Chen Xudong, chairman of IBM Greater China: artificial intelligence is ushering in a "moment of qualitative change"

21st Century Business Herald reporter Yang Qingqing and intern He Sijia reported from Beijing

"The emergence of ChatGPT proves that the big language model is a feasible road to future AI, and it also means that the development of AI has been accumulated in terms of algorithms, computing power, and data for decades, and the 'moment of qualitative change' has arrived."

Recently, Chen Xudong, chairman and general manager of IBM Greater China, pointed out when talking about the current status of artificial intelligence development.

In order to meet the current "qualitative change moment", IBM is also accelerating its actions. Following the announcement of IBM watsonx products in July this year, IBM watsonx has also officially landed in the Chinese domestic market.

In fact, back in 2019, IBM shifted its focus to hybrid cloud and artificial intelligence. As the demand for enterprises to deploy generative AI increased, IBM accelerated its pace, releasing watsonx, its enterprise-grade AI and data platform, in May, and starting in July, various modules will be available one after another. It is expected that all modules will be completed around the end of the year to the beginning of next year, and will be put into use by customers one after another.

In Chen Xudong's view, after realizing the current boom in large models, it is not difficult to understand the reasons behind IBM's rapid response. "First of all, we see that 'making AI a core productivity' has become an urgent need for business leaders, and at the same time, this market opportunity provides a golden opportunity for IBM to accumulate decades in the field of AI."

In IBM's view, it is more feasible to use the company's own data according to its own business needs, use its own data, and tailor generative AI solutions and models for it.

Chen Xudong, chairman and general manager of IBM Greater China, source: photo provided by interviewee

From "+AI" to "AI+"

According to a recent survey released by the IBM Institute for Business Value, three-quarters of CEOs surveyed believe that deploying advanced generative AI will give their organizations a competitive advantage, while 61% of CEOs also expressed concerns about the sources of data used in generative AI.

"This concern reflects the challenges on the road of enterprise AI." Chen Xudong pointed out to the media, including the 21st Century Business Herald, that the current enterprise AI road is first facing technical challenges, especially the preparation, application and governance of data; The second is the talent challenge, enterprises need to quickly transform and improve personnel skills to embrace the wave of AI; The third is cultural challenges, the transformation of skills is often accompanied by the renewal of organizational culture, how to make the two achieve each other, bring about productivity improvement, which requires excellent management wisdom.

At the same time, under the current AI boom, enterprises have begun to shift from the paradigm of data-first "+ AI" (that is, local application of AI) to AI-first "AI+", and all walks of life are seeking how to embed AI into the core of their strategy.

What is pushing enterprise-level AI to an inflection point is the foundational model technology, which makes it possible for enterprises to adopt and scale AI at scale.

Specifically, the base model is built on a specific type of neural network architecture (called a Transformer architecture) designed to generate sequences of related data elements, such as sentences. The Transformer architecture helps the base model understand unlabeled data and transform input into output to generate new content, and the base model is the source of generative AI derivation. By training on large amounts of unlabeled data, the underlying model can adapt to new scenarios and use cases.

It is important to note that although the base model also requires a significant upfront investment, it amortizes the initial work of AI model building each time it is used, as fine-tuning the data requirements for other models built on top of the base model is much lower than building from scratch. This can both significantly improve the return on investment (ROI) and significantly reduce time to market.

Therefore, the underlying model should be the focus of the enterprise, which makes it possible for enterprises to accelerate and scale generative AI, which is an indispensable component of the next-generation AI workflow.

"There is a difference between the basic model and the deep learning model," Xie Dong, chief technology officer and general manager of IBM's R&D center in Greater China, told media including the 21st Century Business Herald.

In two years, IBM expects the underlying model to power about one-third of AI in enterprise environments. In IBM's early work applying the underlying model to a client, the client's time to value was 70 percent faster than traditional AI methods. In this regard, IBM is working to develop foundational models for business scenarios that need to leverage large language models (LLMs), IT automation models, digital workforce models, cybersecurity models and many other specialized models.

"Today's enterprises want to use artificial intelligence in their core applications to enhance actual productivity, and we need to move from the previous data-first '+AI' era to the AI-first 'AI+' era." Xie Dong said, "When we come to a new architecture such as AI+, we consider the basic capabilities of artificial intelligence that the enterprise has established, and on the basis of this basic capability, we let it combine the company's own data and its different business goals to build new applications, which is the transformation process from +AI to AI+." ”

Watsonx landed in China

The new AI paradigm triggered by the basic model undoubtedly requires new tools and project management methods for training and deployment.

In the first half of this year, the ChatGPT fire caused a surge in interest in generative AI and large-language models, and an eagerness to apply new technologies to improve competitiveness. In response to customer demand, IBM further introduced Watsonx.

According to reports, watsonx is a new generation of enterprise-level AI and data platform for basic models and generative AI, providing a toolkit including AI development platform (watsonx.ai), lakehouse integrated solution (watsonx.data) and AI governance (watsonx.governance) to accelerate the use of trusted data while responsibly applying AI at scale.

Among them, watsonx.ai is a next-generation enterprise-level AI development platform, empowering AI builders such as engineers and data scientists to assist in training, verification, tuning, and deployment of AI models. It supports both traditional machine learning and generative AI based on pre-trained base models, allowing users to quickly build AI applications with small amounts of data.

watsonx.data is an all-in-one data storage solution that helps enterprises meet data challenges. It has a data storage based on the integrated architecture of open lakehouses, which can be specially optimized for regulatory data and AI workloads, and finally realize data access and sharing in open data formats.

watsonx.governance is an AI governance toolkit responsible for AI governance and governance to build trust in the AI lifecycle of enterprises. It has end-to-end complete tool support and complete enterprise data and information security to support responsible, transparent, and explainable AI workflows.

Currently, watsonx.ai and watsonx.data are online. Chen Xudong revealed to the 21st Century Business Herald reporter that watsonx.governance is expected to complete the listing at the end of this year to the beginning of next year.

It is important to note that these three are not independent of each other, but work together to massively scale and accelerate the impact and value of AI through trusted data. watsonx.ai leverage the base model to automatically search, discover, and link data in watsonx.data, watsonx.data leverages the data of internal governed enterprises to seamlessly train or fine-tune the base model watsonx.ai enable fine-tuning models for governance and governance through industry-leading governance and lifecycle management capabilities.

Miao Keyan, general manager of IBM's Greater China Technology Division and general manager of China, said that the "X" in WatsonX represents unknown infinite possibilities, and the "X factor" is the key to miracles. "This X-factor today has three aspects: open technology for good, visionary leadership, and an ecosystem that works together."

For more information, please download 21 Finance APP