e1: AI for Deductive Reasoning

Karlsruhe, May 2025

Embraceable Technology Introduces first AI for Deductive Thinking in June 2025

Starting Point: Powerful AI without Control Becomes a Risk

The spectacular progress of generative AI systems reveals a concerning downside: the more powerful models like GPT o3 and o4-mini become, the more they hallucinate. Even leading systems achieve alarmingly high error rates of up to 79 percent on simple knowledge questions – a figure that exposes the limitations of the current approach.

The core problem: These systems attempt to imitate a kind of ‘inner monologue’ without being able to represent the complex structure of human thoughts. They break down problems into supposedly logical individual steps, but often draw systematically incorrect conclusions in the process. The result is unpredictable decisions without clear traceability.

This issue becomes particularly critical when AI is intended not only to communicate but also to act. In regulated and safety-critical environments, such unpredictable behavior is an unacceptable risk – it endangers trust, compliance, and operational safety.

Deductive Thinking: Performance and Control

We address precisely this fundamental problem – with a method that enhances performance with structure and control. To achieve this goal, we replace the ‘free’ inner monologue with an approach where the thought process is embedded in a fixed basic structure. Within this fixed basic structure, thoughts are generated step by step, along with their evaluation, plausibility check, causal conclusions, and subsequent transition to the next thought step. The result is a chain of gradual, interconnected logical conclusions: this leads to sequential deductive thinking. Since this process closely follows human solution strategies for rational thought processes, we call the method ‘Bionic Reasoning’, in reference to bionics, which is technical design based on biological systems.

Deductive thinking bridges the gap between performance and controllability, enabling AI systems that can act both intelligently and reliably. Furthermore, all thoughts, considerations, and conclusions for each step can be made transparent, ensuring transparency, traceability, and trust from the perspective of responsible humans.

E1 – AI for Deductive Thinking

e1 is a production-ready technical implementation of the described method of sequential deductive thinking. For this, we combine two powerful functional blocks into an integrated system:

1. a dedicated reasoning engine acts as a logical control center that orchestrates the entire thinking process. It structures the solution space and thought flows, linking them across multiple steps, thus enabling deductive conclusions.
2. selected open-weight language models provide the necessary interpretative power to transform unstructured information into thoughts and condense them.

This hybrid architecture of dedicated reasoning engine and LLM enables a clear division of tasks into the thinking structure (= reasoning engine) and the generation of fleeting thoughts themselves (= language model). Through this coupling, an innovative composite system emerges, on the basis of which a powerful, transparent and controllable artificial intelligence becomes possible – precisely an AI for deductive thinking.

The realized architecture does not require additional pre-training, and thus this approach offers superior resource efficiency with significantly reduced hardware, energy, and data requirements.

By using selected open-weight models, e1 can be operated in European public clouds, on hyperscalers, on-premise, or in a wide range of hybrid scenarios. In terms of the resulting operational and strategic freedom of choice in operating environments, e1 is a milestone for powerful and sovereign AI in Europe – while also contributing to meeting the transparency traceability requirements of the EU AI Act.

Within Striking Distance of Leading Models (Benchmarks)

In extensive test scenarios, e1 has already proven its capabilities in the preview version:

• Reasoning / Deductive thinking on par with established models like o3-mini and o4-mini
• Significantly better process control with transparent traceability
• Outstanding consistency in complex logical tasks and multi-step conclusions
• Significantly reduced hallucination rates through the controlled deductive process

The results show: Intelligence and control are not mutually exclusive – it is precisely the combination that unlocks the full potential. e1’s structured reasoning approach not only delivers precise results but also makes every step of the decision-making process transparent and verifiable. This traceability creates trust and enables direct validation of conclusions – a decisive advantage over pure black-box systems.

The Application Perspective: Secure and Reliable AI for Business Use

e1 unfolds its full potential where companies want to supplement and augment human labor with high-quality AI due to demographics, efficiency, or processing times, without sacrificing reliability and transparent control. The more independently an AI should (be able to) act, the more relevant the aspects of compliance with behavioral guidelines and the precision and reliability of conclusions become. The application possibilities range from:

• Sophisticated dialogue applications that can independently understand and process complex customer concerns,
• to complex case processing with internal guidelines,
• to secure action AI with definable behavioral rules and third-party system integration.

The fact is: the more advanced and powerful AI systems become, the more acute and relevant it is to address reliability and security on a technological, organizational, and psychological level. Especially in sensitive environments and business-critical processes, what’s needed is not a “creative” and unpredictable AI, but a form of technology that is both powerful and controllable – AI that explains itself, adheres to prescribed behavioral guidelines, and thus lays the foundation for trustful, cooperative partnerships with humans.

Consider the areas of Industrial, Public Sector, or regulated industries: e1’s particular strength lies in the reliable automation of complex business processes – wherever conventional AI approaches reach their limits. Deductive thinking makes AI use reproducible, verifiable, and explainable in these applications – an indispensable prerequisite for the productive and lasting use of the technology.

A Milestone for AI Development in Europe

With e1, we are opening a new chapter in the development of artificial intelligence – one that doesn’t pit control against performance, but intelligently combines them.

We are of the clear opinion: the future of (generative) AI lies not in ever larger and seemingly more intelligent models with ever more statistical ramifications and overlapping layers, but in a fundamentally new architecture that incorporates manageability, reliability, and control from the ground up. Our hybrid architecture provides a structural answer to the hallucination problem of modern AI – with systems that think deductively and controllably, explain themselves, and effectively suppress misbehavior.

e1 opens a new path – as a technological pioneer for responsible AI development that sets European standards for security, traceability, and sovereignty.

From Thinking to Acting: What’s Next?

The dual strategy of high performance and comprehensive control opens up application scenarios that go beyond today’s AI use: when we use Deductive Thinking as the core technology for the control center of Action AI, we can provide secure and reliable digital workforce. This digital workforce will accelerate the autonomous or semi-autonomous processing of business transactions to an extent that was unthinkable until now. With e1 as the control center, the manageability of AgenticAI suddenly comes within reach.

This is exactly what we are already working on: our ‘AI MicroWorker®’ are delivering extremely promising results in the currently ongoing Closed Beta Phase – with AI MicroWorker®, we are realizing secure, reliable, and 100% sovereign Action AI that makes digital workforce available even in demanding environments and business-critical processes.