Introducing Hummingbird Labs: A Space for the Development of Compliance-Grade AI
Introduction
Today Hummingbird announces its newest initiative – Hummingbird Labs! Here you’ll find a front-row seat to our explorations and experiments with the latest AI technology. This is our space to show our work, share our perspective, and take you with us as we work on building true compliance-grade AI.
Why we’re Doing This
Without a doubt, the most exciting new technology of the past year has been Generative AI. It’s no stretch to say that it is the most exciting technical advancement in recent memory. The launch of generative AI applications such as ChatGPT, Midjourney, and Synthesia have sparked imaginations and spurred the call for new business applications. Every industry is now asking: “What could AI do for us? Could this automate or help improve my work?”
As self-professed tech optimists, we’re excited and intrigued by these new developments.
As compliance professionals, on the other hand, we take this news with more than a few grains of salt.
It’s true that the generative AI explosion presents us with a new opportunity to automate tedious work that previously had been handled by a human worker. It’s also true that we at Hummingbird see a number of potential compliance applications for artificial intelligence, and that – as providers of a tech-forward compliance platform – we are in a unique position to build them.
But truer than either of these things is the fact that when new technologies come out, excitement often outstrips applicability. Blindly applying technologies without fully considering the limitations of the technology or the nuanced needs of the application will result in functionality that either does not serve its purpose, or worse, serves it in an ineffective or dangerous way. This is especially true with AI, where the quality is so variable, the results so difficult to understand or explain, and the security model so new. And in compliance, where even small mistakes can have catastrophic impacts to individuals, businesses, and the financial institutions that steward them, it’s hard to imagine a more dangerous mix of ingredients.
Our Position on AI
As you might expect, we've seen a lot of AI-specific marketing in the financial compliance market lately. Some of it seems to be stretching the definition of AI. We’ve seen more standard forms of automation given a misleading AI label; we’ve also seen publicly available AI models like GPT-4 wrapped around a thin compliance wrapper.
These quickly constructed applications often make compelling demos under controlled scenarios, but struggle to produce results when confronted with the range and complexity of scenarios that are common in the wild. Chatbots also demo well, but they put the onus on the user to craft queries properly and so can have limited impact on actual efficiency.
And in all of these cases, there is always the question of whether or not the AI application has the requisite security measures in place and is safe to use.
We believe that compliance leaders will set a high standard for what they need from an AI-enabled platform. As a result, we wanted to take a slightly different approach – to build in the open, to be as transparent as possible, and to only launch products once we are confident that they are ready to support compliance work, with all the nuanced complexity that implies.
It may not be the flashiest or the fastest way of doing things, but we believe its the only way to build compliance-grade AI tools.
The Unique Demands of Compliance AI
Hummingbird has always endeavored to effectively bring the most advanced technologies into the compliance space. Our platform was born out of a desire to help fight financial crime by empowering companies with better, more sophisticated investigation tools.
Our experience in this area has given us a thorough understanding of the challenges that we will face as we work on building our AI-enabled compliance tools.
So that you’re aware, and can be sure that you address these challenges with any AI solutions provider, we’re sharing a breakdown of our top concerns below:
Top Challenges in the Development of Compliance-Grade AI
- Data Sensitivity
The data involved in compliance investigations is some of the most sensitive data you can imagine. Cases regularly include personally identifiable information (PII), financial account details, and transaction data. In addition, regulations heavily restrict the sharing of any information pertinent to the investigation of a potential financial crime. Any AI technology will need enterprise-grade security to prevent unintended data exposure. This is particularly challenging given how new and unique the security model around these technologies is. - Complex Requirements
Financial compliance involves intricate regulatory frameworks, often with a multitude of overlapping and sometimes contradictory rules and guidelines. Navigating this complexity requires a deep understanding of financial instruments, markets, and intricate financial operations. AI-enabled products will need to be sufficiently advanced to handle this complexity. - Rigorous Audits
During an audit, it’s critical to be able to demonstrate what happened, how it happened, and why it happened for any case worked by the compliance team. Any AI functionality will need to be rigorously logged with detailed explanations and verifiable logic. - Severe Repercussions
Even minor errors in financial compliance can lead to significant consequences, including hefty fines, legal penalties, loss of licenses, and reputational damage, which can severely impact the financial stability and viability of an institution. Relying on new technologies before fully vetting their capabilities is a risky endeavor. This is particularly true with modern AI technologies, where hallucinations can result in an AI confidently returning false results.
The Takeaway: Move Fast, Build Iteratively, Lead with Safety
We are excited about the potential that AI technologies, especially new generative AI technologies, have to impact the compliance practice area. We’re especially excited about the potential AI has to automate compliance work when coupled with the other automations products we’ve already announced. It’s because we’re so excited that we’ve chosen to invite you to take the journey with us through Hummingbird Labs.
The AI space is evolving so quickly, the development process is naturally a space where experimentation is rewarded. We've been examining all sorts of different compliance use cases, and we want you to know what we think works – and what doesn't! This is why we've created Hummingbird Labs. Not because we're waiting to develop AI-enabled tools (far from it!), but because new technologies like this demand that we hypothesize, experiment, refine, capture findings, and, only when results meet the standards required by our customers, launch into market.
We hope you’ll join us in our explorations and we hope you enjoy the journey as much as we do!
