Lei Tao, CEO of Techchuang, Tianyun Data: Building a Cognitive System for Correctly Understanding Digital Intelligence

Lei Tao, CEO of Tianyun Data
Won the highest national AI award: Wu Wenjun Artificial Intelligence Science and Technology Invention Award.
Winner of high-end leading talents in Zhongguancun in 2020; the first batch of CCF China Computer Society big data committee members; leading and participating in the planning and construction of multiple billion-level large-scale informatization projects such as HSBC, China Mobile, China Unicom, and Industrial and Commercial Bank of China.

—
Text | babayage
Edit | smile

After more than two years of research, "Science and Technology" has gradually formed a set of value judgment standards: sharing values> sharing methods, sharing low-level cognition> sharing experience, sharing problem-solving principles> sharing single-question answers...choose growth experience as the narrative carrier This is because the key choices and decisions in life best reflect their cognition, principles, and values.

Since value is the purpose and narrative is the means, it should not reduce the efficiency of value transmission for greedy means. Therefore, "Technologists" decided to make an exception for some of the predecessors who are good at systematically sharing cognition, principles and underlying logic. The content of the sharing is abandoned and the source code is submitted.

The first one is Lei Tao, CEO of Tianyun Data.

Paradigm and empiricism fail
Destructive power comes from data native

Technologists: More and more people realize that this is an era when traditional paradigms and empiricism fail, and it is an era when the correct answer is re-searched. Where do you think the power to impact the existing methodological system comes from?

Lei Tao: The macro characteristics of each era must not be caused by a single factor, but there must be some factors that can shape it. In my opinion, the part of the knowledge system based on transcendentalism rather than scientific logic is being replaced and disintegrated by a new knowledge production system based on data.

At the beginning of this year (2021), Google Cloud Artificial Intelligence Application Artificial Intelligence Engineer Dell Markowitz invested in an interesting study: let artificial intelligence learn the scientific reasons behind the crispness of the cookies and the softness of the cake, thereby completing an AI baking menu. Can you decide whether to make biscuits or cakes from the baking process? The result of machine learning is that the baking process can no longer be clearly defined, and can only be defined from the ingredients of the raw materials.

The machine learning process of biscuits and cakes, as well as countless other similar cases, tell us a fact: the knowledge in the digital world is built on the input (components) and results, and the white-box process reasoning that we humans are accustomed to. Knowing (recipe) is very different. We must realize that in the digital world, many human knowledge judgments are often invalid. For example, how do we identify the Turing test method of machine intelligence? The Google conference has falsified the Turing test to identify machines and people.

Data native will bring about a new revolution in knowledge production. The white-box process reasoning cognition will be replaced by the black-box digital long-range calculation, which is more accurate and stronger.

y=f(x), data native promotes the knowledge production revolution

Technologist: Most voices still refer to the changes brought about by digitalization as transformation, but you use the word revolution decisively. How do you define a knowledge revolution?

Lei Tao: We conclude from the four development stages experienced by the production of knowledge:

1 Scientific experiment: the ancient fire drill to Galileo's Leaning Tower of Pisa, knowledge is produced from practice;

2 Theoretical reasoning: Newton uses the mathematical tools of calculus to derive, and knowledge is produced from axiom formulas;

3 Simulation calculation: Based on the known simulation modeling of the physical world, knowledge is produced from scale calculation;

4 Data native: To solve the uncertain process of answering, knowledge is produced from massive data association;

The digital economy is undergoing a stage of development from "data twins" to "data native". The former applies human knowledge to the digital virtual world and is still in the third stage, but the latter produces new cognitions adapted to the digital economy. The existence of levels.

The stage in which data twins promote knowledge production is to try to use existing cognition and knowledge structure to solve problems in the virtual digital world, use our knowledge white box to build a model, do high-performance computing to reason, and calculate more knowledge. Relying on existing knowledge automation based on axiom cognition, the computing power infrastructure is the HPC high-performance computing cluster of the supercomputing center. Faced with complex dynamic and personalized computing scenarios, the simulation of white box modeling encounters ceilings. For example, weather forecasting based on classical geophysical theory modeling fails in the solution of sudden extreme weather such as local weather and thunderstorms.

How digital native produces new knowledge beyond human cognition. Just like Alpha Go, it does not learn from human best practices and excellent chess records, or from existing knowledge. Instead, it learns the process of intermediate uncertainty from behavioral data (black and white moves) for results (wins or loses), and produces Generate new knowledge and reconstruct new business processes and practices. For example, the e-commerce recommendation algorithm reconstructs the retail business, and the planning algorithm of the taxi-hailing software reconstructs the business organization form of supply and demand.

Digital native is reconstructing human cognition.

Technologist: What is the transmission mechanism of data-native knowledge production, from producing knowledge to changing the physical world?

Lei Tao: Take an example of AI business applications. The business model of modern enterprises is undergoing business restructuring from process-driven to data-driven transformation. Artificial intelligence can replace traditional experience, rules, and processes, reconstruct business practices, and promote new models. Business decisions.

Here is a functional formula to express the Schumpeter growth model in the DT (Data Technology) era, that is, the core value is embodied as y=f(x), y is the result, x is the data, and f can be approximately understood as a certain law—— However, it must be emphasized that the understanding of laws by numbers is different from that of human beings. People are good at abstracting and summarizing simple laws, while numbers use complex understanding of complex. Taking the financial sector as an example, when we input a large amount of consumer behavior data (x) and the result data y of capital transactions, through the processing of the database and the AI PaaS platform, we obtain the anti-fraud risk assessment model f, and f can become Smart applications for rapid expansion and replication of 1 by 100 do not need to move data around. As long as f is invested in various "anti-fraud application scenarios", value can be created, and f can be used as a new production factor, thus in the transformation of the information industry. Obtain high growth.

New production methods and changes in production materials have brought about iterative improvements in efficiency. The role of the machine has changed from rigorously executing human instruction programs to iterative learning based on objectives, expressing the uncertain process in the input and output process as a Software models or smart applications will greatly improve the production efficiency of software. For the information industry, this is also a disruptive change in itself. The income of technology companies in the DT era can be reflected in the composite nature of platform tools + data science services. income.

The future of human and machine symbiosis

Technologist: When machines begin to produce knowledge, should the relationship between humans and machines undergo a disruptive change? In your opinion, what is the relationship between humans and machines in the future?

Lei Tao: In the past, in the face of a large amount of information, human beings have always believed that they are the spirit of all things, "you give me the information, I will control and then make judgments." In the process, countless masters and experts were born. So what do experience and experts give us? It is a series of reports. You can see a series of content such as sales figures and sales volume for this month, and then make decisions based on these figures.

But after the emergence of AI, the highest value of people is no longer to process information, but to nurture AI. Before we design a sophisticated algorithm engine, we throw it on the production line, and then plan the design of the engine itself. Like the Roman Arena three thousand years ago, let two deep learning monsters pit against each other to get the best result.

Under this model, the roles of humans and machines have been redefined, and the intervention of AI has improved our roles. We are no longer a simple participant, no longer repeated workers on the production line, but more To engage in some high-precision work.

The digital world ≠ the physical world
Break the cognitive bottleneck and be wary of the "minke" of digital intelligence

Technologist: You have mentioned many times that the way of generating knowledge from data is in a black box state, which cannot be accurately observed and understood by people, so we can't simply understand the "f" of AI calculation as a law?

Lei Tao: Early AI also tried to find some rules. For example, the scoring system that we commonly use in credit cards, is it three thousand yuan or thirty thousand yuan? But we are no longer relying on simply expressing things, but on complexity, and on digital expressions.

AI restores our understanding of the complexity of the entire world. When humans see a tree, they are more accustomed to abstract thinking, no matter what color it is, how many branches, etc., our first reaction: This is one A tree; but when the machine sees this tree, it will try to capture all its details. This is the strength of the machine, and it is easier to express complexity.

It must be admitted that there are many problems in this world. Humans cannot abstract simple rules. For example, we use a lot of visual computing now. How can we make pictures recognize that this is a cat or a dog? Using human language and thinking to describe picture information is very limited, and these limited elements cannot restore complex content; similarly, how do we use Alpha Go to abstractly describe the overall view and style of the 160,000 chess players?

Human language is incapable of "reducing complexity", and deep learning undoubtedly gives us a way to describe the complex world, using a complex mathematical system and distributed computing capabilities to deal with it, and deep learning also provides We have found a way to recognize maps and stitch maps.

Using complexity to deal with complexity, mankind has a new method to gain a wider range of cognition.

Technologist: So will the "black box that cannot understand the digital world" affect people's acceptance of this method, and thus affect the popularization of this productivity?

Lei Tao: I do have this concern. In the face of the knowledge production revolution of digital native, the biggest constraint is that many people are accustomed to using the physical world to understand and define everything. In my opinion, this recognition will hinder the popularization and development of digital native.

Every knowledge revolution is accompanied by a break in "cognition." People cannot fly with wings. What really makes airplanes fly into the sky is aerodynamics; when cars appear, people just need a faster one. Horse, the term horsepower continues to this day; when the ancients kneaded earthenware to make pottery, they would have never imagined that today's lithography machine burns integrated circuits on monocrystalline silicon wafers...

The new knowledge revolution will inevitably bring about a new cognitive system. On the other hand, the wrong cognitive system will inevitably drag down the pace of the knowledge revolution.

end
The Qin people 2000 years ago still ate the same grains as the ancients 100,000 years ago, but everything from trains to the Internet is accelerating. In the past two centuries, the remains of organic matter we burned are fossil raw materials formed after hundreds of millions of years of transformation. These burnings have already caused huge consumption in the fourth quarter of the planet, and have also profoundly affected the evolution of the planet’s diverse life balance. In the process of development, responsible leaders set a carbon neutral goal, master the law of power law, and learn to use more "momentary" technological power to consume increasingly scarce and short time-space.
——Excerpt from "Data Native View of Time and Space" Author: Lei Tao