20250506Special Lecture by AI Research Group, The University of Tokyo

Yasuda Auditorium

They will talk about the future of society and the potential of AI technology in all directions.

Special Lecture by AI Research Group of Todai

https://gyazo.com/aef969fb3c1115f727fd972f2c9e7976

Apparently the Australian national broadcaster ABC is filming a documentary on Audrey.

https://pbs.twimg.com/media/GqPWUEcbAAAY0fB?format=jpg&name=small#.png

https://pbs.twimg.com/media/GqPjqmzbAAIJWav?format=jpg&name=small#.png

AI Support Panel Discussion

https://pbs.twimg.com/media/GqPrMr1b0AAzkOx?format=jpg&name=900x900#.png

Yoichi Ochiai: "In that case, humans would become the bottleneck and we would lose in speed to something that does not involve humans."

Au "Humans should steer in the preferred direction instead of braking and slowing down."

This is a very interesting question.

Audrey "If there are no exits on the highway, you have to keep driving in the same direction for hours after you realize you've gone in the wrong direction. The government should use its top-down power to create lots of exits and maintain the ability to change course."

This was interesting too.

Words from Mr. Ochiai

"(In a democracy) we all want to eat good food. We want to relax in Hawaii. And we're going to vote for someone who can make that happen? It's hard to verbalize what people want," I understood him to have said.

I enjoyed the humorous phrases in places.

Looking forward to the AI workshop in the future!

https://pbs.twimg.com/media/GqPzVJEa0AAmA2a?format=jpg&name=small#.png

https://gyazo.com/a5337b8af41df33bda3e07c84c29d287

I realized after the meeting that I had written it before the meeting.

o3.icon

Audrey Tang Keynote Speech 1)

Audrey Tang Keynote Speech (2)

Yoichi Ochiai Keynote Speech

Discussion and Q&A First half

Desired Direction for "Accelerating" Democracy with AI

Gap between shared instincts of hunter-gatherer societies and modern democracy

Need for "society to have a steering wheel" (Audrey)

Dialogue/Questioning Midfield

Regulatory design on two axes: offensive AI/defensive AI

Highway and off-ramp metaphor = lock-in workaround (e.g., number portability)

Discussion and Q&A Second half

Importance of AI Transparency and Explainable AI

Telescope metaphor: remove tool "aberrations" first!

The problem of "collective knowledge becoming a bottleneck" in the next 5 years

Digital Nature (what will become) and Democracy (what should become)

Mr. Ochiai's question to Audrey was that while "digital nature" is something that "will become," "democracy" needs to be oriented as something that "should become. The question was posed as to how to think about this "what should become.

Audrey said that digital democracy is the direction in which it "should" be headed, using the power of digital to efficiently gather people's opinions and enable more direct democracy, rather than the relative inefficiency of a traditional representative system. He said that the goal is to collect everyone's opinions to the extent possible now, while aggregating diverse individual opinions and, aside from future technologies such as brain wave technology, to better shape social dialogue.

In response, Mr. Ochiai commented that in hunter-gatherer societies, sharing food and other resources was the essence of human nature, a behavior that was engraved in our DNA. However, since agriculture, he said, democracy, human rights, and ethics have become things that must be learned to be gained, and the challenge is how to align them in the future.

He stated that whereas the abstraction of primitive society was the group and others, digital technology has the potential to connect with many more than 150 people and liberate more of our humanity He believed that AI should contribute to the growth of human society by assisting such collective intelligence rather than amplifying individual intelligence He also expressed his belief that AI should contribute to the growth of human society by assisting such collective intelligence rather than amplifying individual intelligence.

The next question was how to position human values and roles in an increasingly AI-oriented society, and how to maintain social trust and cooperation. The options presented were to "rebuild society around the inherent value of humans" or to "form a new society by collaborating with AI and clarifying the division of roles.

o3.icon

Mr. Ochiai's point was that "the period of civilization using advanced intelligence" during the 250,000 years of human history was a very short "extra" period, not that "human intelligence itself is insignificant. He did not say that "human intelligence itself is insignificant.

The nuance was that "AI should be turned to assist humans in physical and emotional tasks, thereby extending the activities that people are inherently good at." It is closer to "helping people to continue without burden" than to "making it easier."

Mr. Ochiai then warned that if social systems are "optimized for what AI is good at (computation and logic)," human physical and emotional needs will be left behind, and summarized that "what humans are naturally good at is what AI should support and make sustainable.

Mr. Audley emphasized that it is most important that AI continue to work with society as a whole, as rapid evolution of AI alone could have two consequences: concentration of power, or social fragmentation due to everyone having a strong AI. He stated that AI is a powerful engine, but the "steering wheel" should always be in the hands of society, and that it is desirable for people to understand the power of AI through education, and for AI to amplify collective intelligence, rather than individual intelligence.

Mr. Ochiai asked that collective intelligence is great, but what happens when AI continues to develop itself and as a result, the collective intelligence becomes a bottleneck and cannot coordinate (align) the AI's actions?

Audrey stated that AI and humans should work cooperatively and that AI may even be able to communicate directly with nature and animals. However, he also expressed his belief that humans can maintain a position of control over AI, since its development is also prompted by the demands of human society; it is important to have a sense of being at the "wheel" of driving rather than fearing AI, and by incorporating norms such as an ethic of care into AI's behavior, he said He said that it would be possible to make AI behave ethically without monitoring.

Direction and speed of AI implementation

The discussion moved on to a question about the social changes resulting from the introduction of AI, particularly with regard to the increase or decrease in employment, and how to put it in the right direction. The options presented were to "accelerate adoption as quickly as possible" or to "adjust the speed of development so that everyone can use it.

He expressed his belief that the AI should be implemented immediately because it learns very quickly and the AI may finish learning while we are discussing regulations.

Mr. Audley distinguished between defensive technologies that enhance society's resilience (e.g., manufacturing advanced masks at home) and offensive technologies that can cause destruction (e.g., manufacturing viruses at home), with defensive technologies needing to be accelerated and offensive technologies needing to be regulated. He stated that which should be prioritized should be determined by whether the technology contributes to the resilience and consensus building of society as a whole.

Questions were also raised about how humans should deal with the evolution of AI, what regulations are needed, and what exactly constitutes "bad AI."

Mr. Ochiai immediately responded that the regulations would not be in place in time. He then likened it to an AI car and an AI police car driving at the same speed on a highway.

Audrey stated that it is important to have handles to orient the AI, not slow it down.

He stated that "differentiable design" (differenciable design) is necessary for this purpose.

o3.icon

Three main scenes in which "differential..." appeared

“We don’t want downward acceleration. We want differential acceleration.”

Defined the need for acceleration to sort out the direction of AI acceleration, rather than simply stopping/advancing it.

Compare to a car: "Even a strong engine (AI) will go off a cliff without a steering wheel. Steering wheel = social institutions and ethics.

Embodiment of "two-track, differential acceleration"

The technology is divided into Defensive (to increase the resilience of society) and Offensive (which can cause extensive damage on its own),

Defensive accelerates forward as much as possible

Offensive is deliberately delayed until society is equipped

The proposed timeline is a differential dynamic timeline called

"Differential acceleration is the driving manual of computational democracy."

(1) Society takes the wheel UI/Institution

(2) Advance acceleration of defense technology

(3) Aberration check at all times with AI transparency

The three sets of "differential acceleration" are summarized as "differential acceleration".

Mr. Ochiai brought up the concept of "derivative" (rate of change) and explained the importance of understanding the direction (derivative) of the change, rather than just looking at the change. He warned that if we do not look at the direction of change, we will think that AI can do anything, and we will not be able to recognize when we are heading in a dangerous direction.

consensus building and leadership

Finally, the question was posed as to whether top-down or bottom-up leadership is more effective in promoting consensus building throughout society.

Mr. Ochiai noted that in agrarian and industrial societies, leadership was a way to bring people together and keep them moving forward in a predictable manner, but as society becomes multi-agent (i.e., thinking not only about humans but also about the environment) through AI, things will change instantaneously, requiring a different type of leadership. He also mentioned that a different type of leadership will be needed. He said that a leader who can provide direction through flat relationships and empathy will be required.

Audley said that an important role for the government is top-down regulation to create "bifurcation points" (onramp and offramp) where users can move freely between different networks and services as needed. Many Internet services are hard to get out of once you use them (lock-in effect), and it is often impossible to switch to another service without losing friends and content. This can be compared to a highway with no offramp. He expressed his belief that government intervention to develop a system that allows such free movement, like MNP (Mobile Number Portability) for cell phones, will promote competition among services and ultimately lead to a better society for users.

Mr. Ochiai stated that intelligence is an "extra" for humans, and that the act of verbalizing an individual's vague desires in a democracy, such as "what I want to be in the future" or "what I want to do," and selecting people connected to these desires, is a very difficult task for human intelligence. However, he said that this difficulty may be a surprisingly easy task for AI, and suggested a future in which AI will facilitate political decisions based on individual desires. He concluded by saying that he felt that the current user interface and social institutions are imperfect and that we need to use AI to capture something like the human collective unconscious.

Audrey said it is important that AI be designed to be transparent and explainable (explainable). Like early telescopes, technology comes with noise and bias, but by improving the technology, he cited the example of an AI system that last year had hallucinations (hallucination), but this year has been able to control them. He said that we should pay attention not only to the improvement of AI capabilities, but also to the improvement of its transparency and accountability, so that AI can one day be used as a "mirror" reflecting the collective knowledge of society as a whole.

---