Artificial Intelligence (AI) and Machine Learning (ML) have become a big part of our everyday lives in a very short period of time. From virtual assistants like Amazon’s Alexa and Google Home to the arrival of self-driving cars, we are beginning to see the positive net effects of these innovations on our productivity and happiness.
However, for all the positive impacts that these innovations can bring us, we’re also becoming more cognizant of the negative impacts of AI. While most popular discourse centers on data privacy concerns, one of the more serious negative impacts is the bias that lives within the data that feeds these systems and uses it to become more ‘intelligent.’ Take for example, the growing body of evidence that Facebook’s newsfeed influenced the outcome of the 2016 election, which has subsequently affected global politics and international economic fortunes.
If you look beyond these broad challenges and look more closely at how organizations could start to use AI in their everyday decision making, the picture becomes even darker. From decisions related to eligibility for healthcare, to determinations about criminal intent, guilt, or innocence, such biases can have serious consequences for all of us. And, what makes the arrival of AI into our daily lives even more challenging is that AI decision making is difficult to audit. AI has brought us the era of “black-box” decision making that is casting aside traditional decision-making processes, the kinds that have been made by consensus in meetings between people or driven by human-programmed business rules.
Even with these cautions it’s highly unlikely that we’ll walk away from AI. So, the question that is typically posed is when do we use AI and when do we not? But I’m of the opinion that this question creates a false dichotomy; let me explain my point of view.
Let’s take a regulated industry – Financial Services – and one such process – entity resolution in financial crimes. Over the years Pitney Bowes Software has helped large banking clients successfully resolve and link customer information across systems of records, insights and engagement using deterministic and probabilistic techniques with full transparency around how those decisions were made. These decisions were primarily driven by subject matter experts that set and validated the foundation of those rules, and more, often than not, seem to be more art than science.
To help right the balance between art and science our practice has evolved to link customer information across obvious and non-obvious dimensions to paint a holistic picture of a given entity and the network of relationships, in context of a process, a transaction, or an interaction. While many fintech startups have embraced ML-based ‘black box’ techniques to reduce the amount of work needed to invest in the development of rules to simply and automatically resolve entities, we’ve taken a more measured approach.
There’s no denying the power of ML techniques to the process of entity resolution, but we also realize the challenges of taking a black box approach to this problem, especially in a regulated industry like Financial Services. And especially, when you are trying to justify to a regulator why you think someone might be committing money laundering.
That’s why I feel that the best approach is one that does not automate the process of resolving entities within a black box, but one that can use the intelligent techniques to furnish the subject matter expert with the best set of rules to apply to resolve those entities. In doing so, full transparency over the rules that are used to make a decision can be maintained. The subject matter expert can then accept, reject, or use this information as a basis to come up with their own rules using the recommendations of the system. This will drastically accelerate the process and use the intelligence of the system to account for outliers and bias reduction across a broader dataset, rather than working narrowly within their own human bounds and interpretation of a limited dataset.
Ready to learn more about how to improve the accuracy and efficiency of your institutions financial? Here are some great resources.