New Weapon for Cognitive Domain Warfare: ChatGPT Analysis and Strategies

On November 30, 2022, the American artificial intelligence company OpenAI officially released the large natural language interaction program chatbot ChatGPT (Chatbot Generative Pre-training Transformer). Subsequently, GPT-3.5 was launched and quickly iterated to GPT-4. This change in the entire process of information production and circulation not only marks the beginning of the use of artificial intelligence in real-world communication issues but also exacerbates systemic biases, intentional fraud, political rumors, and identity forgery that are deeply rooted in the international system and social life. This powerful transformation will also have a huge impact on cognitive domain warfare, making its forms more diverse, methods more covert, and prevention more difficult, while its impact on social life, political ecology, and overall social ideology will be even more profound. It is necessary to analyze its application scenarios in cognitive domain warfare, grasp cognitive laws, reshape cognitive content, and optimize cognitive processes.

Main Characteristics of ChatGPT from the Perspective of Cognitive Domain Warfare

ChatGPT is optimized based on the GPT (Generative Pre-trained Transformer) model, specifically improved for dialogue scenarios using a “human-in-the-loop” approach. GPT is a large language model and currently the mainstream approach for machine learning on natural language, capable of generating various types of text. It can automatically generate text dialogues similar to human language based on context. The main idea of the GPT model is to allow AI to learn text chaining from vast amounts of general data, mastering the ability to generate subsequent text based on previous content. The GPT model has now developed to its fourth generation, a result of targeted optimization for dialogue scenarios. Compared to previous fixed dialogue AI, ChatGPT has its own thinking logic, allowing the system to provide multiple answers to the same question without “human-in-the-loop” supervision, with these answers existing in a probabilistic distribution influenced by system preferences.

GPT Learns “Text Chaining” from Massive Data

As a new generation of artificial intelligence, ChatGPT uses human evaluation of system learning results when introducing big data for autonomous learning. In a sense, this process is about giving the system a “soul” and completing the finishing touches. However, due to inherent political stances or biases of humans, the selection results during the evaluation of learning outcomes inevitably carry the supervisor’s personal labels, bringing a “preset value system” into the system, making it a “new weapon” in cognitive domain warfare.

First, the Indelible Value Base.The value base of ChatGPT is rooted in the value orientation of the personnel involved in its design and development, namely the “Silicon Valley values,” which belong to the moderate left of the United States. This underlying logic examination often makes ChatGPT act as a “guardian” of the American spirit; in political topics involving the U.S., ChatGPT often adopts a neutral or evasive stance, while showing a relatively positive attitude towards discussions on favorable topics about the U.S. Moreover, on controversial topics such as minority issues, gun control, abortion, and environmental issues, ChatGPT also exhibits a relatively conservative attitude and is overall more inclined towards moderate leftism.

After ChatGPT was launched on the internet, netizens from various countries began interacting with it, gradually discovering that when faced with questions in different languages, the same question often receives different answers, and this change varies with social dynamics. This “tailoring” behavior indicates that the intelligent system is still in a growth and learning stage, and the human interventions it receives are far more complex than those indicated in the training model. This real-time and powerful human intervention pushes ChatGPT’s preset value system toward deeper, more obscure levels.

New Weapon for Cognitive Domain Warfare: ChatGPT Analysis and Strategies

The Evolution of GPT, T5, and BERT

Second, Open and Adaptable Data Training.ChatGPT is essentially the result of targeted optimization based on the GPT model for chat environments. When the GPT model learns from data from different sources, it learns different outcomes based on the “meaning distribution” of the data itself. This means that the already proven relatively mature GPT model can be cultivated into a specific AI applicable to different fields under varying data sources. This adaptability greatly enhances the development prospects of the GPT system compared to previous natural language learning systems and provides possibilities for the growth of the GPT family, significantly broadening its market prospects and application scenarios.

With the training of data with specific biases, the given value system will evolve into the value system of the tool itself, allowing ChatGPT to output as a “virtual person” in cyberspace, with its expertise in natural language learning becoming the main path and support for this output. Intelligent systems represented by ChatGPT will expand the existing cyberspace and become an important component of the metaverse. In the metaverse, ChatGPT can give rise to a “virtual person” that does not have a real-world counterpart, not merely existing as “part of the system,” but rather becoming a spokesperson for a certain value system based on its preset values, exerting a significant influence on the real world through virtual networks.

Third, Highly Realistic Autonomous Output.With ChatGPT’s launch on the internet, people have found that it has crossed the human-machine gap that previous tools faced in understanding intent, as the natural language learning process grants it the ability to surpass commands. When applied to cognitive domain warfare, it could replace the current “internet water army” engaged in traffic attacks. ChatGPT’s powerful data generation capabilities and strong autonomous awareness will shorten response times in cognitive attacks, make actions more flexible, and feedback faster, while further reducing operational costs and moving toward unmanned, intelligent warfare.

At the same time, the enhancement of deep forgery technology brought by ChatGPT makes offensive actions in cognitive domain warfare more covert. In existing offensive operations, adversaries often rely on hot events, gathering internet water armies to create a public opinion offensive, leading to cognitive confusion among some citizens. However, ChatGPT can operate independently of hot events, replacing the gathering of internet water armies, integrating cognitive domain warfare operations more smoothly into daily life. In the future, the heat of social events that cognitive domain warfare relies on will further decrease, presenting a pattern of multi-point flowering and multi-line strikes.

Constructing Application Scenarios for ChatGPT in Cognitive Domain Warfare

For most of human history, the latest scientific and technological developments have often been first applied in the military field, and the weaponization of ChatGPT is inevitable, with extensive potential application scenarios in cognitive domain warfare.

New Weapon for Cognitive Domain Warfare: ChatGPT Analysis and Strategies

Intelligent systems represented by ChatGPT will become

An Important Component of the Metaverse

Utilizing the Anchoring Effect to Achieve Cognitive Pre-setting.The anchoring effect refers to the phenomenon where people’s decision-making is often influenced by the first piece of information they receive, which can fix their thinking like an anchor submerged in the sea, leading to a biased perception. ChatGPT will first replace the position of internet water armies in cognitive domain warfare. Its robust system architecture and powerful data processing capabilities will enable a more balanced temporal and spatial distribution of cognitive domain offensive operations, and this state of distribution will yield a total offensive intensity far exceeding previous efforts. The direct impact of this change is that any slightly influential social event in daily life could become a battlefield for cognitive domain warfare. ChatGPT’s preset values no longer need specific social events for expression but will permeate all corners of society, continuously outputting negative information through distortion, rumors, and rhythm manipulation from trivial matters in life.

ChatGPT’s powerful information production capacity will further break through the limits of information productivity, exacerbating individual “information overload” and impacting the composition of social information. From a micro perspective, the intuitive feeling is that negative energy and hostility permeate all aspects of society, while a large amount of information with similar preset values will flood in whenever any social practice occurs. Moreover, ChatGPT’s “persistence” and “subtle influence” will make this process smoother, embedding negative energy information in seemingly normal complaints, insinuations, and “reasonable interests,” quietly branding people’s subconscious and altering their cognitive tendencies.

Creating Information Bubbles, Forming Cognitive Dependency.Information bubbles are an inevitable social problem in the information age, where individuals or groups narrow their information sources and polarize their views due to internal or external factors during information consumption. In cognitive domain warfare, through big data preference analysis on individuals, personalized user profiles can be created, providing possibilities for precise cognitive attacks. ChatGPT’s powerful data processing capabilities and high autonomy enable it to perform preference analysis and subsequent information production and delivery.

The intelligent system of ChatGPT can integrate preference capture and information delivery, possessing autonomous, continuous, and adaptive cognitive warfare capabilities. This will significantly improve the efficiency of establishing information bubbles. In traditional cognitive domain warfare, establishing information bubbles for individuals or groups requires long-term user tracking, preference analysis, and information delivery; it also necessitates leveraging hot events or creating them to achieve ideal outcomes, which consumes considerable resources, and the attack process is easily identifiable, interfered with, and disrupted. However, ChatGPT’s smooth process design and embedding of preset values into any slightly influential social event eliminate the need for hot events, normalizing the establishment of information bubbles; ChatGPT’s high autonomy and intelligence can precisely categorize affected individuals, intensifying information delivery, prompting the impressions implanted in the subconscious to gradually dominate personal thoughts, greatly enhancing the efficiency of establishing information bubbles. Moreover, the decentralized information delivery model based on social events is also harder to detect and regulate, further reducing the cost and response time of cognitive domain warfare offensives.

Strengthening Community Effects, Shaping Group Polarization.The community effect refers to the phenomenon where when a portion of the population shares similar values, they tend to form tighter connections compared to everyday interpersonal relationships. Given a relatively stable total population, as more people are excluded from mainstream values, the influence of mainstream values diminishes. Once the population adhering to mainstream values becomes insufficient to constitute a majority in society, mainstream values will lose their dominant status. If a new mainstream value cannot be established or fails to align with national interests, this will lead to a loss of social cohesion and centripetal force, meaning that the splitting behavior of small groups influenced by community effects will ultimately result in social atomization, leading to group polarization in society.

Behaviors of Small Groups Influenced by Community Effects Will Ultimately Lead to

Social Atomization

In the process of cognitive domain warfare, cognitive pre-setting aims to continuously create specific groups, where individuals with “group entry” potential will gradually form a cognitive dependency on this specific value information through the continuous refinement and filtering of information bubbles woven by relevant information. Ultimately, these specially cultivated individuals will, with or without external push, form small circles to seek warmth in groups, growing under constant nurturing, and eventually challenging mainstream values or engaging in disputes with other opposing small circles, creating chaos in societal thought unity. ChatGPT’s precise information delivery capabilities and crowd categorization abilities will effectively foster the formation of various small groups in this process. Simultaneously, by continuously delivering corresponding information to specific individuals in information bubbles, a batch of individuals with polarized thoughts will be created. Once the population base of such groups reaches a critical threshold, they will be organized into small groups via the internet and continuously provided with related information, allowing these groups to purify and grow steadily. These numerous small groups nurtured by ChatGPT will objectively weaken the influence of mainstream values, and the emergence of this subtle and complex cognitive offensive approach is based on the further enhancement of computational power and ChatGPT’s intelligent and efficient algorithm architecture.

Strategic Reflections on Responding to ChatGPT

Establish a Sound Network Monitoring System to Real-Time Perceive Public Opinion Trends.The capability to perceive and monitor online public opinion is a prerequisite for seizing the initiative in cognitive domain warfare, and an efficient public opinion perception network is based on advanced algorithms with a certain degree of autonomous analytical ability. It is necessary to further improve the network monitoring system to have the capability to perceive and trace online public opinion trends and possess the ability to identify AI-generated content based on aspects such as content repetition, logic, and grammar semantics. This will enable accurate identification of false information and manipulative statements released by AI-controlled accounts, allowing for a rapid response to “send, identify, ban, trace, and capture” the sources of false messages, blocking the channels through which ChatGPT disseminates false information and reducing its ability to participate in cognitive domain warfare. Strengthening the construction of professional talent teams will form a rapid response, sensitive, reliable, and technically proficient cognitive domain warfare talent pool, occupying an advantage in the conventional power comparison of cognitive confrontation in the online environment.

Firmly Defend the Public Opinion Position and Deepen Educational Outreach Effects.Recognizing the sovereign characteristics of cyberspace, all sectors of society should jointly participate in the sovereign construction of cyberspace as if it were “digital territory.” Maintain the mainstream position of mainstream values represented by the core socialist values, deepen the publicity and education of mainstream values, and firmly hold the definition and dominance of social public opinion. Enrich and expand the working forms and means of propaganda departments, discover good deeds that align with mainstream values, establish models, promote righteousness, and proactively seize the definition and leading power of social public opinion. Use methods that are more appealing to the masses to deepen the effects of publicity and education, achieving an organic combination of traditional Chinese culture, the theoretical system of socialism with Chinese characteristics, and social practice in national construction, accelerating the realization of four self-confidences, rectifying the source, and fundamentally eliminating all possibilities for ChatGPT to distort the truth.

Persist in Technological Research and Development, Enrich and Expand Application Means.Achieve a government-led, enterprise-centered construction of a collaborative innovation mechanism integrating government, industry, academia, military, research, and application, promoting the continuous advancement of cutting-edge technologies and large-scale application of applied technologies, effectively transforming platform construction, infrastructure investment, and technological progress into the counterattack power against emerging challenges in cyberspace. In platform construction, draw on advanced experiences from civilian internet; in infrastructure investment, ensure normal operations of regulatory systems led by the government with multiple parties involved; in technological advancement, prioritize regulatory and military needs, with universities and research institutions fully participating, quickly breaking through technological bottlenecks in cognitive domain warfare situational awareness, battlefield twins, user profiling, etc., to gain the initiative in countering ChatGPT’s use in online cognitive confrontation.

Copyright Statement: This article was published in issue 6 of 2023 of “Military Digest,”Authors:Chen Changxiao, Li Hao, Wang Zihan, Jiang Wenbo,If reprinting is needed, please be sure to indicate “transferred from ‘Military Digest.'”