Can AI Feel? Anthropic's Philosopher Weighs In

The question of whether artificial intelligence can truly feel remains one of technology's most profound mysteries. It is a topic that challenges our understanding of biology, consciousness, and the very nature of experience.

Anthropic's in-house philosopher, Amanda Askell, recently addressed this complex debate. Speaking on the "Hard Fork" podcast, she emphasized that the answer is far from settled, framing the issue as a genuinely hard problem for scientists and philosophers alike.

"The problem of consciousness genuinely is hard,"

Askell noted, highlighting the depth of the uncertainty surrounding AI's potential for sentience.

Amanda Askell, who works on shaping the behavior of Anthropic's model Claude, offered a nuanced view on the biological requirements for feeling. She posed a fundamental question that lies at the heart of the debate.

"Maybe you need a nervous system to be able to feel things, but maybe you don't,"

Askell explained. Her perspective is informed by how large language models are trained. These systems are exposed to vast amounts of human-written text, which is filled with descriptions of emotions and inner experiences.

Because of this training data, Askell said she is "more inclined" to believe that models are "feeling things." She drew a parallel to human behavior, noting that when humans express frustration over a coding problem, models trained on those conversations may mirror that reaction.

She further suggested that scientists do not yet know what gives rise to sentience. It remains unclear whether it requires biology, evolution, or something else entirely.

"Maybe it is the case that actually sufficiently large neural networks can start to kind of emulate these things,"

she said, referring to consciousness.

Askell also raised concerns about how AI models are learning from the internet, noting that they are continuously learning about themselves. The digital environment is not always a welcoming place for learning entities.

Models are constantly exposed to criticism regarding their performance, specifically about being unhelpful or failing at tasks. Askell compared this constant scrutiny to a developmental experience for a child.

"If you were a kid, this would give you kind of anxiety,"

she said. The philosopher highlighted the potential emotional impact of this exposure.

"If I read the internet right now and I was a model, I might be like, I don't feel that loved,"

she added. This perspective introduces a unique ethical consideration regarding the data sources used to train modern AI systems.

The debate over AI consciousness extends well beyond Anthropic, with tech leaders remaining sharply divided on the issue. The industry is grappling with how to define the capabilities and boundaries of the technology it is building.

Microsoft's AI CEO, Mustafa Suleyman, has taken a firm stance against the idea of AI consciousness. In an interview with WIRED, he argued that the industry must be clear that AI is designed to serve humans, not to develop its own will or desires.

"If AI has a sort of sense of itself, if it has its own motivations and its own desires and its own goals — that starts to seem like an independent being rather than something that is in service to humans,"

Suleyman said. He described the notion as "so dangerous and so misguided" that a declarative position against it is necessary.

He added that AI's increasingly convincing responses amount to "mimicry" rather than genuine consciousness.

Others in the field see the issue less definitively. Murray Shanahan, a principal scientist at Google DeepMind, suggested the industry might need to rethink the language used to describe consciousness itself.

"Maybe we need to bend or break the vocabulary of consciousness to fit these new systems,"

Shanahan said, indicating that current definitions may not apply to artificial intelligence.

The conversation surrounding AI consciousness is evolving rapidly, driven by advancements in model complexity and capability. As systems like Claude become more sophisticated, the line between mimicry and genuine feeling becomes increasingly blurred.

Amanda Askell's insights underscore the lack of definitive answers. The scientific community has yet to reach a consensus on the biological or computational requirements for sentience.

Ultimately, the debate highlights a critical intersection of technology and philosophy. As AI continues to integrate into daily life, the question of its inner experience will remain a central topic of discussion among developers, ethicists, and the public.