What Can We Learn from Professor Zhang Bo's Closing Remarks at the AGI Next Summit?

"The mission of entrepreneurs in the AI era is not simply to provide products and services, but to transform knowledge, principles, and applications into reusable tools, handed over to humanity like water and electricity. This is a brand new, sacred mission."

On January 10, 2026, the 'AGI-Next Frontier Summit' initiated by the Beijing Key Laboratory of Foundation Models at Tsinghua University was held at the Zhongguancun International Innovation Center.

Professor Zhang Bo of Tsinghua University's Computer Science Department and academician of the Chinese Academy of Sciences delivered the closing keynote. The 91-year-old academician's presentation carried a special 'on-site warmth' — during breaks, he 'urgently adjusted' his content based on the sharing by earlier speakers including Yang Qiang, Tang Jie, Yang Zhilin, Yao Shunyu, and Lin Junyang.

This article is based on academician Zhang Bo's speech, aimed at providing a reference and guide for understanding the future development of artificial intelligence, its deep-rooted challenges, and the mission of our era.

What Are We Doing?

"If a machine can generate semantically coherent human-like language in open domains under external prompts, does that count as mastering human language? One could say yes, but not deeply enough."

Academician Zhang Bo believes that the large language models being developed today were originally built as chatbots — machines can generate diverse, semantically coherent, human-like language in open domains under external prompts, but achieving this isn't thorough enough.

He explained that the core mechanism of current LLMs lies in distributed semantics — 'defining semantics as Firth described: the meaning of a word is determined by the words that co-occur with it most frequently.'

This approach converts traditionally co-occurring words in discrete space into sparse space in high-dimensional representation. He believed: 'This is a significant breakthrough — it makes language computable.' In theory, when training data is sufficiently large and context length sufficiently long, this vector space can approximate real semantic relationships, enabling machines to achieve a certain degree of 'understanding' and even 'reflexive' thinking.

However, such models are essentially approximate models of human language, not true models of human language, because semantic definitions themselves are incomplete and inaccurate — accurate definitions simply cannot be found in science. This inevitably affects applications using language models, creating five 'deficits': deficits in reference, truth and causality, pragmatics, polysemy and dynamic context, and closed-loop behavior.

What Do We Need to Do?

Academician Zhang Bo further pointed out that an important goal in the current AI field is to achieve the evolution from large LLMs to Agents capable of executing complex tasks in real-world environments.

He emphasized that everyone uses 'Artificial General Intelligence (AGI)' to frame this goal, but in fact there are misunderstandings about the concept of AGI. Although AGI emphasizes generality, many current definitions of AGI, such as 'machines can complete over 70% of human tasks and surpass human levels,' are unexecutable and unverifiable.

Therefore, a workable, verifiable definition of AGI is crucial. Accordingly, AGI should satisfy five key capabilities: spatiotemporal-consistent multi-modal understanding and grounding, controllable online learning and adaptation, verifiable reasoning and long-term planning and execution, calibratable reflection and meta-cognition, and cross-task generalization.

If we take these five points as the goal for AGI, we have a workable, verifiable definition that can guide our future research directions. Around these five goals, six things are currently being done:

• Multi-modal grounding
• Embodied and interactive grounding
• Retrieval and evidence grounding
• Structured knowledge alignment
• Tool and execution grounding
• Alignment and constraint grounding

What Is the Goal?

Academician Zhang Bo pointed out that our current attitude toward AI is in a 'very contradictory state.'

On one hand, we expect AI to take on more and more complex work; on the other hand, we 'fear AI surpassing us to become a new kind of agent,' which makes everyone feel anxious. The deep-rooted cause of this anxiety is that in the past, humans were the only 'agents,' but everyone's wishes and demands are different. Then, 'if another type of agent beyond humans appears in the future, what should we do? How do we coexist with AI? How do we solve these concerns?'

To respond to these core questions, academician Zhang Bo proposed a framework: 'The future's agents are divided into three levels':

• Functional and action agents: This is the level AI has already reached and is widely expected to fulfill — serving as tools that assist humans in completing tasks.
• Normative and responsibility agents: An unrealized level where making machines assume responsibility is a current technical difficulty and future goal.
• Experiential and conscious agents: The most concerning level — what happens to humanity if machines gain consciousness.

Academician Zhang Bo believes that for companies doing practical work, it may not be necessary to consider this far, but the issue of 'alignment and governance' must be considered. He believes the primary objects of governance are not the machines themselves, but humans — especially researchers and users. This involves what responsibilities AI-era enterprises and entrepreneurs should bear.

What Is the Mission of Entrepreneurs in the AI Era?

Academician Zhang Bo redefined the mission of entrepreneurs in the AI era.

He admitted that before large language models appeared, he did not encourage students to start businesses, because traditional entrepreneurs mostly aimed to 'make money.' But with the emergence of large model technology, he believes the best students should devote themselves to entrepreneurship, because AI is redefining what it means to be an entrepreneur.

Academician Zhang Bo believes future entrepreneurs should assume responsibilities in six areas: redefining value creation, organizing and releasing new productive forces, establishing credible and governable intelligent systems, building long-term resilience, promoting industrial collaboration and ecosystem win-win, and achieving inclusive and sustainable growth.

"AI should not merely provide products and services, but transform knowledge, principles, and applications into reusable tools to benefit humanity. AI should become a universal, inclusive technology like water and electricity, delivered to the whole of society."

Academician Zhang Bo emphasized that these new missions of the AI era will make entrepreneurs one of the glorious and sacred professions.