The Gist

  • Copyright concerns. One expert says generative AI tools like ChatGPT are unlikely to violate copyrights due to their unique content generation methods.
  • Data privacy. Organizations must ensure AI systems protect sensitive personal and confidential information to maintain compliance.
  • Legal consultation. Involving legal counsel early in AI projects can help organizations navigate potential legal and operational issues.

Since their recent releases to the public, generative AI tools such as ChatGPT and Dall-E have garnered widespread interest and seem to be poised to be here to stay. This development could have a significant impact on how organizations perform various tasks and generate information.

Generative AI refers to artificial intelligence systems that can create new content, whether text, images or audio — things marketers and customer experience professionals know all too well. Some organizations have already adopted these tools into workflows as a way to streamline operations, automate repetitive tasks, aid employee efforts and produce better marketing campaigns and customer experience programs.

But many individuals are asking if it’s legal to use the content that these AI chatbot tools generate. And if an organization wants to use generative AI tools for business, what does it need to know to cover its bases legally?

Growing Drumbeat Over Copyright Concerns

Much of the early public concern about AI chatbots deals with the question of copyright violation. Specifically, when the chatbot is asked to write text on a specific topic, and it reaches out to the Internet for information in order to compose that text, might it be violating copyright with something that it “grabs?”

The answer is, likely not, according to Eric Wang, vice president of artificial intelligence at software firm Turnitin. That company is the leader in the education space for producing Internet-based anti-plagiarism software, something it has done for more than 20 years. The Turnitin anti-plagiarism tool is also used by some businesses, such as media companies wanting to keep journalists honest. It doesn’t stop an individual from borrowing content from another source, it just alerts the licensed company when an example of copyright infringement is found.

But a chatbot such as ChatGPT doesn’t grab entire chapters, sections or even passages of text, Wang explained. Instead, it “builds” a narrative one word at a time, looking for the most common word it finds in use for each point in the narrative, and for the exact placement of words in the text.

In that sense, ChatGPT writes the absolute “most average” text possible for any topic, Wang said. But that is not how humans think or write. He, therefore, discounts the likelihood that an organization needs to fear copyright violation.

Related Article: How Will Generative AI Change Search?

How Safe Is PII Data From AI Chatbots?

But all is not quiet on this front. Another pressing concern about the new AI chatbots is whether they might share personal or protected information and violate data privacy or compliance regulations.

“I am seeing experimentation with generative AI, but care must be taken with personally identifiable information (PII) and information considered confidential or proprietary to the enterprise,” said Bern Elliot, a distinguished analyst at research firm Gartner. “Often, there is an internal team that is evaluating possible use cases and doing some trials. It may be with Azure OpenAI Services such as ChatGPT and GPT-3, or with open LLM models.” 

There are a range of AI tools available to developers. There is also interest by some organizations in using Mirosoft’s recently announced Copilot embedded generative AI tools.

“The use of OpenAI L.L.C.’s version of ChatGPT is more risky than using the Azure versions. So we advise clients to pursue that direction over the intermediate term as the technology matures this year,” Elliot said.

Related Article: The Future of AI: What's Next for the World's Most Disruptive Technology?

The Healthcare Industry Watching for Symptoms of Data Intrusion

One industry that is watching the release and future development of these  is healthcare.

“The biggest concern of AI solutions like these is with privacy protection rather than copyright infringement,” explains Simon Gao, executive of machine learning and artificial intelligence at Sentara Healthcare. “HIPAA has very strict rules about what type of medical health information can be accessed and shared.”

Learning Opportunities

The worry here is that one probably doesn’t know where ChatGPT is finding data on the Internet, and what data it has access to. ChatGPT obviously doesn’t care about the source — all data that resides on the Internet and isn’t safeguarded is potentially within its reach.

“The biggest worry we would have is with violating HIPAA,” Gao stressed. “We would need to train the models to make sure they don't access or use data that patients didn't give us permission to use. We have access as an internal database company, but we always have parameters around data to make sure it isn’t accessed unless we have a permissible business case to do so. And patient consent, if it's required, is a big issue.”

Organizations Should Always Use Care When Privacy and Security Are at Risk

Confirming the need for caution is Alberto Roldan, a well-known business strategist and AI expert who has extensive experience in the field and is the author of a weekly newsletter on AI strategies.

“Organizations should be cautious about using AI in areas where data privacy and security, such as handling sensitive personal or financial information,” Roldan said. “It is important to ensure that AI systems are designed and implemented in a way that protects customer data and meets regulatory requirements.”

If there is good news on this front, it is that ChatGPT is still in its infancy, and fears over potentially bad behavior may be exaggerated — at the moment, Gao said.

“AI is getting better through improvement both from a vendor perspective and more widespread use,” explained Jason James, a former healthcare chief information officer (CIO) and current technology adviser. “In order for AI to be effective, it must have not only better programming but greater data sets in order to analyze and learn. The more it gets used, the more it learns and adapts. Much like young children, wisdom comes through experience. One expected that the greater number of experiences allow for greater understanding of human behavior.”

In addition, Elliot said he believes software vendors such as OpenAI will develop workarounds for copyright and other legal concerns

“Many valuable generative AI use cases do not infringe on copyright or other legal issues. So clearly those will progress,” Elliot said.

Still, Elliot advises organizations to include legal departments in any discussions about AI investments.

“Projects should involve legal counsel early in their use-case definition process, along with those responsible for business issues, operational processes and CIO staff. In that way, the education of the parties involved will be kept apace and organizations won’t be blindsided by any issue that might arise,” Elliot said.