Deep Ratna Srivastav, SVP, Head of AI and Digital Transformation at Franklin Templeton, speaks with Nazar Labunets, Product Marketing Manager at Ataccama, in a video interview about the role of generative AI, the importance of data quality and governance, AI governance, and capabilities that ensure AI accuracy.
Franklin Templeton is a global leader in asset management with more than seven decades of experience.
Generative AI acts as a UI allowing seamless engagement and cross-collaboration by breaking silos in a meaningful way, says Srivastav. For instance, he discusses taking a data set and analyzing and comparing different reports from different sources to improve operational processes. To be able to connect different sources of information, all the data had to be gathered in one place and humans had to structure it together to make decisions.
Whereas, with generative AI, humans can extract information that resides in different silos sitting in their consoles, says Srivastav. Unlike in the past, now decision-makers can connect silos at operational, tactical, and strategic levels.
When asked about the importance of data quality and governance, Srivastav reiterates that data quality is a critical component of what can go wrong with AI opportunities. He maintains that in the past, humans decided by getting data outputs, and in case the system backfired, implicit controls existed to collect information.
However, a challenge in a generative AI world, Srivastav says, is that the extracted information from one system may be getting fed into multiple AI systems allowing different decisions. While such interconnections help in breaking silos, they also magnify the impact of what can go wrong, therefore data governance is imperative now, he asserts.
There is a need for systems and advanced capabilities such as trend monitoring to flag where things can go wrong and try to control problems at the source, says Srivastav. He affirms that this applies to other areas of AI as well because a lot of these decisions do not just happen in unstructured data but in structured data as well.
However, the only benefit of structured data is that the industries have been working on data management techniques and there are matured capabilities, asserts Srivastav. The challenge is there but doubling down on the right technologies when needed comes to the rescue, but with unstructured data the challenges are significant.
Highlighting GenAI problems like hallucinations, Srivastav states that the capability set to manage risk is not mature yet and that evolution will take its time.
Referring to critical capabilities that ensure AI accuracy, he stresses the explainability side. It is a task to comprehend if the decision has changed due to a change in underlying data or because the models point to something new, he notes.
It is a challenge to be able to understand what is going on and have the right oversight, says Srivastav. Thus, the organization puts immense effort into monitoring the output, monitoring the intermediate steps, creating an explanation of the decisions, and getting transparency.
The issue of bias in data and how it impacts the output is shooting up and solid oversight mechanisms are required, says Srivastav. He asserts that technology teams must come together with subject matter experts to build the right oversight.
When asked about specific technologies to assure AI governance, Srivastav mentions outlier detection as a capability that has a significant impact and continues to grow. He implores organizations to focus on what can go wrong on the extreme edges.
Concluding, Srivastav refers to stress testing, and scenario testing before launching something. He states that creating synthetic data simulations and stress testing in all areas has become increasingly crucial.
CDO Magazine appreciates Deep Ratna Srivastav for sharing his AI insights with our global community.
