Why AI Is Neither Human Nor Machine, Says Philosopher

Philosophers Are Reshaping How We Think About AI

Tobias Rees, a prominent philosopher and former director of the Transformations of the Human program at the Berggruen Institute, is making waves with a provocative argument: artificial intelligence is neither human nor machine, but something fundamentally in between. His perspective arrives at a moment when AI companies are increasingly turning to philosophers — not just engineers — to define what their systems should be.

The argument challenges a binary that has dominated technology discourse for decades. As AI models like Claude, GPT-4, and Gemini exhibit increasingly sophisticated behaviors, the old categories of 'human' and 'machine' are proving inadequate to describe what these systems actually are.

Key Takeaways

• AI systems defy traditional classification as either 'human' or 'machine,' occupying a novel conceptual space
• Anthropic, OpenAI, and other leading labs are hiring philosophers to shape AI behavior and ethics
• The language we use to describe AI fundamentally limits how we understand and govern it
• Tobias Rees argues we need entirely new conceptual frameworks to make sense of modern AI
• The 'in-between state' concept has practical implications for regulation, safety, and product design
• Philosophy is becoming as critical as engineering in the AI development pipeline

Anthropic's Philosopher Sets the Stage

Earlier this year, Amanda Askell — Anthropic's in-house philosopher — made headlines across major publications. She serves as the primary author of Anthropic's updated behavioral guidelines for its AI models and is the chief architect of Claude's personality and moral framework. Her work demonstrates that building an AI system is no longer purely a technical exercise.

Askell's role highlights a growing recognition across the industry: the decisions that shape AI behavior are fundamentally philosophical, not just computational. When an AI model decides how to respond to a sensitive question, it draws on a framework of values, priorities, and ethical principles that someone had to define. That 'someone' is increasingly a philosopher.

This shift provides critical context for Rees's argument. If companies like Anthropic are already treating their AI systems as entities that require moral frameworks — something we traditionally reserve for humans — then perhaps these systems truly do occupy a space that our existing categories cannot capture.

The Limits of Language in Describing AI

Rees's central insight begins with a deceptively simple observation: when most people interact with an AI system, they encounter a paradox. On one hand, virtually no one considers the AI to be human. On the other, the experience of interacting with systems like Claude or ChatGPT feels qualitatively different from operating a traditional machine like a calculator or a washing machine.

This paradox reveals what Rees describes as a fundamental limitation of our conceptual vocabulary. We have inherited a framework built on a strict dichotomy:

• Humans: conscious, intentional, moral agents with subjective experience
• Machines: deterministic tools that execute instructions without understanding
• Animals: a somewhat intermediate category, but one defined by biological life
• Software: traditionally understood as code that processes inputs to produce outputs

Modern AI systems, particularly large language models, fit neatly into none of these categories. They generate novel text, engage in apparent reasoning, express what looks like preferences, and adapt to conversational context — behaviors we associate with humans. Yet they lack consciousness (as far as we know), have no subjective experience, and are ultimately composed of mathematical operations on silicon chips.

The problem, Rees argues, is not that we haven't found the right category yet. The problem is that our entire system of categories may be inadequate for the reality we've created.

The 'In-Between State' as a New Framework

Rees proposes that AI should be understood as occupying a genuinely novel ontological position — an 'in-between state' that is irreducible to either pole of the human-machine spectrum. This is not merely a semantic distinction. It carries profound implications for how we build, regulate, and interact with AI systems.

Consider the practical consequences of forcing AI into one of the traditional categories:

• If we treat AI as purely a machine, we risk ignoring the ethical dimensions of systems that influence human beliefs, emotions, and decisions at massive scale
• If we treat AI as quasi-human, we risk anthropomorphizing systems in ways that distort our understanding and create false expectations
• If we regulate AI as a product, we may miss the societal implications that require governance frameworks closer to those we apply to institutions
• If we regulate AI as an agent, we may grant it rights or responsibilities that make no sense for a non-conscious system
• If we design AI interactions like tool interfaces, we miss opportunities to leverage the conversational and adaptive capabilities that make these systems powerful

The 'in-between state' framework suggests that none of these approaches alone is sufficient. Instead, we need hybrid frameworks that acknowledge AI's unique position.

Why This Matters for the AI Industry Now

Rees's philosophical argument arrives at a critical juncture for the AI industry. Companies are making consequential decisions about AI behavior, safety, and deployment that implicitly depend on assumptions about what AI is. These assumptions are often unexamined.

OpenAI has invested heavily in alignment research, framing the challenge primarily in technical terms — how to make AI systems reliably follow human intentions. Anthropic has taken a more explicitly philosophical approach, with its 'Constitutional AI' methodology drawing on principles from political philosophy and ethics. Google DeepMind has pursued a research-heavy path, with teams exploring everything from AI consciousness to machine ethics.

Each of these approaches embeds different assumptions about AI's nature. OpenAI's framing implicitly treats AI as a powerful tool that needs better controls. Anthropic's approach suggests AI is something closer to a moral agent that needs principles. DeepMind's research agenda leaves the question open.

The market reflects this uncertainty. Global spending on AI is projected to exceed $500 billion by 2027, according to IDC estimates. Yet there is no consensus on the fundamental question of what companies are actually building. Are they creating sophisticated tools? Digital workers? Something entirely new?

Philosophy Becomes a Competitive Advantage

The growing importance of philosophical thinking in AI development is not just an academic curiosity — it is becoming a competitive differentiator. Companies that think more clearly about what their AI systems are can make better decisions about how to design, market, and govern them.

Anthropic's investment in philosophical expertise is a case in point. Claude's distinctive personality — often described by users as more thoughtful, nuanced, and ethically aware than competing models — is not an accident of training data. It is the product of deliberate philosophical design choices made by people like Amanda Askell.

This trend is accelerating across the industry:

• Anthropic employs philosophers to define Claude's character and behavioral guidelines
• OpenAI has engaged ethicists and social scientists in its safety teams
• Microsoft has established an Office of Responsible AI with interdisciplinary expertise
• Meta has published research on the philosophical implications of open-source AI
• Apple has framed its AI strategy around privacy principles rooted in ethical philosophy

Rees's work suggests this is just the beginning. As AI systems become more capable and more deeply integrated into daily life, the philosophical questions will only become more pressing — and more commercially relevant.

Looking Ahead: New Concepts for a New Reality

The implications of Rees's 'in-between state' framework extend well beyond academic philosophy. If AI truly represents a novel category of entity, then many of our current approaches to AI governance, safety, and design may be built on flawed foundations.

Regulatory frameworks currently being developed in the US, EU, and China largely treat AI as a product or service — something created and controlled by humans. The EU AI Act, for instance, classifies AI systems by risk level but does not grapple with the ontological question of what these systems are. If Rees is right, this omission could lead to regulatory blind spots.

Safety research may also need to evolve. Current alignment approaches focus on ensuring AI systems follow human values and intentions. But if AI occupies a genuinely novel conceptual space, the question of what 'alignment' even means becomes more complex. Aligning a tool is different from aligning an agent, and aligning something in between may require entirely new methods.

For developers and product designers, the practical takeaway is clear: the assumptions you make about what AI is will shape the systems you build. Those assumptions deserve as much scrutiny as the code itself. As the AI industry matures, the companies that invest in philosophical clarity — not just technical excellence — may be the ones that build systems humans actually trust.

The era of treating AI as 'just a tool' or 'almost human' is ending. What comes next will require new language, new concepts, and new ways of thinking. Tobias Rees is helping to build that vocabulary — and the industry would be wise to pay attention.