoverable by any agent on any device without a central authority. Because meaning is structurally exact and topologically grounded, hallucination becomes architecturally impossible — not suppressed, but eliminated by construction. Every reasoning step is fully traceable and auditable. The entire system operates at the edge, at approximately one watt of power, without a data center.

The industry is locked in an arms race, betting that sheer hardware scale is the ultimate competitive moat. We believe the future of intelligence will not be won by brute force. It will be won by the right architecture.



Leadership

Guided by pioneers who believe AI should comprehend, not approximate.


Eric Blaettler

Eric Blaettler

Founder, Chairman and Interim CEO

is a knowledge architect and systems thinker who has spent decades exploring how digital systems capture, organise, and verify meaning. As the creator of the Semiotic Web, he approaches artificial intelligence as a structural problem rather than a scaling one — arguing that hallucinations, opacity, and the absence of genuine understanding are not bugs to be patched, but inevitable consequences of building intelligence on the wrong mathematical foundations. His work redefines intelligence not as an asset accumulated by large models, but as a verifiable flow of meaning emerging from a network — and provides the architecture to put that definition into practice..

For more about Eric's background, visit his LinkedIn profile. 

Mike Counihan

Mike Counihan

Founding Strategic Architect

brings over 20 years of experience at the intersection of entrepreneurship, strategic finance, and early-stage venture growth. As Managing Partner of Transcend Partners, he has led more than 20 M&A and fundraising transactions for high-growth technology companies across Europe. A former tech founder and CEO himself, Mike also has deep roots in the startup ecosystem, having taught or coached hundreds of founders and entrepreneurs through his work with leading business schools and accelerators including ESADE, SeedRocket, and Startupbootcamp.

For more about Mike's background, visit his LinkedIn profile.