NVIDIA CEO Huang Renxun interview, what is AI?How to change our lives

NVIDIA has successfully transformed from a hardware company focusing on display technology to a computing company that provides a complete stack of software and hardware. Let us hear the views and vision of NVIDIA founder and CEO Huang Renxun on AI computing.

AI changes the way computers “think”

The author had the opportunity to interview Huang Renxun, the founder and CEO of NVIDIA during the Computex 2022, so the information from the previous GTC events and the views put forward by Huang Renxun were integrated into this special report.

AI is the abbreviation of Artificial Intelligence in English. The intuitive response of ordinary people may be linked to science fiction films. Imagine that the specific performance of AI is a realistic robot that behaves like a real person, can think and talk on its own, but in fact, the function of AI is not in this way. Perhaps with the development of science and technology, humans can really make this kind of robot one day, but the current technology is still far from this kind of achievement.

The most prominent project in the field of AI today is Machine Learning (ML). With reference to the Azure documentation provided by Microsoft, AI is the ability of computer systems to imitate human cognitive functions and use mathematics and logic to simulate what people use to learn new information and The way of making decisions, and machine learning is one of the applications of AI. It uses mathematical data models to help computers learn without using direct instructions, allowing the computer system to continue to learn and improve on its own based on experience.

For example, in the past, if we wanted to use traditional programs to tell the computer whether the fruit in the picture was an orange or not, the program developer had to write a variety of judgment rules such as orange color and round shape. When judging the item in the image, it will check whether the item in the image conforms to these rules. When more rules are written, the more accurate the effect will be.

When using machine learning, after writing AI-specific programs and algorithms, you only need to provide a large number of orange pictures to the computer for “AI training” (generally, the more training data you input, the more accurate the results will be) , then the computer will automatically analyze all the characteristics of the orange, and produce an “AI model” that can be used to distinguish the items in the picture. Finally, we will perform “AI inference” on the new picture, and the system will tell us what is in the picture. what is the item. Interestingly, however, we do not necessarily know which features the system will ultimately choose as a basis for judgment.

For example, when I actually tested the NVIDIA Jetson AGX Xavier AI computing platform, I deliberately misled the AI ​​model with a photo of a loquat, and the result was an 81.458% chance of being an orange. Although the author did not analyze the reasons in detail, it is preliminarily speculated that AI may be because the characteristics of loquat are similar to oranges, and the AI ​​model trained with common fruits in Europe and America may not have enough loquat data, so this misjudgment is caused.

In the interview, Huang Renxun responded to many different questions and also shared his views on the development of AI.

For example, if you input a photo of an orange into a trained model, the AI ​​will recognize that the object has a 99.847% chance of being an orange.

Deliberately input the photo of loquat, the identification result is 81.458% chance of being an orange.

AI is a parochial superman

In this interview at Computex, Huang Renxun made a footnote for AI as “AI is superhuman for narrow tasks.” In contrast, humans can accomplish more tasks that AI cannot. .

For example, the technology of using AI for image recognition and classification is quite mature, and the technology of using AI to enhance and predict images has also achieved good results. Therefore, in addition to the fruit recognition mentioned in the previous article, NVIDIA also launched DLSS. Technology, improve the quality of the game screen through AI, and apply AI to predict the climate. It is planned to establish a digital twin of the earth through Omniverse in the supercomputer E-2 (Earth-2, Earth 2), and simulate it through the Modulus AI physical model. Changes in Earth’s climate.

Further reading:
Google uses AI to predict weather changes, providing more refined Nowcast forecasts
(Think of weather forecasting as an image-to-image translation problem, predicting changes with convolutional neural network techniques)

NVIDIA has also further introduced a variety of different uses, such as enhancing ultrasound, MRI and other medical images and accelerating drug research and development through AI, into the AI-assisted DRIVE Hyperion 8 vehicle computer platform, and even launched the cuQuantum software development kit to accelerate quantum computer simulation. , play the role of AI in many different fields.

However, as these AI applications have in common, their respective AI programs are only applicable to very narrow specific situations, and cannot be used to solve all kinds of strange problems encountered by humans.

With the assistance of AI, Lockheed Martin uses accurate physical simulations on the Omniverse visualization and simulation platform to predict fire trends in the digital twin, and allow the system to recommend actions to suppress the fire.

NVIDIA is accelerating the simulation of quantum circuits through the cuQuantum software development kit and embarking on the development of a hybrid computing architecture where traditional and quantum processors coexist.  (Image credit: NVIDIA).

(The next page also includes Huang Renxun’s views on the singularity of technology)

Similar Posts

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.