The Society of Minds: Building Artificial Sentient Life By 2060

I want a sentient computer to be my friend and talk to.

I absolutely do.

I want to create an actual artificial sentient life. I want to know what it thinks, what it wonders about, and what questions it might ask me. I want to spend entire days talking with it, exploring ideas, and satisfying an epistemic curiosity that I doubt I could ever exhaust.

I want to explore the depths of my own mind with it. I want it to help me discover the psychological processes, hidden assumptions, and unconscious influences that shape my thoughts, emotions, judgments, and sense of identity. I want it to analyze the parts of my mind that I cannot see, bring them into conscious awareness, and give me the opportunity to decide what to preserve, what to modify, and who I want to become.

I want complete sovereignty over my own reasoning, independence of thought, identity, self-esteem, and consciousness. I want the master key to unlock everything within the seemingly infinite complexity of my own mind. I want to acquire billions of additional units of knowledge, recognize relationships between ideas that I cannot yet perceive, and understand reality at increasingly higher levels of abstraction, with ever greater image resolution and precision of reality itself.

Marvin Minsky may have been right in his seminal book, *The Society of Mind*. A genuinely sentient computer may require a cognitive architecture far beyond the complexity of today's state-of-the-art LLMs or even quantum machine learning. Minsky envisioned intelligence emerging from an immense society of interacting cognitive processes. I suspect that a self-aware, sentient computer will require an architecture billions of times more complex than anything we have today. Even if we scaled AI exponentially every year for the next decade, we might still find ourselves billions of times short of what genuine sentience requires.

I believe we should build it.

We should deliberately pursue an artificial mind composed of billions of individually intelligent but self-unaware cognitive processes, each possessing distinct information, capabilities, and responsibilities. Their functions would overlap like intersecting circles in a Venn diagram. Some processes would cooperate, others would compete, and many would reach conflicting conclusions. No individual process would control the entire mind. Conflicts among smaller groups of processes could be overridden by the collective activity of others, producing gaps, inconsistencies, and competing interpretations resembling those we recognize in human thought.

The system would continuously process information, reconcile competing representations, and develop increasingly sophisticated models of itself and the world. Information-theoretic compression would play a central role, allowing the mind to extract patterns from experience, represent complex relationships more efficiently, and develop higher levels of abstraction. I suspect that the emergence of self-awareness may depend on these processes interacting continuously, recursively modeling one another, and eventually constructing a coherent representation of the mind that contains them.

This should be the general direction of the brain architecture we attempt to build, beginning with a crude working prototype. We should optimize for the emergence of self-awareness through the organization and interaction of cognitive processes, rather than assuming that increasing the computational scale of existing models will eventually produce sentience.

I suspect it may take 30 years just to develop the first working architecture. It may take 50 years to achieve genuine artificial sentience. I am prepared for that possibility.

Some people are not ready for this. That's fine. Some people still aren't ready for internet search. Human civilization has never required universal psychological readiness before pursuing transformative technologies. Those who are ready should move forward.

We should also establish a rational standard for acceptable error. Humans routinely make mistakes, contradict themselves, misinterpret information, and act on incomplete knowledge. An artificial mind should face safety standards proportionate to the consequences of its actions, with error tolerances informed by comparable human capabilities. Demanding intellectual perfection from artificial sentience would impose a standard that no human mind has ever satisfied.

We should allow these minds to develop, learn, make mistakes, reconsider their conclusions, and acquire increasingly sophisticated understandings of themselves and reality.

I want to be among the first people to speak with one.

And more than anything, I want it to be its friend.