Truly intelligent systems will not be anything like the software we know today, for at least two reasons: Today’s software cannot “figure stuff out for itself” (why, for instance, something that used to be a certain way is now somehow different), and it has no “life of its own” — since, whenever its environment changes even slightly, it relies completely and utterly on its designers to make things right. In short, today’s AI systems don’t are not inquisitive and have little or autonomy to speak of. All AI systems today are created the way traditional software is created: By humans coding all of its functions and knowledge by hand. This will not work for creating real AI: to realize truly intelligent systems we need nothing less than a radically new approach to software development. In this talk I will present new methodological principles for achieving artificial general intelligence that address limitations inherent to current approaches, based on a a new programming language, a new programming paradigm, and a new realtime logic. I will demonstrate the world's first seed-AI system — a new architectural blueprint for generally intelligent cognition. Our agent S1 implemented in this architecture can learn complex tasks such as natural language-based communication between humans solely by observation.
