The brains behind eYulchon talks to Patrick Dransfield about his algorithmic approach to corporate compliance.
Asian-mena Counsel: How did AlgoCompliance come into being and how does it differ from other compliance software solutions in the market?
Carl Im: We at Yulchon had the idea of eYulchon from as early as 2014. It was to empower our clients to get more out of our services. How? By integrating the very technologies that our clients were already using every day into our service delivery. Then how to reformulate service delivery from the client perspective came trickling in. In 2015, I was at a Volcker Rule seminar and it was quite clear that what really mattered to the 80 front-office attendees was not the law, but the impact of the law on their business. It was all about how to operationalise the abstract advice. So when the anticorruption law was implemented in Korea, we made a Q&A mobile application that covered 4 million scenarios.
We have come a long way since then. We were awarded the In-House Community Visionary Firm of the Year along with numerous other international awards. AlgoCompliance was critical in winning this award.
AlgoCompliance helps companies operationalise the legal and regulatory advice from law firms. In-house lawyers put the relevant standard operating procedure (SOP) on LegalPad, drag and drop it onto the AlgoCompliance application, and the rest happens automatically, including the PC and the mobile applications for the rest of the organisation. For the majority of everyday compliance surveillance issues, these applications help the in-house lawyers to achieve scalability across language barriers and across regions.
Once set up, AlgoCompliance walks the user through each step of the SOP in natural language and tells the user in real-time whether each response is within the company policy. In one application, AlgoCompliance keeps track of the economic benefits each client receives in real-time, for example.
Because in-house lawyers have the option to update the regulatory changes directly, clients can fine-tune the system to get the local regulations just right. Because the fine-tuning process is virtually cost-free, in-house lawyers can create a what-if version and see the impact of the change as often as needed on their own. This solves the lost-in-translation problem and wait-for-global-roll-out issues.
This year, we have introduced AlgoCompliance to the international community. This past June, I had the opportunity to compare notes with other legal engineers in London and then in October I visited the ALFA group at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) to compare our technology to similar state-of-the-art technologies. It was gratifying to realise that our approach was well-received. As a result, we are currently collaborating with other law firms to develop various applications. With the folks at MIT CSAIL, I am collaborating in hopes of publishing a paper in the field of artificial intelligence on context-sensitivity in phenotypes.
AMC: Can you explain the “wait-for-global-roll-out” issue?
CI: Currently, many multinational compliance systems are designed globally and distributed to regional offices. These systems often have a standard interface that works globally. By contrast, AlgoCompliance empowers the regional leadership to get the localisation right for each jurisdiction.
In-house lawyers can input local regulations on LegalPad in their own way, drag-and-drop it to AlgoCompliance and immediately see the final product as it appears to the rest of the organisation. Because adding local regulation is so simple, clients start by capturing the basic parts first, but then they feel that AlgoCompliance empowers them to make fewer compromises between the actual local regulation and what is feasible operationally or budget-wise.
Multilingual scenario-based audit trails make centralised oversight manageable. Firm-wide collaboration for multinationals means, for example, that Korean sales can input their activities in Korean, but any regional control officer can double-click on the scenario and can understand it in English or in French, for example.
AMC: What other products can you share with us that will delight in-house counsel and their stakeholders?
CI: As I mentioned, I just had the privilege of visiting MIT’s CSAIL to focus on incorporating some of the latest AI techniques in the areas of labour law. Working together with Yulchon’s HR team, we have created a sort of Google Map for HR Risk. Once an in-house counsel answers a survey question on where the company policy is at the moment, AlgoCompliance tells you the Policy Risk Score as well as a set of AI-recommended policy changes to get to a lower Policy Risk Score. Clients can take the recommended policy changes as a starting point to get the internal dialogue going. AI Decision-support will be big for us in 2019.
AMC: Why do you think the approach to legal software solutions you have implemented at Yulchon is different to similar attempts by other law firms?
CI: One fundamental problem with the technology and related service associated with traditional law firms is that the deliverable, say a compliance system, once delivered, is very costly and difficult to re-configure. That is because the mindset behind the products always comes from the law firm perspective, not the client, so it is only to be expected that the first version is more about the law and less about the industry insight. We at Yulchon began with a very different approach: We developed a system that a non-programmer lawyer on the client-side can re-configure. Everybody says “build things from the clients’ perspective”, but clients find it difficult to articulate a perfect solution when everything is in the abstract. However, once the lawyers, either outside counsel or in-house counsel, are empowered to make incremental changes directly to the initial version, it empowers the client to use her imagination and ask: “Oh, that’s possible?! If that is possible, then can you do this as well?” That allows her to create a more bespoke solution to her needs. Without this agile approach, many clients can tell that the deliverable is not quite useful yet (eg, scaling up and operationalising the new Markets in Financial Instruments Directive or General Data Protection Regulation), but are not able to articulate how to get there. The virtuous circle is this: empowered client can envision a way to close the gap between the abstract legal advice and the industry domain-expertise; eYulchon empowers the client.
I like to think (it has to be!)
of a cybernetic ecology
where we are free of our labors and
joined back to nature, returned to our mammal
brothers and sisters,
and all watched over by machines of loving grace
Richard Brautigan, All Watched Over by Machines of Loving Grace, 1967
AMC: Does your concept of the relationship between humanity and AI square with the optimistic vision of the Brautigan poem All Watched Over by Machines of Loving Grace?
CI: It turns out that fact is stranger than fiction in the land of AI. What we have experienced so far is neural network-related results suffer from the risk of unintended biases. In fact, unbiasing AI is emerging as the new hot topic of the 21st century. We at eYulchon try to sidestep the whole issue by focusing on techniques that are bias-free. This was the reason for my visit to MIT.
I have coined the term AI Level 0, 1 and 2. Level 0 deals with optimising man-made systems, such as training a robot to get a perfect score in Mario-Kart, or even in the game of Go. This will all be done by machines very soon. Level 1 deals with asking man-made questions about stuff that are not man-made. An example is “Can you predict whether the bee will land on this flower or not?” It should be easy, right? After all, you just take videos of millions of flowers and feed the data to Alpha Go and you’re done. Well, it turns out that the video data is meaningless because bees see flowers in ultraviolet and the data we just fed the machine was shot with a regular camera. Garbage-in, garbage-out. So we need to know if the data we are about to feed to the machine is garbage or not. This is a pretty difficult task, it turns out. There are some indications that AI Level 1 problems may be exceedingly difficult. We at eYulchon aim to master AI Level 0, and try to champion AI Level 0.5, for example.
AMC: Your career spans the Asian financial crisis; how did it affect you and how did it change the business environment in Korea?
CI: My career in fixed income can be characterised as a global legitimisation of Asian credit. As the Asian financial crisis of 1997 was beginning to catch fire, I got a call from my boss’s boss in New York to shut down most of my positions with Asian counterparties. None of the counterparties mentioned went on to default. That experience became a data point for arguing that there should be a more localised approach to Asian business selection.
The crisis was a necessary step for South Korea to become a more mature market. It turns out that the amount of knowledge one needs to be a good risk manager is more than one can experience in a lifetime. You have to respect the market because the truth is that there is no neat way to control it. Suppose you get a report from your junior traders that all of your greeks are flat. Guess what? You are still exposed to reputational risk, document mismatch risk, counterparty risk and so on. Market reform in Korea was accelerated because of that painful experience.
AMC: Who is your mentor?
CI: Richard Feynman was still teaching at Caltech when I was an undergraduate there. He taught me that being a geek can be cool. Lloyd Blankfein, Richard Witten, Pete Gerhard and Danny Yee were my benchmark during my fixed income days.
AMC: What is your hinterland?
CI: I am big on education as a vehicle for transmitting culture. Having an inspiring teacher or peer can change how you think and approach the world. Every employer is looking for young people who are going to make a difference. In order to make a difference, knowing what everyone else knows is less important than seeing what people don’t know. There is not enough emphasis on this in the current environment of quick answers.
Carl Im’s company iolex.com, internet of law, makes augmented intelligence applications. As the founder and the CEO, he has developed and patented an algorithmic approach to corporate compliance risk management. As a senior adviser at Yulchon, he has been driving Yulchon’s digital strategy, including the firm’s new website based on the principles of “content-driven marketing” and “your business your way”.
Carl combines over 20 years of sales and trading experience in fixed income, has a doctorate in physics from Stanford University and was a member of the Caltech Putnam team. His unique background as an academic, a banker and an entrepreneur has contributed to unique perspectives and approaches that have won numerous awards. His work has won the Asian Technology and Innovation Initiative of the Year, the Asia-Pacific Innovator of the Year and the In-House Community Visionary Firm of the Year.
Carl’s work at Yulchon is published under the eYulchon label. The eYulchon team has recently delivered the first Universal Compliance Engine, which empowers in-house compliance and legal officers to produce, completely on their own, mobile compliance apps for firm-wide use. Under the Universal Compliance Engine paradigm, a compliance officer can modify a policy, drag-and-drop it, download the new compliance app, see it in action and modify again, closing the gap between the policy and the practice of compliance.