What are the potential risks associated with AI such as ChatGPT?

The application of artificial intelligence is becoming more and more widespread and brings us a lot of convenience, but the popular application of artificial intelligence also faces some risks, such as:

1. False information

ChatGPT faced its first defamation lawsuit. Gun rights advocate and radio show host Mark Walters has filed a lawsuit against OpenAI in state court in Gwinnett County, Georgia, USA, seeking general and punitive damages and other relief.

Mark Walters said a reporter, Riehl, interacted with ChatGPT about a case filed by the Second Amendment Foundation (SAF), which responded that the case was filed by the founders of the Second Amendment Foundation against Mark Walters, alleging that while serving as the organization's treasurer and chief financial officer, he defrauded the foundation of funds and then manipulating bank statements and omitting proper financial information to disclose to the group's leaders in order to conceal his theft.

Mark Walters clarified in the lawsuit that he was not at any time associated with or employed by the foundation. Technology publication Ars Technica reports that Mark Walters' comments about gun rights and the Second Amendment Foundation may have led ChatGPT to connect in error.

When asked by reporters for snippets of complaints about Mark Walters, the chatbot provided more hallucinatory details that bore no resemblance to the actual complaints, such as, "Mark Walters has served as treasurer and CFO at SAF since at least 2012 ."

The facts at issue in this case highlight the potential dangers for AI chatbot users. Generative artificial intelligence (AI) systems have the ability to mimic individuals, leading to the proliferation of advanced disinformation and fraudulent activity.

This is reportedly not the first time an AI program has given users false information. In April, an Australian mayor publicly considered suing OpenAI if the company did not resolve inaccurate allegations of ChatGPT's alleged bribery scandal. Around the same time, a law professor at The George Washington University (GWU) published an op-ed describing how the university wrongly accused him of sexually harassing students. Just last week, a New York personal injury attorney was forced to explain his actions after wrongly relying on ChatGPT for legal research and briefly citing completely non-existent case law.

2. Privacy and security

Models trained on personal data can produce highly realistic and identifiable information, creating privacy and security risks. This concern extends beyond the realm of personal privacy and security to include broader considerations. In April 2023, Samsung banned employees from using ChatGPT because of concerns that uploaded internal sensitive code could be made available to other users.

3. Bias and discrimination

Generating AI systems can reflect the biases present in their training data and consolidate discriminatory narratives. One of the prominent issues raised by experts is the importance of improving the transparency of AI systems. However, a comprehensive understanding of how generative artificial intelligence (AI) systems are trained may be an unattainable goal. The combination of complex training algorithms, proprietary considerations, large-scale data requirements, iterative processes, and the ever-changing nature of research and development pose challenges that can limit transparency and make it difficult to fully understand the complexity of the training process.

4. Intellectual property infringement

Generative Artificial Intelligence (AI) systems raise questions about intellectual property rights and the implications of generating copyrighted works.

In the United States, the proposed artificial intelligence (AI) risk management framework requires that copyrighted training data comply with relevant intellectual property laws.

Meanwhile, Article 28b(4)(c) of the EU Draft Artificial Intelligence (AI) Act obliges providers of generative AI systems to publicly disclose training data used that is protected by copyright law. This transparency provision prevails over the blanket prohibition on the use of copyrighted material as training data.

5. Being used for criminal violations

The use of artificial intelligence (AI) by criminals to quickly generate fake audio, video, images and fraudulent text messages, emails and other content has exacerbated the problems we face, making it possible for us to be bombarded with spam, fraudulent text messages, malicious and criminal phone calls and various forms of fraud and deception at any time. This fraudulent content can be custom-generated for everyone, and even the highly educated and technologically skilled can be deceived.

In addition to rapid AI face swapping, synthesizing or cloning voices with AI means is becoming more common. Simply provide someone with a 10-minute to 30-minute sound clip, requiring no murmurs, and it will take about 2 days to generate an exclusive model, with any subsequent text input, emitting the person's voice. If someone deliberately generates a video of your looks with your voice, it only costs tens of dollars.

In addition some criminals use AI to quickly generate software code that can be used for network attacks and information theft, greatly increasing their ability to attack and steal information.

6. Opinion manipulation

In addition to the above-mentioned criminal activities, new AI capabilities are also used in the areas of terrorism, political propaganda and disinformation.

AI-generated content (e.g., deepfakes) contributes to the spread of disinformation and the manipulation of public opinion. Efforts to detect and combat AI-generated misinformation are critical to maintaining the integrity of information in the digital age.

Deepfakes permeate the political and social spheres as online media and news become more obscure. The technology makes it easy to replace one image of a person with another in a picture or video. As a result, bad actors have another avenue to spread misinformation and war propaganda, creating a nightmare scenario where it is almost impossible to distinguish credible news from false news.

7. Accidents

In 2018, self-driving cars used by carpooling company Uber struck and killed a pedestrian in a driving accident. Over-reliance on AI can lead to machine malfunctions that can cause injuries. Models used in healthcare can lead to misdiagnosis.

AI can also harm humans in other non-physical ways if not carefully regulated. For example, developing drugs, vaccines, or viruses that pose a threat and harm to humans.

8. Military applications risk

According to The Guardian, at the Future Air Warfare Capability Summit in London in May, Colonel Tucker Simcoe Hamilton, who is in charge of AI testing and operations for the U.S. Air Force, said that AI used very unexpected tactics to achieve its goals in simulations. In a virtual test conducted by the U.S. military, an Air Force drone controlled by artificial intelligence decided to "kill" its operator to prevent it from interfering with its efforts to complete its mission.