The AI Revolution Will be Contextualized
Increasingly, every get-together of “data experts” is turning into a discussion about Generative AI. Inevitably, this leads to my being asked whether Tranquil Data is active in this space. So, let me just say this once: emphatically yes, Tranquil Data is a Generative AI company.
Does that mean that we’re gathering data and training models to generate our underlying context graph? No. I firmly believe that when you’re providing transparency, and proving correct use & sharing of data, there needs to be determinism. What it does mean is that the next-generation of AI Infrastructure will be made up of key enabling components. Without a System of Record for data context that fosters trust and enforces purpose, we won’t be able to tackle the most valuable and critical use-cases.
[ quick aside: my colleague Shawn Flaherty just posted a deep-dive on how this is playing out both in the EU and the Enterprise. I won’t repeat his excellent examples or thorough sources, but definitely go read it. ]
There are (at least) two ways that Generative AI must evolve to unlock its full potential. The first is that models must transition from generalists to domain-specific experts. This requires access to sensitive and identifying data, and data of varying quality, both for training and recommendation. To unlock this data, the platform must be able to demonstrate how the data was used on the way in, and why data was recommended for any given query. The “why” is perhaps the most important, and challenging piece.
In a regulated space, like healthcare or banking, there is always context to why someone is requesting knowledge, and therefore where that knowledge may be applied. It may be legal, for instance, to use an internal market report to take financial action in one part of the world but not another. Health data gathered from FTC devices (like an Apple Watch) provides interesting trends, but often can’t be used to made decisions in an FDA or HIPAA-regulated context. For a model to speak with authority in any given domain, it not only needs to be trained with authoritative data, but it will need to understand the nuances of each vertical and how to decide if something that it can recommend is something that it should recommend.
The second way AI must evolve is to create personalized experiences. Following from the previous example, once a service can speak from authority in a domain like healthcare, of course individuals and doctors alike will want to draw on that service to enhance standards of care. If you go create an account on a service like ChatGPT, however, it will say in no uncertain terms not to share any personal data because it can’t be held responsible for how that data may be used or shared once on the platform.
It will be critical that AI services transparently demonstrate the ability to take on personal and identifying data for the purpose of creating individualized outcomes. Only then will they win the trust of individuals and enterprises alike. More to the point, the examples above about training domain-specific expertise are only unlocked with access to this data (this is why Pharmaceutical companies, e.g., are increasingly working to access what’s called Real World Data and Experience). That will require clear models for affirmative consent, both when users and their data onboard and when data is used down-stream.
Contextualizing users and their data, defining rich policy hierarchies that ensure valid use against stated purpose, acquiring and enforcing terms of consent, and creating external transparency is what we do at Tranquil Data. It’s the reason the company was started. The product we have built is the only System of Record that connects this knowledge to automate correct use and transparent audit at-scale. The full potential of AI will only be realized when the appropriate data platforms are built to support it, and we’re excited to be on that journey today helping companies realize the true value of their data.
