promote the application of large models in digital government
2024-10-06
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
author: sun yu, xie ling
the advent of chatgpt ushered in the popularization of generative artificial intelligence. large models immediately became another focus of industrial development after the internet and received widespread attention from all parties. currently, in the construction of digital government, large model applications are used in many places to promote the digitalization and intelligence of government. the application scenarios are very wide, covering government offices, government services, smart cities, unified management through one network, emergency management, market supervision, and public opinion. monitoring, traffic management, health and other fields. objectively speaking, large models are still in the early stages of development. maintaining rational thinking in the face of technological waves is of great significance for balancing development and security and improving the level of digital government construction.
towards a big model for general purpose technology
in a theoretical sense, large models refer to machine learning models with large-scale parameters and complex computing structures. from an application perspective, large models can be divided into general large models and vertical large models. as the name suggests, general large models can be applied to various scenarios and have greater "breadth". vertical large models focus on specific fields and are secondary developed and trained in vertical fields. they are "customized" large models with good "depth" and "accuracy". the government affairs large model belongs to the category of vertical large models. it is an exclusive industry large model customized for the government affairs field based on the general large model, combining the government affairs industry-specific data and the government affairs industry application scenario characteristics.
in practice, various regions have launched large model service platforms in the field of government affairs, and the digital government field has become one of the main fronts for vertical large models. judging from the large government models that have been released so far, most of them are based on one or more general large models, targeting specific regions or specific fields, and are constructed by comprehensively considering data resources and computing resources.
the "bigness" of a large model is mainly reflected in the large number of algorithm parameters, large data and strong computing power. the "data flywheel" effect in large model iteration is deeply rooted in the hearts of the people, that is, scenarios generate applications, applications generate data, data trains algorithms, and algorithms feed back applications. this may be one of the important reasons why “scenario” has become a “hot word” in digital government construction today. however, there are uncertainties in the development of large models, which are highlighted in terms of technical routes, industrial forms, and business models. the competition of large models has led to the emergence of the law of technical scale, that is, after the parameter scale of most large models exceeds a certain threshold, the model capabilities will increase fissionally, which will lead to unpredictable application effects, which poses a challenge to supervision.
of course, if you think about large models in the long term, it is becoming a general purpose technology. on the one hand, large models are general-purpose. judging from this technical feature, large models are bound to be applied to digital government, just like internet technology. on the other hand, another important technical characteristic of general purpose technology is innovation complementarity. practice has proven that reform drive and digital empowerment are the “two wings” of high-level development of digital government. so, once a large model is applied to digital government, whether its application effect is "empowering" or "negative" depends on the adaptation of the organizational structure, institutional system and governance model, and what kind of reform demands this will bring. complexity. therefore, we need to advance the application of large models in digital government with an open and prudent attitude.
properly handle the three types of relationships
at present, although government affairs large model service platforms have flourished, the openness of the basic large model is insufficient and there is a lack of unified standards for compliance review. data scale and data quality are uneven, there are hidden dangers in data security, and computing resources are scattered. there is a problem. promoting the application of large models in digital government with an open and prudent attitude requires the proper handling of three relationships.
one is the relationship between development and security. judging from the characteristics of general-purpose technology, digital government construction will use large models. however, large models have endogenous, exogenous and derived security risks. endogenous security risks originate from the data, algorithms and computing power of the large model itself, exogenous security risks originate from attacks, and derived security risks originate from the adverse effects of the application of the large model. promoting the application of large models in digital government must coordinate development and security, and insist on equal emphasis on development and security. promote the development of large model applications with an open attitude, and ensure the security of large model applications with a prudent attitude. in practice, we cannot blindly emphasize security and ignore the development of large model applications, nor can we only emphasize the development of large model applications and ignore security. to properly handle the relationship between development and security, we need to adhere to systemic thinking, incorporate large model applications into the unified planning of digital government construction, and incorporate the security of large model applications into the integrated planning of the digital government security system.
the second is the human-machine relationship. with every iteration of technology since the industrial revolution, the relationship between humans and machines has been constantly redefined. from machine assisting people, to human-machine collaboration, to human-machine collaboration. it is undeniable that the previous rounds of technology waves mainly impacted occupations with high repetitiveness or low professional skills. however, large models have cognitive abilities, analytical abilities, reasoning abilities and creative abilities. some are considered to be in the traditional sense. highly specialized occupations will likely be affected. for example, to promote the application of large models in digital government, whether the civil servants are empowered by technology or replaced by technology, whether administrative counterparts feel digital fairness, or whether the digital divide is further exacerbated, all need to fundamentally handle the human-machine relationship. problem. on the one hand, we must welcome the application of large models with an open attitude and improve the artificial intelligence literacy of the whole people; on the other hand, we must regulate the application of large models with a prudent attitude and be wary of technological failures and loss of control.
the third is the relationship between technological innovation orientation and application scenario orientation. there is always a mutual game between promoting the application of large models in digital government, whether to choose technology innovation orientation or application scenario orientation. the former focuses on the research and development of key core technologies for general large models, superimposes influential products with technological innovation, and achieves independent controllability of key core technologies for large models. the latter focuses on the application of technology, that is, with scenario innovation as the core and the creative application of new technologies as the guide. as a basic and pilot project for building a network power and digital china, the application of large models in digital government obviously requires attention to both the technical research and development of general large models and the application of vertical large models in government affairs scenarios. on the one hand, we must promote technology research and development and industrialization development layout in the frontier areas of large models with an open attitude, strive to achieve new breakthroughs in original innovation, and achieve self-reliance and self-reliance in science and technology; on the other hand, we must choose vertical large models with a prudent attitude. the application scenarios of the model in the construction of digital government should balance development and security, handle the human-machine relationship well, select those scenarios that the people expect, are safe and controllable, and apply them first, and then promote them after forming experience and practices. (sun yu, xie ling)
source: study times
