According to Gartner analyst Avivah Litan, some of the biggest risks of generative AI concern trust and security and include hallucinations, deepfakes, data privacy, copyright issues, and cybersecurity problems.
Hallucinations refer to the errors that AI models are prone to make because, although they are advanced, they are still not human and rely on training and data to provide answers.
If you've used used an AI chatbot, then you have probably experienced these hallucinations through a misunderstanding of your prompt or a blatantly wrong answer to your question.
"These fake images, videos and voice recordings have been used to attack celebrities and politicians, to create and spread misleading information, and even to create fake accounts or take over and break into existing legitimate accounts," says Litan.
Like hallucinations, deepfakes can contribute to the massive spread of fake content, leading to the spread of misinformation, which is a serious societal problem.
3. Data privacy
Privacy is also a major concern with generative AI since user data is often stored for model training. This concern was the overarching factor that pushed Italy to ban ChatGPT, claiming OpenAI was not legally authorized to gather user data.
"Employees can easily expose sensitive and proprietary enterprise data when interacting with generative AI chatbot solutions," says Litan. "These applications may indefinitely store information captured through user inputs, and even use information to train other models -- further compromising confidentiality."
Litan says even though vendors who offer generative AI solutions typically assure customers that their models are trained to reject malicious cybersecurity requests, these suppliers don't equip end users with the ability to verify all the security measures that have been implemented.
5. Copyright issues
Copyright is a big concern because generative AI models are trained on massive amounts of internet data that is used to generate an output.
This process of training means that works that have not been explicitly shared by the original source can then be used to generate new content.
Copyright is a particularly thorny issue for AI-generated art of any form, including photos and music.
To create an image from a prompt, AI-generating tools, such as DALL-E, will refer back to the large database of photos they were trained on. The result of this process is that the final product might include aspects of an artist's work or style that are not attributed to them.
Since the exact works that generative AI models are trained on are not explicitly disclosed, it is hard to mitigate these copyright issues.
Despite the many risks associated to generative AI, Litan doesn't think that organizations should stop exploring the technology. Instead, they should create an enterprise-wide strategy that targets AI trust, risk, and security management.
"AI developers must urgently work with policymakers, including new regulatory authorities that may emerge, to establish policies and practices for generative AI oversight and risk management," says Litan.