Yuanwang think tank foresees the future Frontier observation of network information
Source: Intelligence Analysts
Since the launch of ChatGPT (Chat Generative Pre-trained Transformer) in November last year, OpenAI has reached 200 million active users worldwide in just 2 months, with an average of 13 million visitors using ChatGPT every day, which is a miracle in the history of user growth in the field of artificial intelligence.
With the proliferation of users and training data, the application scenarios of ChatGPT have gradually expanded, writing homework, writing papers, writing code, writing PPT, translation, proofreading, and even wearing the skin of Internet celebrity to become a virtual companion.
However, as the artificial intelligence army led by ChatGPT continues to infiltrate all corners of society, the corresponding legal issues and risks need urgent attention.
This article focuses on the current relevant laws and regulations, and discusses the legal risks that may be involved in the use of ChatGPT.
Risk 1 – Fabricated and inaccurate answers
One of the most common problems with ChatGPT and other language modeling tools is the tendency to provide misinformation even though it seems reasonable on the surface. ChatGPT is also prone to "hallucinations," including fabricating wrong answers and non-existent legal or scientific citations.
Reasons why ChatGPT tends to provide misinformation:
1. Limitations of language models: Current language models are trained based on statistical algorithms, so what it learns is limited to a given corpus. If there is incorrect information in the corpus, the language model repeats the errors.
2. Lack of practical experience: The language model has no actual experience, so it does not know the veracity of certain information. For example, a language model might answer a question by presenting an inaccurate news story as fact.
3. Lack of human intervention: In the training process of language models, the role of human intervention is very important. Without sufficient human intervention, language models can be disturbed by misinformation.
Second, the reasons why ChatGPT is easy to "hallucinate":
1. Lack of contextual understanding: ChatGPT's performance depends not only on its language model, but also on its ability to understand context. If ChatGPT doesn't understand the context correctly, it can interpret the error message as real content.
2. Self-confidence in data generation: ChatGPT's understanding of data-driven models is based on a large amount of data, so ChatGPT may confidently answer the wrong information without sufficient data support.
3. Misgrasp of facts: ChatGPT's misinformation may be due to it not grasping the facts correctly. This may be because it does not have enough knowledge or does not have enough experience to understand the facts.
Risk 2 – Data privacy and confidentiality
Legal and compliance leaders should be aware that if chat history is not disabled, any information entered into ChatGPT may become part of its training dataset. If sensitive, proprietary, or confidential information is used for prompts, it may be included in the response to users outside the enterprise.
In daily chat communication, more or less sensitive and proprietary or confidential information will be involved, such as private data, trade secrets, etc. If this information is not legally encrypted or protected, and chat history is not disabled, it is possible that they could be acquired by natural language processing tools like ChatGPT and become part of their training dataset.
Risk 3 – Model and output bias
OpenAI is an artificial intelligence research organization established to create robots with human intelligence. ChatGPT is a chatbot developed by OpenAI that can interact with users in natural language. However, there are also issues of bias and discrimination in such chatbots.
The root of these problems is the training data of machine learning algorithms. ChatGPT is trained on large amounts of chat data by machine learning algorithms that contain people's biases and discrimination. For example, on some social media, people may make discriminatory remarks that become part of ChatGPT's training data.
Risk 4 – Intellectual property and copyright risks
With the emergence and popularization of artificial intelligence technologies such as ChatGPT, more attention needs to be paid to the protection of intellectual property rights.
Training on ChatGPT requires a large amount of data, which may contain a lot of copyrighted material that could infringe copyright or intellectual property rights if used to generate ChatGPT output.
Therefore, legal and compliance leaders need to pay close attention to changes in copyright laws applicable to ChatGPT output, as well as relevant laws and regulations applicable to intellectual property protection of AI technologies.
To ensure that ChatGPT's output does not infringe copyright or intellectual property rights, users need to double-check any output they generate, especially if they may contain content that is considered copyrighted or intellectual property protected.
At the same time, chatbot companies should also strengthen their understanding of laws and regulations related to intellectual property protection, and comply with corresponding laws and regulations when designing and using artificial intelligence technologies such as ChatGPT to avoid infringement.
Risk 5 – Risk of online fraud
Malicious actors have begun using ChatGPT to generate false information at scale, such as fake reviews. Not only that, but applications that use language models, including ChatGPT, are also vulnerable to prompt injection attack techniques.
This technique uses malicious prompts to trick models into performing tasks they shouldn't, such as writing malware code or developing phishing sites that resemble well-known websites. We must strengthen our prevention of such attacks in order to protect the safety and interests of the vast number of users.
Risk 6 – Consumer protection risks
ChatGPT is a chatbot based on artificial intelligence technology that is able to provide effective customer support and services to businesses.
However, many businesses use ChatGPT without disclosing their usage to consumers, which may cause consumer dissatisfaction and concern.
If the company perceives that this situation is not corrected in time, it may lead to a decline in customer trust in the enterprise, which will negatively affect the company's brand image and word of mouth.
In addition, if a business fails to disclose the use of ChatGPT, then it may be at risk of violating the law. For example, under data privacy regulations, businesses must transparently inform consumers how they collect, store, and use users' personal data.
If a company fails to disclose the use of ChatGPT, then it could be accused of violating data privacy regulations, which will have serious legal consequences for the business.