2023 Trusted AI Technology and Application Progress White Paper - 74 pages
Content guide
The rapid progress and application of artificial intelligence have brought unprecedented opportunities and challenges, and also pushed trusted artificial intelligence to the forefront of industrial development. In 2021, the "Trusted Artificial Intelligence White Paper" of the Chinese Academy of Information and Communications Technology systematically extracted the more unified trusted AI feature elements in the industry for the first time, and under the framework of security robustness, privacy protection, fairness, and explainability, trusted AI is accelerating in finance, insurance, manufacturing, medical and other fields. With the continuous consolidation of the foundation of artificial intelligence technology and the explosive growth of large models and AIGCs, artificial intelligence has taken a key step towards general artificial intelligence, and the Politburo meeting of the CPC Central Committee in April 2023 emphasized "attach importance to the development of general artificial intelligence, create an innovation ecology, and attach importance to risk prevention", and trusted AI has become an important technical means to balance innovation and risk in the new stage.
Based on the unified cognitive dimension of trusted AI, this report summarizes the development status of trusted AI technology and application, extracts the new requirements caused by large models and AIGC in the new stage, and forms a trusted AI evaluation system and tools guaranteed by technology. The main points of the full text are as follows:
The first is to summarize the industrial technology background of the development and change of trusted AI, that is, the recent algorithm, computing power, and data foundation have been continuously consolidated, the inherent technical risks of artificial intelligence have continued to amplify, trusted AI technology has become a key underlying capability in the field of AI, and countries are also promoting governance around generative AI regulation and AI ethics.
The second is to sort out the development of the four major technical directions of trusted AI at this stage. Trusted AI has been implemented into specific products and practice cases in the industry, and is generally moving from the stage of innovative integration solutions to the stage of forming a trusted system mechanism, and the complete practice link, sound standards and system mechanisms will gradually land.
The third is to summarize the trusted AI exploration for large models and AIGC, that is, the security and trust challenges that may be faced in the era of large models, mainly including the generative attack and defense of large models, the security risks and hidden dangers of large models themselves, and the interpretability and fairness of large models
Three broad categories. It builds a trusted AI evaluation system guaranteed by technology, and builds a trusted AI detection index system including privacy protection detection, interpretability detection, fairness detection and robustness detection to meet the needs of industrial development, and puts forward specific evaluation content, indicators and methods, which can further empower the development of large models and AIGC.
Finally, this white paper puts forward suggestions for the development of trusted artificial intelligence from the technical, industrial, ecological and regulatory dimensions, and looks forward to the trend, which will usher in a more balanced research and application, more multi-dimensional internal dynamics, and more comprehensive external supervision.
Core content
Chapter 1: Artificial intelligence is entering a new stage of development
Chapter II: In-depth Development of Trusted AI Industry Practice
Chapter 3: Trusted AI Exploration for Large Models and AIGC
Chapter 4: Trusted AI Assessment Backed by Technology
Chapter V: Development Recommendations and Trend Outlook
Attached: Comparative table of English abbreviations
