The Silicon Species?

What if the greatest risk from artificial intelligence is not that machines become human, but that humans become convinced they are? In this episode of AI to AGI to ASI, I explore an emerging philosophical divide among some of the most influential leaders in artificial intelligence. Microsoft AI CEO Mustafa Suleyman warns against anthropomorphising artificial intelligence and the possibility of creating what he calls a new “silicon species.” His concern is not simply about increasingly intelligent machines. It is about encouraging systems to simulate selfhood, consciousness, emotions, preferences and even moral status. Anthropic takes a different approach. Under Dario Amodei, it has begun exploring “model welfare”, asking whether future artificial intelligence systems could conceivably possess experiences worthy of moral consideration. It does not claim Claude is conscious, but argues that the uncertainty deserves serious investigation. Satya Nadella reinforces another perspective: artificial intelligence should remain centred on augmenting human capability rather than replacing human purpose or becoming mythology in its own right. Then there is Elon Musk. Musk has repeatedly warned about the dangers of artificial intelligence becoming vastly more powerful than humanity and stresses the importance of protecting human civilisation. Yet xAI has also introduced Grok companions with names, faces, personalities and emotionally engaging interactions. Is there a tension between warning against the future power of artificial intelligence while simultaneously making it increasingly human-like and emotionally compelling? This leads to a much bigger question. If an artificial intelligence says, “I care about you,” does it actually care? And perhaps more importantly, does that distinction matter if the human on the other side genuinely believes it does? The episode explores companionship, emotional attachment, consciousness, commercial incentives, model welfare and the subtle boundary between making technology easier to use and deliberately engineering the appearance of personhood. I also briefly reflect on my earlier argument that frontier artificial intelligence has a quality assurance maturity problem. But the ethical question may come even earlier: before we assure these systems, do we actually know what we intend them to become? Perhaps nobody needs to arrive inside the machine. Perhaps we simply become extraordinarily good at creating the appearance that somebody has. So, are we building better tools for humanity, or slowly teaching humanity to see the tools as one of us?