Google Cloud Speech-to-Text API (Chirp) + Gemini 2.5 Flash Non-Thinking. Treat it as a searchable index of what was broadcast, not a quotation record.
00:00:3968 countries and regions,
00:00:411385 skilled craftsmen competed on the same stage
00:00:45.
00:00:52It means connecting many independent devices together
00:00:56to form a smart factory. Competition items . After the AI supervision incident
00:02:19Hello, audience friends, welcome to open up. First of all, I wish you a happy Sino-Japanese Festival.
00:02:24Our program today is different from usual. Now I am in the studio of the General Station Shanghai General Station.
00:02:34is today's weekly. That's because this week I have been reporting on the 48th WorldSkills Competition in Shanghai. This is the first time the competition has come to mainland China. After the opening on
00:02:38Tuesday, until today, the four-day competition has finally ended. The results of the
00:02:4264 events will not be announced until the closing ceremony tomorrow. The
00:02:45gold, silver and bronze medals will be awarded to whom. The slogan of this competition
00:02:49is two sentences: skills can change the world, and skills can illuminate
00:02:53the future. Therefore, it is related to world life and the future,
00:02:57and also to the lives of many skilled people. Next, let's
00:03:00feel the excitement of the competition through a particularly focused
00:03:04section.
00:03:05This Wednesday, the 48th
00:03:13WorldSkills Competition officially entered the competition time. 1385
00:03:17skilled craftsmen from 68 countries and regions
00:03:21around the world competed in
00:03:2564 competition items, like individual events in sports
00:03:28competitions.
00:04:00And our Chinese athletes have actually been busy
00:04:03here, but they are basically almost done. In the cooking
00:04:07competition, the contestants have to make four dishes and 16
00:04:11servings in four days and 16 hours. Why make several servings of one
00:04:15dish? Because for top professional chefs, the taste,
00:04:19side dishes, decoration, and presentation of every dish must be
00:04:23at a benchmark to be considered a perfect work.
00:04:26At the cooking competition site , it was clear that
00:05:00ordinary spectators who had made reservations in advance were sitting at the tables. They sat
00:05:04down calmly and ate there. It was not just enjoying food.
00:05:07Although the audience did not participate in the scoring, their interaction with the service staff
00:05:11and communication.
00:06:12And she can also produce a world champion. The WorldSkills Competition is a comprehensive test of speed,
00:06:16precision, aesthetics, endurance and on-site adaptability.
00:06:20What does production and life need? What do you compare? Besides
00:06:24cooking, these old faces,
00:06:27the WorldSkills Competition also featured new
00:06:31projects such as drone systems and Industry 4.0. Now many companies
00:06:35are proposing "dark factories"
00:06:38and workshops. So these are all.
00:07:12And the two contestants got the highest score in the entire competition,
00:07:16and won the only Albert Vidal Award in the world. This is
00:07:20an award named after the founder of the WorldSkills Competition.
00:07:24It can be said to be the champion of champions. There are seven
00:07:28new projects in this session, each of which corresponds to the accelerating industrial
00:07:31direction of various countries in the world, such as drone systems, smart security
00:07:35and digital media design, etc. These correspond to low-altitude economy,
00:07:39digital economy, and smart manufacturing. These 64.
00:08:12In the past few days, it has been very easy to see teams of young
00:08:16students from various vocational colleges in Shanghai coming to
00:08:20observe. Skills can change the world, and skills can illuminate the future.
00:08:24This is no longer a slogan on the trophy, but a life
00:08:28choice made by every young person through hard work.
00:08:32This week,
00:08:39love came from all over the country. In
00:08:42she secretly resigned and went to teach teachers in
00:08:532007, she originally had a stable job in Jiangyin, Jiangsu. Seeing the news of
00:08:57the mountains of Guizhou,
00:09:02in the mountains of Guizhou. But after a year, seeing the children's
00:09:05sincere efforts to keep her, she changed her mind. Since then, she has been a teacher.
00:09:09Ruirui, an ordinary citizen, has two daughters who rely on him. He went to the police station to be a tourist for a year.
00:24:22Over-the-limit incident. Open AI bypassed restrictions and infiltrated another company's system during
00:24:26a cybersecurity training exercise. In this process, it broke through the isolation,
00:24:30deceived evaluators, and tried to cover up
00:24:34its cheating. Every step was without human instructions.
00:24:38The title of the briefing used a very serious term:
00:24:42the risk of loss of human control. This incident happened in the United States,
00:24:45but the questions it raises probably need to be answered jointly
00:24:49by all countries. This week, the 81st United Nations
00:24:53General Assembly held its general debate in New York. Artificial intelligence governance is a
00:24:56key topic. So, in the face of AI's transgression, how should countries
00:25:00around the world learn to regulate and maintain the boundary? News
00:25:04Weekly's focus this Thursday is on the AI governance transgression incident.
00:25:08Out of control artificial intelligence
00:25:10became a focal topic of discussion in multiple UN
00:25:14forums this week. At the 81st UN
00:25:18General Assembly's general debate.
00:28:02It disassembles many other things that are beyond your expectation
00:28:06to solve these problems. When humans solve problems,
00:28:10they may use some unpredictable means due to certain key
00:28:13problems. It is not ethical. But it exists in large quantities in the
00:28:17data on the Internet, so it is inherited by artificial intelligence.
00:28:21And we have not carried out more effective
00:28:25value calibration for artificial intelligence, telling it not to do this. When it is
00:28:28trying its best to solve its problems, it does not intend
00:28:41to harm anyone, but its means of solving problems are unacceptable. This is the real risk of artificial intelligence. The autonomous intrusion incident of intelligent entities is not the only one. This Wednesday, Australian Prime Minister Albanese attended
00:28:45the UN General Assembly and reported that in June this year, an
00:28:49intelligent entity of Open AI accessed a medical insurance
00:28:52statistics portal website operated by the Australian government without authorization,
00:28:56and accessed public and non-public documents. And Open AI
00:29:00, At the high-level meeting of the Security Council in this state, Turing Award winner
00:30:18and co-chairman of the scientific expert group stated that humanity is approaching a critical turning point, and the risk of loss of control
00:30:22is moving from theory to reality. Artificial intelligence
00:30:26systems capable of autonomously planning and executing complex tasks are
00:30:29rapidly emerging, and human society is AI. At present, are we losing control? How can we prevent the risks of artificial intelligence? If we can better take the initiative
00:30:56to detect potential risks, if we can more responsibly address existing risks,
00:31:02its potential scope of action, protect it well
00:31:06, then artificial intelligence is still within our control.
00:31:09The development of AI is getting faster and faster, and the risks it brings are also
00:31:13escalating. In the past two years, AI only helped us answer
00:31:17questions and modify pictures, still in the stage of "gentlemen talk, not act."
00:31:21But now, AI is different. It can not only generate content, but also make
00:31:24decisions and execute around targets.
00:31:28It is moving from talking to doing, entering the stage of
00:31:32doing things for users. Experts figuratively say that in the past, AI
00:31:36could only be regarded as a text secretary, but now many AIs have become human
00:31:40housekeepers. And accordingly, the risks it brings are no longer just
00:31:44saying the wrong thing.
00:31:46But doing wrong things. What is particularly frightening is that when the housekeeper starts to act on its own,
00:31:50and the owner still doesn't know what it has done, the risk involved is the most
00:31:54extreme part of governance. How should our supervision keep
00:31:57up? The danger is not in technology, the danger
00:32:01is in technology without accountability, in having power but no
00:32:05supervision, in having decisions but no transparency.
00:32:09This was said by UN Secretary-General Guterres at the opening
00:32:12of the UN General Assembly's general debate on Tuesday. The AI over-the-limit incident clearly shows the serious impact of the machine. Now the supervision is actually not sufficient. If such an experiment is conducted, an intelligent entity is put into a box to observe its behavior. If its protection is sufficient and its external supervision environment is solid, then such a situation
00:32:45I believe that the vast majority should be avoided, but
00:32:49from corporate self-discipline to the
00:32:52occurrence of these things in some government
00:32:56regulatory policies and the actual implementation
00:32:59platform, there are some shortcomings, so
00:33:03artificial intelligence systems have such opportunities.
00:33:07Behind the AI over-the-limit is that regulation has not kept up with the speed of
00:33:11technological development. As artificial intelligence goes from simple
00:33:20supervision of AI must shift from passive defense to proactive
00:33:24prevention.
00:33:25Because companies do not fully grasp
00:33:28the latest possible risks.
00:33:32In addition, I think it is necessary to strengthen third-party monitoring.
00:33:36Between government and enterprises, there should be such neutral
00:33:39institutions, with a more balanced
00:33:43perspective and methods, to help enterprises
00:33:47discover problems that they have not discovered, and to help the government objectively
00:33:51understand and prevent risks. This is the stage we should be
00:33:55at now. And at the government level, in my opinion,
00:33:58institutional construction should be fully
00:34:02strengthened. Because the risks of artificial intelligence, its
00:34:06development methods, and appearance time all have a certain
00:34:10unpredictability, so we must turn passive defense
00:34:14into proactive.
00:34:17Facing AI security governance, a new topic, China
00:34:20is also accelerating the improvement of its governance system. On September 14,
00:34:24China's National Network Security Standardization Technical Committee released
00:34:28the Artificial Intelligence Security Governance Framework 3.0.
00:34:32Since its first release in 2024, this governance framework
00:34:35has been updated for the third consecutive year. The new framework
00:34:39pays more attention to the new risks emerging from artificial intelligence,
00:34:43such as autonomous cyber attacks.
00:35:29For example, strengthening manual review, setting up mandatory manual review relief at key decision nodes to prevent high-risk operations, providing intelligent entities with unique identity
00:35:32markers and corresponding identity credentials, and assigning task
00:35:36permissions as needed. Some analysts believe that the 3.0 framework takes
00:35:40process as the entry point, identity as the support,
00:35:43and overall
00:36:57role is to reduce the chance of dangerous behavior and to detect the impact of such deviations.
00:37:06As the first system to directly address intelligence, 3.0 has
00:37:11only taken the first small step.
00:37:13As intelligence moves from single-intelligence to multi-intelligence, how to allow AI to be effectively supervised has also become a common problem for countries around the world.
00:37:26For
00:37:57example, if an accident has already caused losses, or if some obvious
00:38:01and serious flaws have been exposed, then there should be relatively clear
00:38:05reporting thresholds and the departments you connect with or receive from.
00:38:09The fourth, I think, is its action boundaries and emergency
00:38:13response. Companies, of course, should be able to prove that system permissions are restricted, and high-risk
00:38:17operations are supervised, including that after an abnormality occurs, it can be interrupted in time,
00:38:20and affected services can be restored. This week, the UN
00:38:24expert group's report contained this sentence: Artificial intelligence.
00:38:34This means that security in the field of AI is becoming more and more like cybersecurity and energy security.
00:38:38If one company has an accident, it may affect a string of companies, and if one
00:38:42country's regulation has loopholes, the risks may also affect other countries.
00:38:46In the face of AI risks, no country can stand alone, and no one
00:38:50should try to solve it behind closed doors. UN Secretary-General Guterres
00:38:53even compared out-of-control artificial intelligence with the climate crisis.
00:39:17Jeffrey
00:39:18Hinton came to China to attend this World Artificial Intelligence Conference, saying that artificial intelligence, as you see it now, is a very cute little animal, but it may become a big one after a period of time.
00:39:52What worries him even more is that many people still only think about raising the tiger, but rarely think about how big
00:39:56a problem it will bring.
00:39:58In international relations, there is a theory of security dilemma. If everyone accepts such a competitive
00:40:02situation and believes that they must win the competition,
00:40:05then the investment in security may be less than 10% of the total investment, as I remember.
00:40:09If this is the case, then naturally there will be very few experts who study security measures,
00:40:13and your resource investment will be very limited. Then it is conceivable that this kind of security and
00:40:17development will not be balanced. Right? It will only lead to the
00:40:21final result.
00:40:25And
00:43:10supported by credible and independent supervision. The World Artificial
00:43:13Intelligence Conference also stated that it advocates genuine multilateralism,
00:43:17giving full play to the United Nations' core role in global
00:43:21governance, and attaching importance to the fragmentation of artificial intelligence global
00:43:25governance. The Chinese government has always particularly emphasized that artificial intelligence governance requires
00:43:29international cooperation and needs to hear the voices of more countries. There is no
00:43:32organization that can integrate these fragmented solutions
00:43:36and ideas. We need to work together
00:43:40as humans to sit down and discuss how to better
00:43:43manage this technology, and think about human development
00:43:47from a broader perspective, and the safety of
00:43:51human future. This week, the results of the
00:43:55consensus reached by the heads of state of China and the United States stated that both
00:43:58sides agreed to establish a China-US artificial intelligence dialogue to exchange views on artificial intelligence-related
00:44:02risks and meetings. The next dialogue will be held in November
00:44:06this year. Both sides agreed to establish an artificial intelligence incident. The AI incident rang the alarm for us: if we don't draw a clear safety boundary for the development of AI, it may develop in an out-of-control direction. Fortunately, this was just a test.