Big Tobacco May Be Influencing AI-Generated Content Online

Summary:

• Big Tobacco has been funding and publishing misleading information for years. Since the industry links aren’t always obvious, large language models (LLMs) may present industry-linked content without nuance or full context.
• New AI tools are being built, including an LLM specifically related to the tobacco industry which can recommend new flavors and products to users.
• Tobacco control advocates and public health researchers are also using AI tools. Canary monitors social media for instances of tobacco marketing, and the University of Bath is testing a tool that assesses tobacco industry claims for accuracy.

You don’t need to look far to find evidence that the tobacco industry tries to buy the public’s trust and influence public opinion online. These efforts can be seen in familiar corners of the internet, such as on news sites, research platforms and social media.

But Big Tobacco appears to be doing even more below the surface to promote its business and products. Research suggests AI tools are being used by Big Tobacco to help spread misinformation online, promote profit-driven industry narratives and possibly expand its product offerings.

To respond to these threats, the public health sector is also embracing AI to monitor and report on industry activity. While AI continues to develop at a rapid pace and the landscape of its effects continues to evolve, here are some of the key findings researchers have captured so far.

Big Tobacco’s tactics for influencing AI-generated content

Ask an AI-enabled search tool or chatbot a question, and it will sweep the web and give you an answer based on what others have already said about the topic—whether true or not. Large companies are seizing this opportunity to influence what AI says about them by flooding all corners of the internet with specific, brand-aligned content about their products or services.

The tobacco industry has many decades’ worth of misleading content already online, ready to influence or train these AI-enabled tools.

Take, for example, industry-linked science. This is research either conducted or paid for by the industry. It often serves to mislead people about the harms of its products and argue against tobacco control regulations, and its links to the industry are not always disclosed or easily detectable. Not only can AI-enabled search tools pick up this industry-aligned content as a search result, but the large language models underlying them that compose AI-generated responses could have been trained on industry-funded science, corporate communications, advertorials and other industry-shaped, biased material that has been online for years.

This content, even when technically accurate, may still be misleading through the industry’s selective emphasis on certain points of its narrative, the omission of harms or extensive “reduced-risk” framing.

To this end, the industry may also be ramping up its presence on Reddit, which AI models are mining for “authentic” and “trustworthy” content. Though it’s difficult to prove an industry link, researchers have found content on Reddit promoting snus and nicotine pouches. Posts related to nicotine pouches increased from 2019 to 2021, and more than half of the posts portrayed nicotine pouches in a positive light.

Researchers have also cautioned that the industry may be using bots or coordinated misinformation campaigns to interfere with AI-enabled tools used for tobacco control. Such content could influence social media monitoring and sentiment analysis programs, as well as public consultation forums for tobacco control legislation, distorting public opinion metrics. Research, policy and consumer decisions all risk being influenced by industry misinformation, serving the industry’s interests over those of health.

These efforts may be only the beginning—there are undoubtedly other ways the industry is employing AI to influence opinion and policy and promote its products. And they could soon be dwarfed by larger, tobacco-industry-specific AI tools. Tobacco Titan, an AI-powered large language model reportedly in development by Canada-based GenAI, promises to give the tobacco industry and consumers a “wide range of AI-powered intelligence and information” using proprietary and public datasets. This is reported to include information about products, regulations, “health and safety insights”—and even recommendations for new flavors, brands or accessories.

Using AI to protect public health and hold the industry accountable

In the right hands, AI tools can help protect public health. Research suggests AI chatbots, if configured appropriately, can provide effective, personalized smoking cessation interventions. And programs such as Canary, an AI-assisted social media surveillance tool used to document tobacco marketing online, are already being used to monitor and report breaches in marketing regulations.

Another promising tool shows how industry misinformation can be monitored in real-time online. Researchers at the Tobacco Control Research Group at the University of Bath have tested a beta AI tool that assesses tobacco-related claims online for accuracy. After cross-checking the program’s assessment against that of subject matter experts, researchers found the tool to be highly accurate and, importantly, much faster. More testing is needed, but tools like this hold promise in helping advocates track industry misinformation and hold the industry accountable.

As tobacco control regulations push the industry and its brands further out of physical spaces in many parts of the world (via tobacco advertising, promotion and sponsorship bans, for example), the industry has become more present online, where regulations haven’t caught up and where enforcement can be challenging.

If left unchecked, the industry will continue spreading its influence and normalizing its presence online—reaching millions of young people in the process. Repeated exposure to this content online risks sustaining the smoking epidemic and creating new epidemics.

Governments can take action to protect young people online. Researchers have put forth AI governance recommendations to ensure that AI is used for tobacco control in a way that is as transparent, ethical and equitable as possible. Governments should also follow the World Health Organization Framework Convention on Tobacco Control’s Article 13 Guidelines, which include ways to prevent online tobacco marketing and cross-border online advertising.

With almost 75% of the world’s population using the internet, governments must bring the same resolve to fight tobacco industry tactics in the digital world as they do in the physical world.