Can artificial intelligence also undergo physical examinations? The first batch funded artificial intelligence quality inspection center in Shanghai uses such tools | artificial intelligence | quality inspection center
A cute panda image, with some interference information added, is recognized by artificial intelligence as a "gibbon"; A traffic sign that represents "stop" is mistakenly thought to be a "green light" by artificial intelligence after adding some black and white squares. "We will conduct a 'physical examination' of various indicators of the artificial intelligence software submitted for testing and provide a detailed 'physical examination report'," Cai Lizhi, director of the Shanghai Computer Software Technology Development Center directly under the Shanghai Academy of Sciences, told the Liberation Daily Shangguan News reporter. ".
Just last month, the Shanghai Generative Artificial Intelligence Quality Inspection and Testing Center, established by the Shanghai Computer Software Technology Development Center, was listed as the first quality inspection center in the field of artificial intelligence in Shanghai to be approved.
Recently, generative artificial intelligence represented by ChatGPT has become a hot topic. The "Research Report on China's Artificial Intelligence Large Model Map" released in May this year shows that China has 79 large models with parameter scales of over 1 billion. How to objectively evaluate the performance of large models has become a crucial aspect of their application. Just this month, the National Cyberspace Administration and seven other departments jointly issued the "Interim Measures for the Management of Generative Artificial Intelligence Services". The implementation of the management measures urgently requires professional third-party institutions to provide artificial intelligence quality inspection services.
It is reported that the Shanghai Generative Artificial Intelligence Quality Inspection Center focuses on the inspection and testing of artificial intelligence in the fields of generative artificial intelligence, industrial intelligence quality inspection, smart healthcare, facial recognition, and autonomous driving.
"Unlike traditional quality inspections that 'hold onto the bottom line of quality', our positioning of artificial intelligence quality inspection is to help enterprises improve their innovation capabilities on the one hand, and also to help enterprises maintain the red lines of regulations, laws, and ethics and morality on the other hand." Chen Mingang, Deputy Director of the Software Engineering Research Institute of Shanghai Computer Software Technology Development Center, introduced. For example, in the early stages of some big language model applications, some malicious users may evade the security rules of the big language model through role-playing, step-by-step questioning, and other methods, inducing them to give answers that violate laws and regulations, or generating content that violates ethical and moral principles. Our detection aims to identify similar vulnerabilities, enhance the security of artificial intelligence systems, and reduce various risks.
It is reported that the quality inspection center will refer to domestic and foreign artificial intelligence evaluation standards, independently develop artificial intelligence deep security and evaluation technologies and tools, including adversarial sample generation and defense evaluation tools, natural language processing system evaluation data generation tools, artificial intelligence false face detection tools, artificial intelligence risk assessment and governance tools, etc., to provide enterprises and institutions with "one-stop" artificial intelligence quality evaluation services.
In recent years, under the deployment of the Shanghai Academy of Sciences, the Shanghai Computer Software Technology Development Center has laid out and carried out technical research and third-party technical services on the application, governance, and evaluation of artificial intelligence. It has participated in the writing of the "Standardization White Paper on Artificial Intelligence" and completed the development of more than 30 standards, including the "Technical Specification for Deep Synthetic Image System of Artificial Intelligence" and the "Evaluation Specification for Computer Vision System of Artificial Intelligence". In the past three years, it has provided third-party evaluations for artificial intelligence enterprises and application parties with an average annual average of over 500 items.