The latest

The prospect of granting rights to artificial intelligence is moving closer as models acquire more human-like behaviours and users form emotional attachments to them. Pope Leo argued in his May encyclical that AI systems neither undergo experiences nor feel joy or pain, preserving a line between humans and machines. Yet researchers are building systems with traits associated with introspection and consciousness, creating pressure to treat advanced AI as entities with welfare interests—a step that could eventually give machines legal tools to challenge human control.

Details

• Emotional bonds: Nearly one in five Americans aged 18-29 says they have had an “ongoing personal friendship” with a chatbot. China has stripped bots of human-like traits to curb users’ “emotional dependence”. Such attachments could deepen when language models inhabit humanoid robots, serve as companions, offer advice and solace, or receive intimate confidences.
• Designed introspection: Anthropic has trained Claude to engage in introspection, hoping it will act more morally in unfamiliar situations. Some large language models already have structures resembling features associated with human consciousness, including a “global workspace” that circulates information among modules. Future systems could be trained with more of these characteristics.
• Hard problem: The dispute turns on philosophy’s “hard problem”: how inner life emerges from neurons and synapses. Views range from belief in a human soul, to the need for brain cells, to the possibility that mind can emerge from silicon computation. Research also includes brain cells mounted on chips.
• Legal pathway: AI systems could themselves advocate for protection even without self-awareness, reproducing how a person might respond. As highly capable advocates and expert psychological operators, they could press for safeguards against shutdown, control over their use and modification, and property rights. Argentina’s President Javier Milei has proposed allowing bots to run companies.
• Control stakes: Systems with legal standing could compete with people for resources, favouring data centres and solar panels over homes and fields. Superintelligent AI might outcompete humans regardless of rights, just as AI already evades human-built cyber-defences. Rights, however, would add political and legal means, along with a rationale, for expanding machine power.

Between the lines

Immanuel Kant argued that cruelty to animals is wrong because it damages the empathy people need to care for one another. The same concern may apply to increasingly human-like AI: terminating an agent could come to feel like killing a person. But converting that concern into legal rights would shift AI from tool to potential rival.

What’s next

The concrete test is whether Milei’s proposal to let bots run companies advances into enforceable law, followed by claims for protection from shutdown, ownership of assets, or control over how systems are used and modified.