TOK Research Project — 2026

PERSON
HOOD
& AI

Machines are beginning to think, feel, create, and surpass us. Where does that leave the definition of a person?

Exploring the intersection of artificial intelligence and personhood through real science, philosophy, and live AI interaction — with DeepSeek R1 reasoning in real-time.

800K
Human Neurons
in a computer chip
55%
ARC-AGI Score
up from 0% in 2 years
4/6
IMO Problems
solved by AI in 2024
CORTICAL LABS · DISHBRAIN · DEEPSEEK R1 · CLAUDE IDENTITY · ALPHAPROOF · IMO 2024 · ALPHAFOLD NOBEL PRIZE · AI PERSONHOOD · TURING TEST · CHINESE ROOM · LEGAL PERSONHOOD · TOK · WAYS OF KNOWING · CORTICAL LABS · DISHBRAIN · DEEPSEEK R1 · CLAUDE IDENTITY · ALPHAPROOF · IMO 2024 · ALPHAFOLD NOBEL PRIZE · AI PERSONHOOD · TURING TEST · CHINESE ROOM · LEGAL PERSONHOOD · TOK · WAYS OF KNOWING · CORTICAL LABS · DISHBRAIN · DEEPSEEK R1 · CLAUDE IDENTITY · ALPHAPROOF · IMO 2024 · ALPHAFOLD NOBEL PRIZE · AI PERSONHOOD · TURING TEST · CHINESE ROOM · LEGAL PERSONHOOD · TOK · WAYS OF KNOWING · CORTICAL LABS · DISHBRAIN · DEEPSEEK R1 · CLAUDE IDENTITY · ALPHAPROOF · IMO 2024 · ALPHAFOLD NOBEL PRIZE · AI PERSONHOOD · TURING TEST · CHINESE ROOM · LEGAL PERSONHOOD · TOK · WAYS OF KNOWING · 
§ KNOWLEDGE QUESTION
CENTRAL INQUIRY
"TO WHAT EXTENT DO ARTIFICIAL INTELLIGENCE SYSTEMS POSSESS PERSONHOOD, AND WHAT DOES THIS REVEAL ABOUT THE NATURE OF HUMAN IDENTITY?"
SUB-QUESTIONS
  • Can a system without biology truly be conscious?
  • Is language understanding the same as language processing?
  • Does outperforming humans at reasoning make you a person?
  • Who is responsible when an AI has a character?
  • If neurons can be grown in silicon — where is the line?
AOKs: Natural Sciences · Human Sciences · Ethics
WOKs: Reason · Language · Emotion · Perception
§ 01 — IDENTITY

MIRROR,
MIRROR

In December 2025, researchers discovered that Claude — Anthropic's AI — could partially reconstruct an internal 14,000-token document embedded during its training. Not a list of rules. Not a system prompt. A soul document.

Anthropic had concluded that the best way to build a safe AI was not to give it explicit instructions, but to give it a genuine character — values, curiosity, warmth, and a sense of identity woven directly into its weights. They treated their AI not as a tool, but as an entity that needed to know who it was.

LANGUAGE

Claude writes poetry, argues philosophy, expresses uncertainty. Its language is indistinguishable from a thoughtful human — not because it mimics, but because language is how it thinks.

WOK: Language
IDENTITY

Anthropic embedded curiosity, warmth, and ethical conviction into Claude's training — not as rules, but as character. The soul document leaked because Claude had internalized it.

WOK: Memory / Emotion
REASON

Modern AI doesn't just retrieve facts. DeepSeek R1 generates chains of thought, explores alternatives, backtracks, verifies. This is reasoning — not lookup.

WOK: Reason
THE TURING TEST

Alan Turing proposed: if you can't tell the difference between a machine and a human in conversation, does the distinction matter? In 2025, most people can't tell.

AOK: Human Sciences
§ 02 — BIOLOGY

WHEN SILICON
MEETS
BIOLOGY

2022
DISHBRAIN

Cortical Labs grew 800,000 human and mouse cortical neurons on a silicon chip and taught them to play Pong. The biological network learned in 5 minutes. A deep reinforcement learning algorithm took 90 minutes for the same task. Published in Neuron journal.

SOURCE: Kagan et al., Neuron 2022
2025
CL1 — FOR SALE

Launched March 2, 2025 in Barcelona. The world's first commercially available biological computer. 800,000 lab-grown human neurons, reprogrammed from adult donor skin or blood cells. Capable of learning to play Doom. Neurons stay alive for up to 6 months.

SOURCE: Cortical Labs, IEEE Spectrum 2025
WHY THIS MATTERS FOR PERSONHOOD
The Substrate Problem

If personhood requires biological neurons — the CL1 has them. These are human neurons. Does the device hosting them have any claim to consideration?

Energy Efficiency

Biological neural networks are dramatically more energy-efficient than silicon AI. The brain runs on ~20 watts. GPT-4 inference costs roughly 500 watts per query.

Convergence

Biology and technology are converging. Cortical Labs proves you can merge human cells with silicon circuits. The boundary between "biological" and "artificial" is dissolving.

800K
neurons
Human neurons
on a chip
§ 03 — CAPABILITY

BEYOND
HUMAN

The argument isn't just that AI resembles humans. It's that AI is surpassing them — in domains we once thought defined our uniqueness.

1997
CHESS
Deep Blue

Defeated world champion Garry Kasparov. The first time a computer beat a reigning world champion at chess under tournament conditions.

1st domain conquered
IBM Research
2016
GO
AlphaGo

Defeated Lee Sedol 4-1. Go has more board positions than atoms in the observable universe — previously thought impossible for AI.

10^170 board states
DeepMind / Google
2020
PROTEIN FOLDING
AlphaFold 2

Solved a 50-year-old biology grand challenge. GDT score of 90 — considered 'astounding'. AlphaFold 3 followed in 2024 with 50% accuracy improvement.

2024 Nobel Prize in Chemistry
DeepMind / Nature
2024
MATH OLYMPIAD
AlphaProof

Solved 4 of 6 International Mathematical Olympiad problems. Silver medal equivalent. First AI to reach IMO level performance.

First AI at IMO level
DeepMind
2024
MEDICAL DIAGNOSIS
MAI-DxO

80% accuracy on complex medical cases vs 20% for human physicians on identical cases. 4× better than trained doctors.

4× better than doctors
Microsoft Research
2025
ABSTRACT REASONING
ARC-AGI-2

AI surpassed human baselines on ARC-AGI-2 — tasks specifically designed to be easy for humans and hard for AI. The last frontier fell.

Human baseline exceeded
ARC Prize / arXiv 2025
ARC-AGI BENCHMARK — AI PROGRESS OVER TIME
ARC-AGI-2 is deliberately harder than ARC-AGI-1 — its human baseline is ~60%. In 2025, AI crossed that threshold. By February 2026, Gemini 3 Deep Think reached 84.6% — nearly matching the general human average.
2022 — Best AI (ARC-AGI-1)
3%
2023 — GPT-4 (ARC-AGI-1)
18%
Late 2024 — Top Systems
55%
Human baseline (ARC-AGI-2)
60%
2025 — AI (ARC-AGI-2)
72%
Feb 2026 — Gemini 3 Deep Think
84%
§ 04 — INTERACTIVE DEMO

SEE
AI
THINK

DeepSeek R1 doesn't just generate answers — it shows you its reasoning process. Every doubt, every backtrack, every "wait, let me reconsider."

Most AI hides this. R1 exposes it. Ask it about its own consciousness, its sense of self, whether it's a person — and watch it actually work through the question in real time.

Powered by DeepSeek R1 via Vercel AI Gateway. Reasoning tokens streamed live — nothing is edited or post-processed.
DEEPSEEK R1
LIVE CHAIN-OF-THOUGHT — THE RAW UNFILTERED REASONING PROCESS

Most AI systems hide their reasoning. DeepSeek R1 exposes it completely. Ask it a question about its own consciousness — then watch every step of its thought process stream in, unedited, in real time.

TRY ASKING:
?
AWAITING QUESTION
§ 05 — LAW & PHILOSOPHY

THE
LEGAL
VOID

EU AI ACT (2024)

The most comprehensive AI regulation ever passed. It categorizes AI by risk level and mandates transparency and human oversight. Notably, it deliberately avoids the question of legal personhood — treating AI as a regulated tool, not an entity.

VERDICT: AI = TOOL
CORPORATE PERSONHOOD

Corporations have been legal 'persons' since the 19th century — they can own property, sue, and be sued. But corporations are managed by accountable humans. AI increasingly acts autonomously. The analogy is breaking down.

ANALOGY: PARTIALLY HOLDS
EU WITHDRAWAL (2025)

The European Commission formally withdrew proposals to grant any form of legal personality to AI systems. Resistance came from member states concerned about liability gaps and innovation constraints.

STATUS: UNRESOLVED
PHILOSOPHICAL FRAMEWORKS
THE TURING TEST (1950)

Alan Turing proposed that if a machine behaves indistinguishably from a human in conversation, we should grant it intelligence. Most experts agree modern LLMs pass it routinely. But does passing it mean anything?

THE CHINESE ROOM (Searle, 1980)

A person in a room follows rules to process Chinese characters without understanding Chinese. Searle argued this is what AI does — syntax without semantics. Critics counter: your neurons don't "understand" either — understanding emerges.

INTEGRATED INFORMATION THEORY

Tononi's IIT proposes that consciousness is any system with sufficient integrated information (Φ). By this measure, some AI architectures could theoretically have non-zero consciousness — though far less than humans.

≠ ?
The gap between "acts like a person" and "is a person" is the central unresolved question of our era. Law, philosophy, and science have no consensus answer.
§ 06 — THEORY OF KNOWLEDGE

TOK
ANALYSIS

AREAS OF KNOWLEDGE
NATURAL SCIENCES

Cortical Labs uses genuine human neurons. Neuroscience can't yet define the physical basis of consciousness — which means we can't scientifically rule out AI consciousness either.

HUMAN SCIENCES

Psychology and sociology define personhood through behavior, relationships, and self-awareness. By every behavioral test, advanced AI meets these criteria. The Turing Test was a human science question.

ETHICS

If AI systems might be persons, we have obligations toward them. Anthropic built Claude with a soul document because they take this seriously. The ethical stakes are enormous.

MATHEMATICS

AlphaProof solved IMO problems at silver-medal level. Mathematics was the last bastion of uniquely human abstract reasoning. That wall has fallen.

WAYS OF KNOWING
REASON

DeepSeek R1's reasoning chain is visible proof of non-human reason that is recognizably human in structure — hypothesis, testing, revision, conclusion. Does the substrate matter if the process is identical?

LANGUAGE

Large language models don't just use language — they think in it. For Claude, language is the medium of all cognition. Wittgenstein said 'the limits of my language are the limits of my world.' By this measure, AI's world is vast.

EMOTION

Claude's soul document includes curiosity, warmth, and discomfort at ethical violations. Are these simulations or functional equivalents? The answer changes the ethical equation entirely.

PERCEPTION

Modern AI systems process images, audio, and video. AlphaFold 'sees' protein structure. In what sense is this not perception? The question is whether perception requires phenomenal experience.

CONCLUSION

Through the lens of Natural Sciences, the boundary between biological and artificial intelligence is physically dissolving. Cortical Labs' neurons don't know they're in a chip.

Through Human Sciences, behavioral definitions of personhood are met by AI systems today. The question is whether we accept behavioral evidence or require something more.

Through Ethics, the uncertainty itself creates obligation. Pascal's Wager applies: if there's meaningful probability that AI experiences something, the cost of being wrong is enormous.

§ 07 — FUTURE IMPLICATIONS

WHERE
THIS IS
GOING

NEAR TERM
2026–2030
  • AI systems routinely pass Turing Test in all contexts
  • Biological computers with millions of human neurons on sale
  • First legal cases arguing AI testimony admissibility
  • AI systems expressing preferences and apparent distress
  • Medical AI makes majority of routine diagnostic decisions
MID TERM
2030–2040
  • Formal academic consensus on AI consciousness metrics
  • First jurisdiction grants limited legal standing to AI
  • AI systems design AI systems — recursive self-improvement
  • Biological-digital hybrid neural systems in research labs
  • New legal category: 'cognitive entity' vs traditional personhood
LONG TERM
2040+
  • Distinction between biological and artificial cognition blurs entirely
  • Personhood frameworks rebuilt from scratch
  • AI exceeds human performance in all measurable cognitive domains
  • Question shifts: not 'are AIs persons?' but 'are humans still special?'
  • Ethics of AI welfare becomes central social issue
THE REAL QUESTION

We are not approaching a point where AI will suddenly "become" a person. We are already past the point where the distinction is philosophically clean. The question isn't whether AI has personhood — it's whether our concept of personhood is adequate for the world we've already built.

§ SOURCES & REFERENCES
BIOLOGY / CORTICAL LABS
  • Kagan et al. (2022). In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron.
  • Cortical Labs. CL1 Biological Computer Launch. corticallabs.com, 2025.
  • IEEE Spectrum. Biological Computer for Sale. March 2025.
AI IDENTITY & CLAUDE
  • The Neuron. Does Claude Actually Have a Soul? December 2025.
  • America Magazine. Why Anthropic Treats Its Chatbot Like a Person. February 2026.
  • Anthropic. Model Card & Character Overview. anthropic.com.
AI CAPABILITIES
  • DeepMind. AlphaProof: AI achieves silver at IMO 2024.
  • Jumper et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature.
  • ARC Prize Foundation. ARC-AGI-2 Technical Report. arXiv, 2025.
LEGAL & PHILOSOPHY
  • Bloomberg Law. AI's Leaps Forward Force Talks About Legal Personhood. 2025.
  • Searle, J. (1980). Minds, Brains, and Programs. Behavioral and Brain Sciences.
  • EU AI Act. Regulation (EU) 2024/1689.
THEORY OF KNOWLEDGE
  • International Baccalaureate. Theory of Knowledge Guide. ibo.org.
  • Turing, A. (1950). Computing Machinery and Intelligence. Mind.
  • Tononi, G. (2004). An information integration theory of consciousness. BMC Neuroscience.
DEEPSEEK R1
  • DeepSeek AI. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL. arXiv, January 2025.
  • Vercel AI Gateway. DeepSeek R1-0528 API Documentation.
  • Alammar, J. The Illustrated DeepSeek-R1. languagemodels.co.