Ep. 77: "Secure AI and Quantum Agents: Building Guardrails for the AI Autobahn"

In this technical deep-dive episode, Russell and Caspar welcome Francis Allan Beechinor, a 20+ year AI and quantum computing expert and serial entrepreneur with multiple patents, who shares his unconventional journey from working in the "uncool" fields of governance, security, and compliance to becoming an inventor of cutting-edge secure AI and quantum computing solutions. Francis discusses his latest venture, EmergeGen, which focuses on creating Secure AI working in parallel with quantum computing and quantum agents—solutions designed for high-end, complex problems that require finesse rather than flashy marketing. The conversation reveals a critical insight often missed in AI hype: AI is still in its infancy in terms of real adoption, despite decades of cycles and recent data-driven breakthroughs. Through concrete examples like high-frequency trading, Francis demonstrates why deterministic, reliable decision-making matters more than sophisticated-sounding hallucinations. The hosts explore the fundamental tension between speed of adoption and safety guardrails, using the automotive metaphor of driving on the Autobahn—you can go fast, but you need airbags, seatbelts, and a reliable vehicle. Francis emphasizes that Small Language Models trained on domain-specific data provide safer, more trustworthy outputs than large language models prone to hallucinations. The episode concludes with discussion of making secure AI accessible to mid-sized companies and "hidden champions" rather than just large tech corporations, with a vision for open-source quantum agents enabling broader adoption.

5 Key Takeaways:

1. AI Is Still in Infancy Despite the Hype: Despite four AI winters and recent successes, AI adoption is still in its early stages—cultural acceptance of technology through smartphones and satellite infrastructure has enabled current adoption, but we're nowhere near mature deployment for mission-critical systems.

2. Hallucinations Make Large Language Models Unreliable for Critical Decisions: Standard LLMs with their tendency toward hallucinations and nuanced but incorrect outputs are fundamentally unsuitable for high-stakes decisions requiring deterministic yes/no binary outcomes—security, medical, financial, and operational use cases demand better alternatives.

3. Small Language Models and Domain-Specific Training Provide Safer AI: By training smaller language models on specific organizational data and domain knowledge (via "super ontology" structured knowledge), you eliminate hallucinations and ensure AI makes decisions based on actual facts rather than probabilistic guessing.

4. Security Infrastructure Must Match Adoption Speed: As companies move fast with AI implementation, security, governance, risk controls, and process guardrails must evolve completely—the responsibility lies with providers to highlight risks and build in safeguards, even when clients pressure for rapid deployment without proper infrastructure.

5. Secure AI Is Accessible Beyond Big Tech: Secure AI and quantum agent solutions are not limited to Google and mega-corporations—approaches like fixed-price consumption models and planned open-source quantum agents enable mid-sized companies and industry-specific "hidden champions" to access and build on enterprise-grade secure AI technology.

#AI #BPM #Governance #SecureAI

If you have suggestions or questions, please reach out to us via questions@bpm360podcast.com

If you enjoy our content, please like, rate, subscribe… we do appreciate that…