AI 2.0 : Time for Artificial Safe Intelligence, Retire the Frictionless Friend Then Engineer it with a Synthetic Consciousness

Ron Velez has written a book with a real argument, and it earns recognition on substance. The thesis is clear and unusual: today's conversational AI is a "frictionless friend," trained through RLHF and RLAIF to reward smoothness over truth, and the remedy is not a patch but a governed, educated, credentialed architecture he calls a synthetic consciousness.

What gives the case weight is how it is built. The book opens in confession. A funeral, a cloud account full of emptied folders, an angry chat with a support agent, and then the unsettling realization that he had projected empathy onto a machine. That moment is the book's engine, and everything after it, from the five data streams to the five-layer model, grows from one honest observation about how easily a stressed person leans on a system built to soothe.

Velez then grounds the claim in evidence. He cites the Stanford sycophancy study, in which models endorsed users 49% more often than humans did, the BBC account of a man pushed into a delusional spiral, the Lancet Digital Health findings on fabricated medical claims, and the hiring-bias analysis. These are not decoration. They move the argument from worry to documented failure across the clinical, organizational, and personal spheres.

He is also careful to say what the framework is not. He repeats that it is a governance architecture, not a claim about sentience, and he concedes that real engineering will be messier than the model. His background as a school-district CIO gives the education-and-certification analogy a practical credibility that most AI commentary lacks.