How to apply artificial intelligence "hot technology"

Guangming Daily reporter Yuan Yufei

Artificial intelligence composing poems, writing couplets, the latest progress in neuromedicine artificial intelligence research, artificial intelligence traffic fusion perception and digital twin solutions, precision medicine auxiliary diagnosis platform... On October 26, the 2021 Artificial Intelligence Computing Conference was held in Beijing, and a number of innovative achievements in the application of artificial intelligence technology attracted many audience interactions.

At the meeting, experts, including Wang Endong, academician of the Chinese Academy of Engineering and chief scientist of Inspur, discussed in depth how computing transforms into intelligent computing under the new pattern of digital economy, how intelligent computing can empower scientific and technological innovation, social governance and industrial upgrading, and answered hot issues in the artificial intelligence industry such as how artificial intelligence develops human logic like human logic and how to combine with application scenarios.

"Artificial intelligence has changed from 'black technology' five or six years ago to 'hot technology' today, and we are seeing cutting-edge research emerge, such as predicting human protein sequences through the AlphaFold2 model, and letting monkeys play games with their minds through brain-computer interface research." At the same time, we also see that artificial intelligence is deeply integrated with various industries, changing the production mode of the primary, secondary and tertiary industries, and various industry brains and unmanned operation models are emerging, and these new infrastructures are accelerating the arrival of the era of wisdom. Wang Endong said that the key to artificial intelligence becoming a "hot technology" lies in strengthening new infrastructure and releasing the value of multiple computing power, of which the innovation of computing systems is the key.

In 2020, the total computing power of AI-accelerated chips exceeds that of general-purpose CPUs

The theme of this year's Artificial Intelligence Computing Conference is "Intelligent Computing , New World". At the conference site, the world's largest Chinese AI giant model "Source 1.0" developed by Inspur Artificial Intelligence Research Institute became the focus of the audience, and a large number of participants lined up to interact with "Source 1.0" to personally experience the transformation of content production methods driven by artificial intelligence.

"In 2020, the sum of computing power delivered by AI acceleration chips has already surpassed that of general-purpose CPUs (central processing units). It is expected that by 2025, the computing power provided by acceleration chips may exceed 80%. Wang Endong said.

"With the large-scale development of artificial intelligence, computing power has become a decisive force, and smart computing is the core productivity of the wisdom era." Wang Endong said that artificial intelligence brings exponential growth in computing power demand, and the computing industry is facing the trend and challenge of diversification, huge quantification and ecological discretization. On the one hand, diversified intelligent scenarios require diversified computing power, and huge quantitative models, data and application scale require a huge amount of computing power, which has become the top priority of the continued development of artificial intelligence; on the other hand, there is still a huge gap in the conversion from chips to computing power, and the value of multiple computing power has not been fully released. How to quickly complete the innovation of multiple chips to computing systems has become a key link in promoting the development of the artificial intelligence industry.

How artificial intelligence develops logic like humans

How artificial intelligence develops cognitive abilities like humans with logic, consciousness and reasoning is the direction that artificial intelligence research has been exploring.

"At present, training huge models with very large parameter quantities through large-scale data is considered to be an important direction for achieving general artificial intelligence." Wang Endong believes that with the rise of giant models, giant quantification has become a very important trend in the future development of artificial intelligence.

The world's well-known AI leading companies have invested heavily in huge models, and Google, Microsoft, Nvidia, Inspur, Zhiyuan Research Institute, Baidu, Ali and other companies have launched their own huge models.

Wang Endong introduced that one of the core characteristics of giant quantization is that there are many model parameters and a large amount of training data. "Taking 'Source 1.0' as an example, the number of parameters is as high as 245.7 billion, and the scale of the training dataset reaches 5,000 GB."

Applications are facing a dilemma, how to combine artificial intelligence with application scenarios

Many people will have such a confusion: artificial intelligence is so good, but how to combine with my business and application scenarios? I want to do intelligent transformation through AI technology, but no one understands the algorithm to understand the model, and there is a lack of easy-to-use AI development platform, there are so many algorithm models, how to find the optimal combination of different algorithms in the application?

"People who understand these things are often concentrated in scientific research institutions or head companies." These places concentrate the best AI talents, but lack in-depth understanding of the demand scenarios and business laws of traditional industries. For the current dilemma faced by artificial intelligence from technology to application, Wang Endong pointed out.

According to a survey report from Accenture, more than 70% of research institutions and technology companies with technology lack demand scenarios, domain knowledge and data, and more than 70% of industry users lack technical talents, AI platforms and practical capabilities.

Wang Endong believes that at present, the technology and industrial chain of artificial intelligence are disconnected, and ecological discretization has become a bottleneck restricting the level of artificial intelligence technology, the scale of application, and the level of industry. "In order to release the value of multiple computing power and promote the innovation of artificial intelligence, we must not only attach importance to the innovation of intelligent computing systems, increase the construction of new artificial intelligence infrastructure, design the chain from technology to application, form a situation of clear division of labor and collaborative innovation in various fields such as architecture, chip design, system design, system software, and development environment, and accelerate the construction of open standards, through unified and standardized standards, transform diversified computing power into schedulable resources, and make computing power easy to use and easy to use."

Guangming Daily (08/10/2021)

Source: Guangming Network - Guangming Daily