Author | Wang Qilong, Tang Xiaoyin
Listing | CSDN (ID: CSDNnews)
[CSDN Editor's Note] In the development history of artificial intelligence, which has experienced several ups and downs, and in the milestone evolution of artificial intelligence, we will always see the figure of IBM, a century-old enterprise. In addition to the well-known 1997 IBM Deep Blue computer defeating world chess champion Garry Kasparov in a chess match (and rematch), there are also many events such as the checkers human-machine war in the last century. Today, in the era of large models, IBM has brought a new platform for the basic model watsonx and generative AI - watsonx.
Chen Xudong, Chairman and General Manager of IBM Greater China, said: "The emergence of ChatGPT proves that large language models are a viable path to the future of AI. It also means that the development of AI has gone through decades of algorithm development. , the accumulation of quantitative changes in computing power and data, the 'moment of qualitative change' has arrived."
Source: Photographed by the author in IBM Beijing
In 2011, IBM (International Business Machines Corporation) officially entered the ranks of the century-old enterprise. It is one of the most influential technology companies in the world, with a rich history and excellent technological tradition. IBM held a grand celebration to commemorate its 100-year glorious history, used a short film to condense IBM's century-old history into a few intoxicating minutes, and provided a full year of volunteer activities in all corners of the world.
Until now, we can still travel through time and space through the IBM 100 web page, looking back from the mechanical tabulation machine to browsing every innovation of atomic rearrangement technology, Fortran, RISC, mainframes, personal computers, minicomputers, Deep Blue, and Watson.
2023, the twelfth year after IBM's centennial celebration, is the year of AI. Companies from all walks of life are jumping on the intelligence bandwagon, developing large-scale models of all kinds, launching generative chatbots that can plan our vacations, AI smart assistants that organize corporate data, and AI services that can create images and videos. In August this year, IBM followed this trend and announced the launch of a new generation of AI and data platform, IBM watsonx, to provide power for enterprise-level basic models and generative AI.
This is another landmark open AI technology platform launched by IBM in the Greater China market after Red Hat OpenShift open hybrid cloud technology platform.
"Watson" is the name of IBM founder Watson Sr. and the name of Thomas J. Watson Sr., the founder of IBM culture. Over time, the name became more symbolic, representing IBM's ambition and innovation in the field of artificial intelligence. If we want to trace back these histories, we have to start with those humble chess pieces.
AI that started with chess
Computer chess has been associated with the field of artificial intelligence since the emergence of artificial intelligence and the first computers in the late 1940s.
The two fields intertwined many well-known figures, such as Alan Turing, the founder of modern computer science; John McCarthy, the coiner of the term "artificial intelligence"; Boolean logic pioneer G. Claude Shannon, who was the first to support the use of chess as an appropriate starting point for intellectual development.
Among others, Turing, undeterred by the lack of technology and established precedent at the time, sought support for his theory and wrote the algorithm for the world's first chess program. Unfortunately, this program, written down on paper and implemented by Turing himself, was quickly and easily defeated by a colleague.
Since then, a large number of programmers have devoted time and energy to studying chess, believing that the game will lead to breakthroughs in the field of artificial intelligence. These researchers saw in the structure of chess a simplified model of large-scale problem solving. From a human comparison perspective, the hierarchies of chess players that already exist around the world provide engineers with a yardstick by which to easily and accurately measure machine prowess.
Decades of research have made little progress, and many computers have played against numerous chess grandmasters, with no winners. It was not until 1985 that Feng-hsiung Hsu, a graduate student at Carnegie Mellon University, began working on his thesis project: ChipTest, a chess machine. Xu Fengxiong's classmate Murray Campbell also participated in it and was hired as a staff member of IBM Research in 1989. They joined the IBM team and named the project "Deep Blue."
With its powerful hardware system and innovative evaluation functions, Deep Blue One launched a six-game chess match in 1996 against the then world chess champion Garry Kasparov. To the surprise of many, Deep Blue won the first game directly, marking the first time a machine had defeated a world chess champion. However, Kasparov was not deterred by the early defeat and ended up winning the entire match by a score of four to two.
After updating and improving its chess knowledge several times to counter the strategies used in the previous game, a rebuilt Deep Blue defeated Kasparov in the 1997 rematch.
During the game, the unconventional openings, psychological intimidation and timeout tactics that Kasparov is good at have no impact on the machine, because Deep Blue will only calmly identify and analyze the situation on the chessboard: the threats and threats displayed by human players on the chessboard. Emotions don’t play much of a role in a machine’s assessment of a situation.
The architecture used by Deep Blue was not limited to the competition field, but was quickly applied to financial modeling, data mining, and molecular dynamics. Eventually, the Deep Blue was retired and stored in the Smithsonian Museum in Washington, D.C., ending its glorious life. In the field of AI, IBM is not slacking off, but has invested in another major challenge: building a computer that can beat human champions in more complex games.
Can AI think?
In 2011, IBM's Watson computer competed on the television quiz show "Jeopardy!" against two of the show's greatest ever champions. The Watson computer, developed by IBM Research, can run a software called Deep QA. While the main challenge of this project is winning Jeopardy!, Watson's real goal is to create a new generation of technology that attempts to defeat standard search techniques and find answers from unstructured data more efficiently.
Sound familiar? Is it very similar to the popular conversational artificial intelligence-ChatGPT? For IBM scientists, this is not a difficult idea to think of, but the miracle it emerged failed to appear in advance in 2011. Instead, it required 12 years of data accumulation.
On the show, Watson excels at understanding complex problems and finding the best answers. IBM scientists also provided an explanation, pointing out that Watson cannot actually think.
"The goal is to build a computer that can understand and use natural language more efficiently, but doesn't necessarily need to think in the same way as a human, rather than imitating the human brain," said Nicolas, who has worked at IBM Research for 15 years. says David Ferrucci, who specializes in natural language problems and finding answers in unstructured information.
Jeopardy!'s questions are full of subtlety, puns, and wordplay, and these classic brainteasers are packed with elements that a computer would struggle to handle. Computers have never been good at finding answers. Search engines don't answer questions, they just provide thousands of search results that match keywords.
University researchers and company engineers have long worked to create the perfect question-and-answer software, but the best software can only understand and answer simple, straightforward questions—and even today in 2023, large question-answering models often fail They lied to us with a bunch of false facts.
Watson runs on a set of Power 750 computers, including ten racks, 90 servers, and a total of 2,880 processor cores. It can hold the equivalent of approximately one million books. IBM spent years allowing Watson to absorb a vast amount of information, including text from commercial sources as well as sources that allowed their content to be publicly reproduced (such as Wikipedia).
When a host poses a question to Watson, more than 100 algorithms analyze the question in different ways and find many different reasonable answers simultaneously. Another set of algorithms ranks the answers and gives them a score. For each possible answer, Watson finds evidence that might support or refute that answer. However, in a Jeopardy game, if the top possible answer is not rated high enough to give Watson enough confidence, Watson will decide not to press the button to avoid losing points for a wrong answer.
In late 2010, Watson was put through a round of testing, winning about 70 percent of its games against former Jeopardy! Then, in February 2011, the latest iteration of Watson beat Jeopardy's human superstars Ken Jennings and Brad Rutter.
"I wanted to create something that could be introduced into every other retail industry, transportation industry, etc., because everywhere time is of the essence and cutting-edge information needs to be delivered to the front line," said John Kelly, the father of Watson Decision makers. Computers need to move from logistics to tools that enhance human decision-making intelligence.”
Deep Blue and Watson respectively represent IBM's two peak moments in the field of AI. Deep Blue's victory demonstrated the potential of machines in highly specialized fields, while Watson introduced AI into new areas of natural language processing and business applications. Both milestones have laid a solid foundation for IBM's future exploration in the field of AI.
Embrace basic models and generative AI
In the 1980s, IBM was like a behemoth on top of Silicon Valley, waiting for challenges. Microsoft and Apple are eyeing it, and countless start-ups are eager to give it a try. IBM has been at the forefront of breakthrough artificial intelligence technology for decades. From Deep Blue to Watson to the cloud native era, AI supercomputers have been built on the basis of cloud computing. However, the AI winter has been too long, and many companies have given up exploring the field of artificial intelligence and invested more resources in other fields.
As ChatGPT ignited this cold winter, enterprises have paid increasing attention to generative AI and large language models, and are eager to apply new technologies in the AI field to enhance competitiveness. At a press conference in August this year, IBM combined the technology and experience accumulated in the enterprise-level AI field over the years with the progress made in basic model research in the past five years to launch a new generation of data and AI platform IBM watsonx.
watsonx is a comprehensive platform launched by IBM to solve the challenges of enterprises in data management, model development, verification, deployment and governance in artificial intelligence applications. The platform includes three key components: watsonx.ai, watsonx.data and watsonx.governance, which together form a complete AI life cycle solution.
watsonx.ai is the core of the platform, focusing on the training, verification, fine-tuning and deployment of AI models. It provides enterprises with powerful computing resources to handle large-scale model training tasks. With watsonx.ai, AI builders can use IBM's models and Hugging Face's models to complete a series of AI development tasks.
These models are pre-trained to support a range of natural language processing (NLP) type tasks, including question answering, content generation and summarization, text classification and extraction. watsonx.ai allows enterprises to choose a model architecture that suits their needs and evaluate and select the best model based on benchmark results of model performance.
In his speech at the press conference, Xie Dong, Chief Technology Officer of IBM Greater China and General Manager of the R&D Center, also gave a detailed introduction to IBM’s recent AI research: “Recently, many people have asked me a question: Is IBM still continuing its AI research? Developing artificial intelligence? I want to tell you that IBM has always been a leader in hybrid cloud and AI.
IBM has been a technology sponsor of several major sporting events over the years, including the Masters, Wimbledon and the U.S. Open.
The reason why I mentioned these sports events is not only because I love sports, but more importantly, the generative artificial intelligence technology based on Watsonx used in these competitions is also the AI technology we use to empower applications in all walks of life. ——Provide more accurate predictions (such as a player's winning rate in games) based on data and knowledge in specific professional fields (such as tennis, golf).
But someone may ask, if artificial intelligence is so powerful and interesting, can it be easily applied to various fields? I want to tell you that this is not the case. Although many people are familiar with large language models such as ChatGPT, the actual application of AI requires consideration of multiple factors, including model selection, data utilization, development environment, computing power, etc.
This is the challenge many businesses are facing right now. As large language models continue to emerge, enterprises need to carefully consider which model to choose, how to make full use of their own data, how to develop applications and deploy them, and other issues. IBM is committed to solving these problems and providing complete artificial intelligence solutions for enterprises. "
Xie Dong, CTO of IBM Greater China and General Manager of R&D Center
watsonx.data is a key part of data management, helping enterprises to efficiently prepare, filter and clean data for training models. Classification and tokenization of data are important steps in the data preparation process. watsonx.data also allows enterprises to use their own data or IBM data stacks to meet model training needs, and provides data business cards and version control for traceable data management.
Data - this word has been lingering in the minds of countless technical people, and it is an important page in the history of IBM. IBM had a profound impact on the database field in the 1970s. In 1970, IBM scientist Edgar F. Codd proposed the theory of relational database, which later evolved into SQL (Structured Query Language) and became the relational database management system (RDBMS). laid the foundation for its development.
In the era of IBM Watson mentioned above, Watson uses large-scale databases to build knowledge graphs to help users obtain answers to complex questions. Watson is also used in the medical field to assist doctors in formulating diagnosis and treatment plans by analyzing millions of medical documents and case databases.
Today, Watson’s name continues. Later this year, watsonx.data will leverage watsonx.ai’s foundational models to help simplify and accelerate the way users interact with data, enabling them to discover, enhance, optimize and visualize it using natural language with a conversational user experience. Data and metadata.
watsonx.governance focuses on compliance and governance of enterprise-level AI. Just like the name of this conference-"The Future of Enterprise-level AI". It ensures that models and data comply with laws, regulations and ethical standards, especially when handling sensitive information and private data. During the speech, the importance of governance was emphasized to ensure that enterprises can trust their AI systems. watsonx.governance also provides fact tables for recording model and data details for monitoring and updating.
A cross-century AI explorer
Deep Blue and Watson are both early AI projects. The author has seen countless forum netizens and celebrities and scholars criticizing them. These machines have been criticized as having "no real intelligence."
Although Deep Blue is very powerful at calculating chess positions and searching for possible moves, it has been criticized as a "hard-coded" approach, that is, it has no real intelligence or reasoning capabilities. Deep Blue's victory mainly relied on its excellent calculation speed and search algorithm, rather than its understanding of the chess game.
Criticisms of Watson mostly relate to its uncertainty management and reasoning capabilities when answering questions. Although Watson can handle large amounts of information and text data, it sometimes gives inaccurate answers because it cannot truly understand the context of the question and only bases its answer selections on statistical probabilities.
Now, big models are getting the same flak.
It still cannot reason, think, or have emotions, but it is as useful as IBM's several attempts and has injected new vitality into the entire industry. Just as Deep Blue and Watson have made significant contributions to IBM's professional fields after being put into use, large models are also nourishing all walks of life.
The field of artificial intelligence is still developing, and countless companies and scientists are working tirelessly to pursue more optimized methods to handle cognitive tasks and reasoning problems to achieve higher levels of intelligence. Despite the challenges, the future of AI remains promising.
