The unreal - making synthetic, authentic

We are entering a world of synthetic realness, where AI-generated data convincingly reflects the physical world. In this world of synthetic data, images and chatbots, spoofing and fakes, we face questions: What’s real, what’s not and perhaps more importantly, when do we care?

Synthetic realness can push AI to new heights in healthcare. Synthetic data can represent patient datasets for use in research, training or other applications. Synthetic content, such as AI-generated text, video and audio, could be used to counter malicious deepfakes and misinformation in healthcare by spreading truth from trusted sources to counter damage bad actors.

Indiscernible fakes exist, so as synthetic realness progresses, we must focus on authenticity. We’ll begin to evaluate "Is this authentic?” based on four primary tenets:

• Provenance – what is its history?
• Policy – what are its restrictions?
• People – who is responsible?
• Purpose – what is it trying to do?

That said, using these technologies pushes healthcare into controversial terrain. It raises tough questions about how to use generative AI in an authentic way within the context of bad actors using these same technologies to create deepfakes and disinformation that undermine trust.

Healthcare enterprises are becoming architects of the unreal world. First, synthetic data is being used to train AI models in ways that real-world data practically cannot or should not. This realistic (yet unreal) data can be shared, maintaining the same statistical properties while protecting confidentiality and privacy. It can also be made to have increased diversity to counter bias, thus overcoming the pitfalls of real-world data.

A second use of synthetic data is to be more human-like for creation and interaction. In people’s everyday lives, chatbots and virtual assistants are increasingly common and convenient – and new technologies could make them more realistic than ever. However, while pursuing these technologies in healthcare we must remember: Just because we can use AI, doesn’t mean we should.

Currently, the biggest threat for healthcare organizations in the synthetic world are the actors who use it maliciously. Researchers have found that deep fakes can easily infiltrate medical imagery, such as MRIs, X-ray scans and CT scans. They found that deceptive attackers can easily surpass systems with outdated security to add or remove tumors on imagery. The potential results? False detection of tumors and misdiagnoses that lead to lack of treatment or unnecessary treatment.⁵

These bad actors could also come across as regular citizens sharing information via social media through synthetic content. Deepfakes could have significant negative impact on patient outcomes. If healthcare consumers are duped by synthetic content, they’ll blame the healthcare organization – or they won’t become patients in the first place. Lives are at stake as are the enterprises’ reputations.

Using generative AI in an authentic way means taking heed of provenance, policy, people and purpose. By observing these four tenets, healthcare organizations can gain confidence not only in their decisions to trust others but also in their use of generative AI such that others can trust them.