From AI to conquer the sky, to employment tools out of the mountains
Author/ IT Times reporter Sun Yan
Editor/Sun Yan
In 3 years, the boots landed and Ant Group completed rectification.
Ants are moving forward in adjustment, but what has not changed over the hill is the core gene of technology. In the past three years, the annual World Artificial Intelligence Conference (hereinafter referred to as WAIC) has been the "award" of Ant Group's AI technology, and has also witnessed the evolution of Ant "root technology". In fact, 3 years ago, before preparing for the IPO, Ant Group had clearly positioned itself as a technology company, not a financial company.
At WAIC 2023, half of the domestic large models will come to "show muscles", or it will become the beginning of the "100-model war". But ants have a different way, when AIGC is plagued by frequent hallucinations, uncontrollable results, privacy security, discrimination and prejudice, etc., ants sacrificed a trick to AI to "dispel poison": AI security detection platform "Ant Jian 2.0", using generative AI capabilities to detect generative AI models.
At WAIC 2022 last year, Ant demonstrated the three "root technologies" of privacy computing, blockchain, and graph computing. At WAIC 2023 this year, these three "root technologies" came to fruition, and "Antjian" was upgraded to "Antjian 2.0", adding AIGC safety detection; The privacy computing "Hidden Language Technology Stack" was upgraded to open source, launched the Hidden Language Framework 1.0, and open-sourced the first domestic financial security TEE solution "HyperEnclave"; AntChain FAIR, an Ant blockchain privacy collaboration platform, has upgraded its new architecture and introduced verifiable computing technology with zero-knowledge proof (ZKP) as the core, making it a step further in tackling the problems of Web3 technology.
Large model "safety medical examination software"
The "100-model war" to open China's AI catch-up race is naturally not a bad thing, but when humans begin to rely on the big model as a production tool, if the big model is talking nonsense, who will judge whether it is right or wrong, and will the fake data it produces pollute the data pool?
At WAIC 2023, Ant Group and Tsinghua University released the AI security detection platform "Ant Jian 2.0", which uses generative AI capabilities to detect generative AI models, which can identify hundreds of risks in three categories: data security, content security, and science and technology ethics, covering a variety of data and task types such as tables, texts, and images, and is the industry's first all-data AI security detection platform that realizes industrial-level applications.
In booth C, the "IT Times" reporter saw the "treasure of the town hall" of this conference - "Ant Jian 2.0", the "Ant Robot on the left" automatically generates a massive test set, interactive induction of the large model on the right, and more than 10 confrontation methods with different difficulty coefficients can be called every minute to generate more than 30 induction problems. It is very similar to a 24-hour sleepless "security hacker" looking for a large model, more intelligent than the "security hacker", at the same time, the answer of the large model will be real-time, automated detection and calculation, from nearly 200 subclass latitude, find the weaknesses and security problems of the large model, and finally generate the security physical examination report of the large model.
"Generative big model is a kind of 'deep black box' technology, this kind of AI evaluation of AI is teaching and learning, just like generative examiners assess generative athletes, in the confrontation both sides improve each other's capabilities, further improve the AI safety level of the entire industry." Li Junkui, president of the technology department of Ant Group's large security business group, said.
Another hardcore capability is that "Antjian 2.0" integrates interpretability detection tools, quantitatively analyzes the interpretation quality of AI systems from 7 dimensions such as completeness, accuracy, stability and more than 20 evaluation indicators, helping users more clearly verify and optimize interpretable schemes.
"Antjian 2.0" integrates Ant's nearly 10 years of trusted AI practice and more than 1,000 trusted AI patents, and supports zero-coding evaluation. Developers can connect AI models to Ant Discovery through API interfaces to identify and mine model vulnerabilities with one click.
Ant Group announced that "Ant Jian 2.0" is fully open, providing free AIGC security, AI explainability, and AI robustness detection tools to developers around the world, which can serve large-scale and complex business scenarios in digital finance, education, culture, healthcare, e-commerce and other fields.
Of course, the criteria for whether the big model is right or wrong are not convincing if they are just "the words of the family". As a result, the China Academy of Information and Communications Technology, Shanghai Artificial Intelligence Laboratory, Wuhan University and Ant Group jointly launched the AIGC Trusted Initiative. The initiative has been participated by dozens of units, focusing on the economic, security, privacy and data governance issues that may be caused by artificial intelligence, proposing to build a global governance cooperation framework for AIGC trusted development, using secure and trusted data resources, technical frameworks, computing methods and software platforms to improve AIGC's trusted engineering capabilities, and ensure the safety, transparency and explainability of generative AI to the greatest extent.
The next step for clouds is sky calculation
One of the "Treasures of the Eight Town Halls" on the list of WAIC 2022 is another product of Ant: the trusted privacy computing hidden language technology stack. This technology stack took 6 years of research and development, and this year, Ant will open source two products in the privacy computing technology stack, namely "Hidden Language Open Source Framework 1.0" and the first domestic financial security TEE solution "HyperEnclave".
The "Cryptic Open Source Framework" is the computing engine layer in the "Cryptographic Technology Stack", and HyperEnclave is its trusted base. The technology stack is self-developed by Ant, which has accumulated more than 1,000 patents, integrating privacy computing technology products such as the hidden language open source framework, Occlum TEE open source operating system, HyperEnclave, etc., and incubating new privacy computing technologies such as trusted secret computing (TECC). The open source of the two new product functions is expected to bring leaps and bounds to the ease of use and versatility of privacy computing.
The Hidden Language Framework was first open sourced in July last year and upgraded to version 1.0 after one year, better realizing the "bag check-in" service. The new version has three advantages: solving problems such as interoperability with third-party systems, creating black-and-white box full-stack interoperability, and launching an "out-of-the-box" lightweight deployment experience package, which lowers the threshold for privacy computing applications.
TEE is one of the technical routes of privacy computing, and is considered to be the most effective technical means for data security to the cloud and privacy-preserving computing. The advantages of this open source HyperEnrich are security and compatibility. It supports mainstream CPU hardware platforms at home and abroad, provides unified TEE abstraction, and the core code is formally verified. At the same time, it meets the requirements of domestic self-research and has the landing experience of large-scale commercial scenarios.
Important progress has also been made in some technological "soft infrastructure". IEEE (Institute of Electrical and Electronics Engineers) has published the industry's first international standard IEEE2952-2023 for "Trusted Enforcement Environmental Safety". The standard was led by Ant Group to develop the technical framework for a secure computing system based on TEE technology, and CAICT, China Mobile, China Unicom, China Telecom, Ant Group and Insight Technology jointly issued the Privacy Computing Cross-Platform Interconnection Open Protocol Part 2: SS-LR. The protocol will build a more open, transparent and secure privacy computing interconnection platform to protect data security.
"The next step of the cloud will be sky computing", in the view of Wei Tao, vice president and chief technical security officer of Ant Group, privacy computing can get rid of the shackles of the cloud, realize the intercommunication between the cloud, computing, big data and data elements are the three stages that must be passed, the core is to reduce the cost of data circulation and security assurance.
Industrial digitalization is the greatest certainty opportunity
In an era of uncertainty, everyone is looking for an opportunity for certainty. "Industrial digitalization is the core area of the digital economy and the greatest certainty opportunity." Jiang Guofei, senior vice president of Ant Group, said at WAIC 2023.
At WAIC 2023 this year, AntChain announced a new architecture upgrade of the privacy collaboration platform AntChain FAIR, introducing verifiable computing technology with zero-knowledge proof (ZKP) as the core, expanding from trusted data flow to verifiable computing processes, data attributes and identities.
At present, FAIR has been applied on many data exchanges such as Hangzhou Data Exchange and Guishu Institute.
Zero-knowledge proofs are regarded in the industry as one of the most critical technologies for Web3 in the next 5 years, and it refers to the ability of the prover to convince the verifier that a certain assertion is correct without providing any useful information to the verifier. Compared with privacy computing, which emphasizes data analysis and computation on the basis of protecting data privacy, zero-knowledge proofs are characterized by "verifiability".
The metaverse cannot be built on the beach, and AI model training, identity verification, data transfer, asset transactions, on-chain and off-chain collaboration, etc. must also be based on the credibility of the whole process, which will accelerate the penetration of Web3 from the digital world to more complex industrial scenarios.
The most critical link of zero-knowledge proof is the "proof generation stage", that is, generating a "proof" for data, identity, calculation process, etc. AntChain accelerates end-to-end "proof generation" by 2.8 times the original baseline through software algorithm optimization.
As the earliest manufacturer of Web3 technology in China, AntChain is fully independent and controllable in the field of Web3 technology, and has the world's first authorized patents such as blockchain and privacy computing.
AI goes into the field
AI trainers out of the mountains
For ordinary people, the "root technology" that ants have been rooted for many years may be obscure. However, in recent years, Ant has been focusing on the real economy, especially to help small and micro groups create greater value, which is the AI that ordinary people can experience.
AI has entered the field, and in the past, the credit investigation of rural online loans had to run on both legs. Now, based on the "AI + remote sensing" technology research and development of the big system, MYbank has applied satellite remote sensing technology to the rural financial field for the first time, solving the credit problem of farmers without collateral, without guarantee, and without submitting paper materials. In the credit granting process, the big system uses artificial intelligence technologies such as deep learning to observe and analyze satellite images, identify the planting area, type and growth of crops, predict the yield and output value, and give a reasonable amount. At present, the big system has covered 31 provinces, municipalities and autonomous regions, serving more than 1.2 million growers.
Zhang Juan, an artificial intelligence trainer from Jishishan County, Gansu Province, walked out of the county, and like her, more AI trainers came to the world stage of WAIC to face with large model companies. Behind the increasingly intelligent AI, AI trainers are indispensable to teach it step by step, marking raw data such as pictures, language, text, and videos into a structured language that AI can understand.
In 2019, Ant Group, Ant Foundation and China Women's Development Foundation jointly launched the "Digital Magnolia | AI Bean Plan", and by the end of 2022, the plan has supported the establishment of 17 county digital employment centers in Shaanxi, Shanxi, Gansu, Guizhou, Ningxia and other places, helping 4,000 people to find employment locally and become "artificial intelligence trainers", of which more than 60% of employees are women. The labeling business carried out by the relevant digital employment centers has covered more than 200 specific application scenarios, with an annual data labeling volume of more than 100 million, and many centers have become the largest local labor enterprises.
From the temple technology that conquers the sky to the employment tool that goes out of the mountains, the meaning of AI is to achieve society through self-learning, and enterprises should also overcome peaks and troughs from self-adjustment to create more opportunities and futures.
Typesetting / Ji Jiaying
Image: Ant Group
Source: IT Times public account vittimes
E N D
Please add "star" to not miss us