There’s a peer of mine who runs our AI hub, which serves as the foundation of our hub-and-spoke platform and modeling strategy for the company. We are inextricable partners in this journey because AI relies on the foundation of good data.

Our permissioning framework, data ownership, and approach to entitlements and access all flow through our AI complex — there’s no central set of entitlements. Instead, data is entitled at the source by its owners, and that structure propagates throughout our systems. Additionally, we share a common ethics framework, privacy structure, and control mechanisms, meaning our success is deeply connected.

Our data complex is also one of the largest users of our AI complex. We’re leveraging AI for data quality, scanning millions of legacy contracts to extract value from unstructured data, and driving our master data strategy. For example, insights from a million customer records are informed by previously untapped PDFs that we can now mine effectively.

Ultimately, there’s a seamless, hand-in-glove relationship between the CDO and the head of AI.

When we step back and look at use cases, I’ve got customer data, product data, or what have you. Some of the artifacts governing that information sit in unstructured documents. We’ve found incredible value in connecting large language models to repositories of PDFs, scanning them, and cross-checking what’s in our databases, essentially validating structured data using unstructured sources.

We’ve gone through our governance life cycle to ensure these models are approved, back-tested, and ground-truthed. That process has been critical. But more importantly, it has allowed us to challenge assumptions about our database accuracy and ensure alignment with unstructured information that may have been platformed just a few years ago. This approach has driven a better operating model.

For example, we can confirm that what we’re contracted for matches what we’re using and ensure we’re utilizing everything we’ve contracted. That works both ways — are we fulfilling obligations as expected, or are we leaving value on the table? Consider major transitions like the LIBOR shift a few years ago. Scanning a million contracts to identify risks and opportunities manually would be both prohibitive and error-prone. Training machines to analyze and interpret that information has been an incredible accelerant in driving commercial outcomes.

“Guardrails” is a wonderful word, and it’s something we’ve truly embraced. It’s so apropos of what we’re doing. There’s immense flexibility in the technology around us, a great deal of curiosity, and a very low barrier to entry. If you can type a question into a text box, you can use a large language model.

But guardrails aren’t just about technology or information security. We have to think of them in terms of literacy, as you mentioned. To use our AI platform, users must complete a training course and obtain certification. It’s not lengthy but it provides essential education on ethics, privacy, our obligations, and the efficacy of the information generated by these models. It also teaches users how much they can rely on AI-generated content, what checks and challenges they should apply, and how our operating model values factor into the responsible use of AI.

For example, can you take information from a large language model and use it in a particular way? We’ve been very prescriptive in ensuring these guardrails are upheld. We see AI as an incredible opportunity for our company and the world, but without vigilance, people may unintentionally use it in ways that don’t align with our values.

So, our approach is threefold:

• Training and certification: Users must complete a course before accessing our AI platform.

• Operational safeguards: AI must be used within specific operating models, ensuring, for instance, that information cannot be transferred without a third-party check.

• Data set considerations: We carefully determine what data sets can be input into a large language model.

Additionally, all our large language models are sandboxed, meaning no information is ever exposed to the public domain. While that approach may work for some companies, it does not align with our value system.

It’s “and” as opposed to “or.” Certain job descriptions require employees to undergo training as part of their role. If you’re going to come into contact with this technology, we are not going to be self-insured — we want people to be trained. I envision that this training will be periodically updated and mandated, much like some of our compliance training.

For other job roles, this training would be an additional tool in their skill set, but we haven’t mandated it yet. That said, we do expect that, over time, the majority of the firm will have completed the training.

So far, we’ve seen a high double-digit percentage of the firm trained, and we anticipate that trend will continue. Most employees choose to take the training, and we’ve been rolling out our platform in stages. This is truly a firm-wide effort, and we’re passionate about the opportunities ahead while remaining vigilant in ensuring the proper guardrails are in place.

Without commenting on our board specifically, I think the culture of the firm has really embraced, in terms of, you know, the pillars and principles of how we run the company, which are formally written down. We think about being more for clients, running our company better, and powering our culture.

At whatever level of management, right up to the CEO, those pillars really flow through the potential we see with our ambitions. We just see the potential to use this technology to run the company better. We think there are manual processes that we could be using these models to run for us or assist us with, freeing our people up to work on more complex tasks. We think we can use these models to provide more insights to our clients. We can be more for our clients using these models.

And in other places, we find that they are great agents in terms of educating our people, synthesizing the knowledge we have inside the firm, and making that knowledge available to people in various job families. So, from stem to stern, the transformation of our company or the way we run our company is deeply impacted by the possibilities that are coming to fruition by using AI.