ChatGPT was developed by OpenAI. Once it was launched in November 2022, it sparked heated discussions, and the number of users continued to skyrocket. At the beginning of 2023, Microsoft successfully integrated ChatGPT into its search engine Bing, and its market influence increased rapidly. At almost the same time, Google announced the launch of Bard, a similar artificial intelligence application it developed. Since then, generative artificial intelligence has set off waves of upsurge in domestic and foreign technology circles and capital markets.
Table of contents
2. The development process and main application fields of generative AI
2. Analysis of the hidden legal risks of ChatGPT
1. Data compliance and network security issues
2. Intellectual property protection issues
What is ChatGPT? What are the potential legal risks? What are the trends in law regulation in AI? How should enterprises and users prevent legal risks under the background of artificial intelligence fever sweeping the world? Will we be replaced by AI?
1. What is ChatGPT?
1. Concept and principle
ChatGPT is a chat robot based on the language model GPT-3.5. He interacts in a dialogue way, which can not only answer questions, but also admit mistakes, question incorrect premises and reject inappropriate requests. ChatGPT is part of the wave of Artificial Intelligence Generative Content ("AIGC" or "Generative AI").
The GPT (Generative Pre-trained Transformer) model is a natural language processing (NLP) model that uses a multi-layer transformer (Transformer) to predict the frequency distribution of the next monomer, by training the language pattern learned on a large text corpus to generate natural language text. According to the official website of OpenAI, the training of the ChatGPT model uses human feedback reinforcement learning (Reinforcement Learning from Human Feedback, referred to as RLHF). language is more human-like. (See Figure 1 for the training process listed on the OpenAI official website)
2. The development process and main application fields of generative AI
Generative AI originated from analytical AI. Unlike analytical AI, the knowledge learned is limited to the data itself. Generative AI can generate samples that do not exist in the data on the basis of summarizing and inductive data knowledge. Since 2010, generative AI has ushered in a stage of rapid development. In 2017, Microsoft's artificial intelligence "Xiaobing" independently created the world's first AI poetry collection; in 2018, Nvidia released StyleGAN to automatically generate pictures that are difficult to distinguish with the naked eye; The breakthrough of 3.5... Each milestone breakthrough marks the inflection point of generative AI technology and business.
The explosion of generative AI is mainly brought about by the upstream and downstream applications of the industry chain in different fields, from text interactive AI represented by ChatGPT to image generation AI represented by Stable Diffusion, from policy generation to virtual human generation, From video generation to game character generation, as shown in Figure 2, AI has had a profound impact on traditional industries in human society in many fields such as search, finance, education, medical care, commodity marketing, and entertainment. Humans can use AI to help them write business plans , learn English, study the composition of biological proteins, draw pictures... use AI to arm the mind to achieve wisdom upgrade and evolution.
2. Analysis of the hidden legal risks of ChatGPT
While people are cheering for the novel experiences brought by new technologies, and venture capital is pouring into a new world full of creativity, we need to think about the chaos that new technologies may cause and the potential risks of harming human society. Combining our experience serving generative AI clients with our observations of this new field, we believe that there is a need to focus on the legal aspects of generative AI in the areas of data and network security, intellectual property, protection of minors, and criminal law.
1. Data compliance and network security issues
(1) Compliance risks of data sources
According to ChatGPT's answer (Figure 3), its training data comes from a wide range of Internet text and social platform information. Although in the era of big data, data crawlers are a common means for enterprises to collect public data, whether personal information or even sensitive personal information will be screened for data capture; similar to generative AI capture training data, whether the crawler technology used is invasive; And whether the data captured in the early stage is only used for training and other issues should arouse our vigilance.
(2) Compliance risks of data processing
Whether business secrets, corporate data rights and personal information rights will be violated in the process of using generative AI is also worthy of attention. Generative AI replaces human work while increasing the risk of data leakage. How should generative AI companies ensure that they do not infringe on business secrets, corporate data rights, and users’ personal information and other personal rights during the interaction with users has become a question worth thinking about. . Regarding this issue, different generative AI companies have different approaches. OpenAI, as the leader of generative AI, is naturally at the forefront in the exploration of privacy protection measures: ChatGPT replied that it promised to take strict confidentiality measures for all data to protect privacy and security ( See Figure 4); and another image generation application of OpenAI, DALL-E, states in the user agreement: It is forbidden to upload recognizable human facial images, and it is forbidden to enter the names of real people in the prompt (prompt), and to add technology and manual review to prevent abuse.
(3) Network security risks
While generative AI helps humans acquire knowledge and solve problems, it may also be used to write malware codes, thereby endangering network security; in addition, generative AI itself, which stores a large amount of interactive information, may become the target of network attacks, thereby further Increased risk of data breaches. In addition, ChatGPT deeply influences human beings to acquire knowledge and make judgments, and at the same time, it will inevitably bring about the danger of manipulating human thoughts. Due to algorithmic bias and political manipulation brought by untrue information sources, the risks to information security and even national security cannot be ignored.
2. Intellectual property protection issues
(1) The issue of copyright ownership
As more and more people use ChatGPT to write papers or do research on papers, ChatGPT has been titled a co-author by some papers, and the issue of copyright ownership of works generated by generative AI has always been a hot topic of discussion—using generated Does the work created by AI belong to the AI, the company that developed the AI, or the person who entered the prompt words?
(2) Copyright infringement issues
For generative AI companies, whether they use commercial software or open source software, they may involve the risk of software copyright infringement. Not only that, due to the limitations of machine learning, works created using generative AI may be sued for infringement by the original authors of works with similar styles because they are not creative enough. Thousands of painters on ArtStation, a world-renowned visual art website, initiated a boycott and prohibited Users put their paintings into the AI painting system, thinking that letting the system learn to imitate paintings is infringing copyright. Stable Diffusion, the originator of AI painting, is not only deeply involved in intellectual property infringement lawsuits, but also attacked by many painters.
3. Protection of minors
After our investigation, we found that because the quality of AI-generated content is uneven, it will indeed pose a threat to minors who lack judgment. At present, most of the generative AI applications on the market are limited to adults, but how to confirm the user in the user agreement? Age restrictions and how to strictly prohibit the use of minors in the registration stage have become difficult problems faced by enterprises and society. Take ChatGPT as an example, according to its privacy policy, children under the age of 13 are not allowed to use ChatGPT, and ChatGPT has not taken any statement and measures at this stage on how to protect the interests of minors aged 13-18; in addition, due to the current ChatGPT can register only with a mobile phone number. How to protect minors when the mobile phone number is not verified by real name, ChatGPT currently does not take any restrictive measures.
4. Criminal risk
With the emergence of more and more multi-functional generative AI technology, there are no technical problems in the synthesis of pictures, audio and video. Criminals and terrorists use AI technology to steal faces, use ChatGPT similar emerging concepts to carry out online fraud, etc. The problem is becoming more and more prominent, and the criminal methods are becoming more and more diversified.