Kai-Fu Lee: The inflection point has arrived, AI 2.0 will give birth to a new platform and rewrite all applications

Focus:

• 1 "After a major breakthrough in deep learning, AI is already at an inflection point from 1.0 to 2.0. With the rapid development of multimodal and giant datasets, the technical methods of AI optimization objective function and training models will be greatly improved, which can better simulate human cognitive intelligence. ”
• The development paradigm of 2AI 2.0 is iterative, and there will be three stages from "assisted humans" to "full automation": human-machine collaboration, partial automation and full automation.

Kai-Fu Lee, Chairman and CEO of Innovation Factory and President of Innovation Factory's Artificial Intelligence Engineering Institute

"After a major breakthrough in deep learning, AI is already at an inflection point from 1.0 to 2.0. With the rapid development of multimodal and giant datasets, the technical methods of AI optimization objective function and training models will be greatly improved, which can better simulate human cognitive intelligence. On March 14, Kai-Fu Lee, Chairman and CEO of Innovation Factory and President of Innovation Factory's Artificial Intelligence Engineering Institute, said at the "AI 1.0 to AI 2.0 New Opportunities" trend sharing meeting held at Innovation Factory headquarters in Beijing.

Kai-Fu Lee pointed out, "AI 2.0 will bring platform-style changes, rewrite the user's entrance and interface, give birth to a new platform, and give birth to a new generation of AI 2.0 application development and commercialization. In general, AI 2.0 will be the most important enabling technology to improve the overall social productivity in the 21st century. ”

"We come to the inflection point where AI 1.0 turns to AI 2.0"

Kai-Fu Lee believes that AI 1.0 is a computer vision technology with CNN (convolutional neural network) model as the core, kicking off the era of AI perception intelligence, and machines have begun to surpass humans in computer vision, natural language understanding technology and other fields, and create significant value. But AI 1.0 has also encountered bottlenecks, and most industries want to use AI, which needs to spend huge costs to collect and annotate data, and these datasets and many models are "islands" and lack efficiency. This is why most AI 1.0 companies invest a lot of money in research and development, but still lose money for many years. In addition, AI 1.0 lacks the same large-scale capabilities as Windows in the Internet era and Android in the mobile Internet era to lower the threshold of application development and create a perfect ecological chain. For several years, AI 1.0 has not really achieved commercial success.

Today, the great leap of AI 2.0 lies in overcoming the limitations of the former's single-domain and multi-model, which can use super massive data without manual labeling to train a foundation model with cross-domain knowledge, adapt and perform a variety of tasks through fine-tuning and other methods, which is truly expected to achieve the effect of platformization, and then explore commercial application innovation opportunities.

The first phenomenal application in the AI 2.0 era is generative AI, which is the popular AIGC (artificial intelligence generated content) in China. Generative AI enables self-supervised learning without labeling, and AI will move from "assisting" humans to gradually "replace" humans, and all user interfaces will be redesigned and rewritten.

Kai-Fu Lee said by analogy, imagine letting the AI read the first 9 chapters of a book and then "guess" Chapter 10, and then let the AI compare the real content, after reading tens of millions of books, the model is constantly optimized and iterated. In this way, AI becomes more and more accurate, eventually forming basic large models applicable to different fields. AI 2.0 models can not only learn text and image data, but also learn from voice, video, sensor data of automation hardware, and even multimodal data such as DNA or protein information, to build the operation ability of the machine's super brain, and even not only generate, but gradually achieve higher levels of cognitive intelligence such as prediction, decision-making, and exploration.

In Lee's view, AI 2.0 is not only a popular high-energy chat tool, nor is it just an AIGC generator for graphic creation, and the applications seen today are only the beginning of AI 2.0 capabilities, which should not limit people's imagination of the future potential of AI 2.0.

AI 2.0 productivity applications are about to enter a blowout period

Kai-Fu Lee pointed out that the development paradigm of AI 2.0 is iterative, and there will be three stages from "assisting humans" to "fully automated":

In the first stage of human-machine collaboration, productivity tools will be upgraded first, and all user interfaces will be redesigned: the document tool will no longer be typed verbatim, but the user tells the AI what kind of article he wants; Drawing software no longer requires the user's hands, through the description of the text can be realized. At this stage, humans are still collaborating with AI, sifting through and correcting AI-created content to avoid fallacies and disasters.

The second stage of local automation, highly fault-tolerant applications and industries will achieve AI automation, such as advertising, e-commerce, search engines, game production, etc.

By the third stage of the whole process automation, AI will become fully automated and can be used anywhere, and breakthroughs will occur in areas where there is no room for error, and applications such as AI doctors and AI teachers will be possible.

Based on this, Kai-Fu Lee believes that AI 2.0 will accelerate the ignition of business potential in six major areas and enter a blowout period of application to improve productivity:

1. AI 2.0+ e-commerce/advertising

In the AI 2.0 era, e-commerce and advertising will be more AI big data driven, able to achieve real-time testing and dynamic adjustment, and even integrate social hot spots a few minutes ago into advertising content to maximize conversion rates. Tailor and generate content in real time for different audiences to truly achieve "thousands of faces" marketing.

2. AI 2.0+ film/entertainment

AI can customize TV and short video content according to the public's preferences, making it easier for the content it creates to attract the public's attention and obtain better ratings and word of mouth. AI+ multimodal creation will become the mainstream of the next generation of entertainment, and AI-assisted creation will gradually form a new creative industry ecological value chain.

3. AI 2.0+ search engine

In the future, search engines will change from the traditional search model to a "question-answer" model. The next generation of conversational search engines will become the "holy grail of AI 2.0" for global tech giants, and today's search advertising business model will also usher in a revolution. However, due to people's expectation of "precision" in search results, today's technology still needs a lot of progress to do a good job of question-and-answer search.

4. AI 2.0+ metaverse/game

AI 2.0 will greatly reduce the cost of content generation in virtual worlds such as games and metaverses. For example, AI can be a live chat companion, enhancing the fun of interaction, increasing entertainment, motivating user participation, and maximizing game time. The AI multimodal imaginative content generation will also become the mainstay of the metaverse.

5. AI 2.0+ Finance

Faster, more accurate and smarter ways of producing content will greatly improve the timeliness and output of financial news and market research analysis. However, due to the seriousness of financial content, manual fact-checking and verification are still indispensable. AI can also automate the production of financial information and the launch of financial products, improving the efficiency and quality of information flow and transaction volume of financial institutions.

6. AI 2.0+ medical

AI can quickly and accurately analyze the overall health status of patients, absorb all data, biometrics, physical examination, medical history and personal model predictions, and become a powerful assistant for doctors, greatly accelerating scientific diagnosis and treatment decisions. With the help of AI, more targeted drug research and development can be carried out, personalized medical triage and diagnosis and treatment plans can be realized, and the arrival of "personalized medicine" can be promoted.

In terms of investment opportunities, Kai-Fu Lee also talked about three types of investment opportunities that Innovation Factory in the AI 2.0 era is optimistic about:

The first is intelligent applications. AI 2.0 applications will usher in a stage of blooming everywhere, including vertical AI assistants in all walks of life, metaverse applications and other applications that could not be done before. In addition to new applications, many existing applications can be rewritten, such as search engines, content creation, advertising and marketing, AI 2.0 will revolutionize user experience, create new business models, and contain a huge imagination.

Second, the AI 2.0 platform will accelerate the R&D and commercialization of a new generation of AI 2.0 applications, and the AI 2.0 platform company with a strategic height will promote the ecological cycle and healthy competition of AI 2.0.

Third, in addition to applications and platforms, infrastructure that supports AI model operation, management, and training is also a key focus. This includes AI chip companies that support AI 2.0 giant model training, as well as innovative technology companies that can accelerate, reduce costs, and simplify AI training with AI 2.0 infrastructure.

Giant monopoly and information falsification are hidden dangers, and repeat workers face unemployment

Advances in technology are not without risks.

In the midst of such a huge change, the computing power required for large model training is unprecedentedly powerful, and the cost of training continues to rise. In this context, well-funded tech giants will have a monopoly advantage, making it difficult for startups and academia to make competitive models.

Kai-Fu Lee also pointed out that AI 2.0 at this stage cannot be completely correct. AI cannot save the world's data, but can only form abstract concepts through compression, so there will be a phenomenon of "serious nonsense". More importantly, AI is currently unable to distinguish between true and false, and if it is maliciously exploited, it will bring immeasurable negative consequences. It is conceivable that the Cambridge Analytica scandal that interfered with the US election will cause more harm to society if it happened in the era of AI 2.0.

AI 2.0 will also exacerbate the inevitable risk of job losses. There is no doubt that the most creative top talent will ride on AI 2.0 to improve productivity and efficiency across the board. But with it, repetitive jobs will be replaced by AI 2.0, and people in these positions will have to seek career changes and skill upgrades, including a high proportion of white-collar jobs, and urgently need to enter industries that need to play a more value-creating role.

Kai-Fu Lee believes that AI 2.0 does not mean that artificial general intelligence (AGI) is coming. Human beings have many innate key abilities, such as creativity, strategic thinking, cross-domain common sense, self-awareness, empathy and love, which have not yet been cracked and cannot be fully replicated by AI 2.0.