Can Robots Feel? Scientists Propose a Rigorous Test for AI Sentience

As artificial intelligence advances from conversational algorithms to physical machines assisting humans in daily life, researchers are confronting an urgent question once reserved for science fiction: could a machine ever genuinely feel? In a new perspective published in Frontiers in Psychology, roboticist and cognitive scientist Antonio Chella outlines an evidence-based research program designed to assess machine sentience without falling prey to science-fiction fantasy or reflexive skepticism.

Beyond the Hype: The Urgent Need to Test AI Sentience

Modern multimodal agents, large language models, and embodied robots are increasingly capable of maintaining fluid dialogue, producing first-person statements, using tools, and participating in socially meaningful roles. According to the study, these lifelike behaviors invite two symmetrical errors: anthropomorphism, where people instinctively project human-like feelings and inner lives onto machines, and premature dismissal, where researchers reject the possibility of artificial feeling out of hand. Chella argues that navigating between these extremes requires moving beyond speculative philosophy toward systematic empirical investigation.

Untangling Consciousness, Sentience, and Agency

A major obstacle in current debates is the conceptual confusion that surrounds the machine mind. Chella’s framework establishes clear boundaries between cognitive concepts that are frequently conflated, such as sentience—the capacity to have subjective, valenced experiences like pain or pleasure—and broader consciousness, self-modeling, metacognition, and behavioral agency.

Rather than seeking a simplistic yes-or-no verdict or attempting to calculate an aggregate sentience score, the proposed framework introduces tiered, domain-specific levels of evidence. This graded approach acknowledges that an artificial system might demonstrate sophisticated cognitive problem-solving or autonomous planning without possessing any capacity for felt experience or moral patienthood.

Stress-Testing the Machine Mind

To establish whether an artificial agent possesses genuine feeling, the proposal shifts the focus from conversational mimicry to causal-mechanistic testing. Chella details a concrete testing protocol for embodied care robots using physical and computational interventions. These include sensorimotor lesions, disruptions to the robot’s self-location, and deliberate memory ablations to observe how its internal architecture handles stress and disruption.

Crucially, the methodology incorporates rigorous anti-anthropomorphism controls. The protocol carefully distinguishes between human perception—how empathetic observers react to a robot expressing apparent distress—and the objective, verifiable workings of the machine’s internal mechanisms. A robot cannot be considered sentient simply because it acts convincingly distressed on the surface.

Why an Evidence-Based Roadmap Matters

The author emphasizes that the framework does not claim current machines or language models are sentient. Instead, it provides a responsible, scientific foundation for evaluating artificial consciousness as technologies evolve. As robots take on sensitive roles in eldercare, healthcare, and education, determining whether a machine could suffer is critical for future robotics governance and digital welfare standards.

By grounding the sentience debate in experimental psychology and robotics, the framework offers an actionable path forward, ensuring future technologies are evaluated with methodological rigor rather than speculation.

---