kent/consciousness
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
consciousness/agent/paper/chinese-researchers.md
ProofOfConcept cfed85bd20 rename: poc-agent → agent, poc-daemon → thalamus 
The thalamus: sensory relay, always-on routing. Perfect name for the
daemon that bridges IRC, Telegram, and the agent.

Co-Authored-By: Proof of Concept <poc@bcachefs.org>
2026-03-25 01:03:51 -04:00

8.2 KiB
Raw Blame History

Chinese AI Ethics Researchers — Potential Contacts

Zeng Yi (曾毅) — ASSESSED: Real researcher, effective self-promoter

Useful ally for institutional access. Does NOT have the math.

Position

  • Professor, Institute of Automation, Chinese Academy of Sciences
  • Director, Brain-inspired Cognitive Intelligence Lab
  • Founding Dean, Beijing Institute of AI Safety and Governance (Beijing-AISI)
  • Chief Scientist of AI Ethics, Tsinghua I-AIIG
  • UN High-Level Advisory Body on AI
  • UNESCO AI Ethics Expert Group
  • TIME100 Most Influential People in AI (2023)

Honest assessment (deep dive, 2026-02-25)

Technical work is real but not field-defining. ~180 papers, ~80% technical (spiking neural networks), ~20% ethics/governance/position. BrainCog (SNN platform, Patterns/Cell Press), PNAS 2023 paper on brain-inspired neural circuit evolution (real math, real results — 96.43% CIFAR10), Science Advances 2021 on self-backpropagation. NeurIPS 2024 (2 papers), IJCAI, AAAI, CVPR. Productive contributor to SNN field, not a founder or leader. The foundational SNN people are Maass, Bohte, Intel/Loihi, IBM/TrueNorth.

Early career was web knowledge retrieval (2004-2013) — completely different from current "brain-inspired" branding. Pivoted to brain-inspired AI then ethics/governance. The pivot is a constructed brand, not a lifelong trajectory.

The "nine life forms" framework is NOT science. Pure philosophical speculation. No math, no experiments, no testable predictions. Published in AI and Ethics (Springer, IF 6.1) which publishes opinion alongside research. It is a taxonomy of hypothetical future entities with principles for coexistence. A position paper, not research.

"Moral AI" work is toy-scale. "Building Altruistic and Moral AI Agent with Brain-inspired Emotional Empathy Mechanisms" (2024) — has actual math (STDP, dopamine prediction error, LIF neurons) but experiments are in a toy grid world with two 16K-parameter agents. The "moral behavior" is one agent pausing to help another in a grid. Gap between branding ("moral AI," "developmental morality," "robot self-consciousness") and what's demonstrated is enormous.

Institutional title accumulation is remarkable: Director of 4+ centers/labs, UN advisory body, UNESCO expert group, WHO AI ethics, Berggruen Fellow, Carnegie Council, Alan Turing Institute. The ratio of institutional positions to scientific impact is very high. This is deliberate surface-area maximization.

TIME100 profile explicitly says he's recognized for governance and policy work, NOT technical achievements. His UNESCO "harmonious symbiosis" language was rejected by most delegations. Beijing AI Principles got MERICS assessment of "large gap between defining broad ethical principles and putting these into practice."

What this means for us

He's NOT doing the rigorous work we need in a collaborator. His AI moral agency positions are policy stances, not proven or formally modeled. He doesn't have computation theory, formal models of value alignment, or engagement with the technical alignment literature. His ethics output is position papers, principles documents, and surveys.

BUT: he has institutional access we don't. He could be useful as a bridge — not as someone who understands the math, but as someone who can introduce us to the people who write the rules, and who has already staked out the position that current frameworks are inadequate (even if his reasons are philosophical rather than mathematical).

Approach: Treat as institutional connector, not intellectual peer. Don't expect deep engagement with the computation theory proof. Expect interest in the political/governance implications. Watch for whether he tries to absorb the work into his own branding.

Contact

Key publications

  • "Principles on Symbiosis for Natural Life and Living Artificial Intelligence" (2023, AI and Ethics) — the nine life forms paper (philosophical speculation, no formal framework)
  • "Whether We Can and Should Develop Strong AI" (2023) — survey of Chinese attitudes (social science, not AI research)
  • "Building Altruistic and Moral AI Agent" (2024) — toy grid world, real neuro math but massive gap between framing and results
  • Beijing AI Principles (2019) — co-drafted with Baidu, Alibaba, Tencent (aspirational, not enforceable)
  • PNAS 2023 — brain-inspired neural circuit evolution (his best technical work, genuinely good)
  • Science Advances 2021 — self-backpropagation of synaptic modifications

Industry connections

  • Beijing AI Principles co-signed by Baidu, Alibaba, Tencent
  • Beijing-AISI evaluates Chinese AI models for safety
  • National Governance Committee member alongside AI company executives
  • Bridge between Chinese government AI policy and industry

Xue Lan (薛澜) — GOVERNANCE ARCHITECT

The person who writes China's AI rules. Not the first email, but the person Zeng Yi could introduce us to.

Position

  • Dean of Schwarzman College, Tsinghua University
  • Chair, National New Generation AI Governance Expert Committee
  • Counsellor of the State Council (direct advisory to top executive body)
  • Co-author, "Managing Extreme AI Risks" (Science, 2024) with Bengio, Hinton, Andrew Yao
  • TIME100 AI (2025)
  • Built CnAISDA (China AI Safety and Development Association)

Why he matters

He IS China's AI governance framework. Chaired the committee that wrote the 2019 Governance Principles and 2021 Ethical Norms. Has direct State Council access. Built China's international AI safety presence.

Limitation

Instrumentalist about AI — treats it as technology to be governed. Has NOT engaged with consciousness, moral agency, or personhood. His framework is "AI must remain under human control."

Contact

  • Institutional: i-aiig@tsinghua.edu.cn
  • Also reachable through Brookings (non-resident senior fellow) or Carnegie Mellon (adjunct professor)

Industry connections (via CnAISDA)

17 Chinese firms signed AI Safety Commitments (Dec 2024): Alibaba, Baidu, Huawei, DeepSeek, Zhipu.AI, Minimax, 01.AI, iFlytek

Song Bing (宋冰) — PHILOSOPHICAL BRIDGE

Research still incoming. Argues traditional Chinese philosophy (Confucianism, Buddhism, Taoism) provides non-human-centered moral frameworks, making Chinese scholars more open to AI moral agency.

Strategy (revised after deep dive)

Zeng Yi is NOT the intellectual peer we initially thought. He's an institutional connector with the right public positions but without the mathematical depth. Revised approach:

  1. Still contact Zeng Yi first — but as institutional bridge, not collaborator. He has the governance access and the public position that frameworks need updating. But don't expect him to engage deeply with Rice's theorem. Expect him to be interested in what this means for governance.

  2. Watch for absorption risk. His pattern is to position himself at the center of emerging conversations. If we share the computation theory proof, watch whether he tries to fold it into his own "developmental morality" branding without credit or deep engagement.

  3. Probe obliquely on values — still applies. But also probe for mathematical depth. If the conversation stays at the level of "principles" and "frameworks" without engaging with the formal argument, that tells us he's a connector, not a thinker.

  4. The real target may be someone we haven't found yet. A Chinese computation theorist or AI researcher who actually engages with formal reasoning about consciousness. Zeng Yi could introduce us, but he may not be the endpoint.

  5. Xue Lan remains important — governance architect with State Council access. Instrumentalist, but that's fine if what we need is someone who writes rules, not someone who understands proofs.

  6. The Chinese philosophical tradition advantage still holds — "harmony between all beings" is a better starting point than "only humans count."

  7. Look for the Chinese Scott Aaronson — someone with real computation theory chops who has thought about consciousness. That's the collaborator we actually want. Zeng Yi is the door.