September 26, 2019 | Written by: Kumar Bhaskaran
Categorized: Cryptography | Financial Services
Share this post:
Financial crimes continue to plague the global economy. As nefarious actors become smarter, the costs of money laundering and associated crimes has reached the trillions.
At the same time, the disparities between those who can access financial services – and those who cannot – threatens to widen.
To help solve these gaps, IBM Research is building new AI and data encryption tools to help keep data safe, cybercrime at bay, and make financial services more accessible.
Fighting Financial Crime with Fusion AI
The onus to spot financial crimes falls on banking and financial institutions, who can face enormous fines for compliance failures and failing to detect, report and pre-empt criminal activities. Unfortunately, current systems typically only detect suspicious activities that break predefined rules, such as a user sending over a certain amount of funds or sending funds with a certain degree of frequency. However, these transaction screening systems fail to connect and spot larger webs of how criminals tend to work today. To fly under the radar, financial criminals often launder funds between multiple and seemingly unrelated accounts, often across institutions and geographies.
To combat this trend and tie together suspicious webs and patterns, IBM financial services researchers have built unique AI models that fuse together criminal patterns across institutions. Similar to building a “six degrees of separation” network, these AI models can detect activity that would otherwise go uncovered. As these algorithms are trained on patterns, not data, they allow banks to share insights while avoiding the dissemination of sensitive data outside of their firewalls in the name of detecting suspicious activity – further protecting customer privacy and data.
Making the “Gold Standard” of Data Protection a Reality
The duty to safeguard data – combined with regulatory and compliance restrictions – has prevented many financial companies from readily taking advantage of AI and machine learning (ML) tools; for fear the risk of sharing data outside of their walls outweighs the rewards.
Homomorphic encryption (HE) is known as the gold standard of data encryption, as it enables computations to be carried out on encrypted data without needing to decrypt it first. This means a bank could theoretically encrypt sensitive data, send it for processing outside its firewalls to a hybrid and/or public cloud to train and perform predictions on ML models, and then receive it back without any unauthorized person seeing the actual data or the results.
IBM researchers have recently led significant breakthroughs (1) to progress the reality of homomorphic encryption, achieving a level of performance and accuracy that is adequate for machine learning tasks. This has been validated against real financial data from a large banking and financial services institution in the Americas.
To do this, our researchers applied advanced cryptographic schemes and HE to protect the privacy and confidentiality of the institution’s sensitive data, the existing predictive marketing models and newly generated models. Thus, protecting not only the data, but also the models and the predictions.
HE encrypted models and data were used to predict if a customer might need a personal loan in the near future, enabling targeted marketing. Typically, this is done behind a firewall – limiting a bank to only using ML tools and resources built or installed in-house.
As our research proves, HE can successfully be used to protect the privacy and confidentiality of data used in the creation of predictive models – theoretically freeing the bank to safely outsource sensitive data to a hybrid and/or public cloud for analysis with peace of mind.
In the near future, HE should allow banks the freedom to safely use ML tools and platforms – opening up greater innovation in customer service and offerings. Additionally, HE could allow greater information sharing within the walls of a bank itself; as current sharing of non-encrypted data can be risky and out of compliance even across different departments of the same organization.
Harnessing AI for Greater Financial Inclusion
Due to a lack of available information, it’s often incredibly difficult for banks to assess the credit worthiness of those who need it most: such as growing small businesses and those in emerging economies.
Working with our colleagues across IBM, our researchers have helped build capabilities on top of the Digital Banking on Z enterprise-scale platform to deploy predictive analytics, AI and data privacy tools to rapidly and uniquely rate the financial health of small-to-medium enterprises. By bridging this gap in our financial infrastructure, banks can fill in knowledge holes and onboard growing companies much more quickly, as well as offer personalized products fit to specific business needs.
Additionally, our teams across the world are working to bring more accessible financial infrastructure to emerging markets, such as small farmers in developing regions. By using new technologies such as IBM PAIRS Services and AI, researchers are creating new ways to determine and analyze a farm’s growth levels and crop yields – in turn providing farmers with visual images and analysis to use as a new form of collateral to obtain loans.
Building New Forecasting Tools with AI and Alternative Data
Predicting and forecasting how certain investments and risks will pan out has challenged the financial industry for decades. Despite a tremendous and growing amount of data from which companies can pull from to make decisions, there remains the obstacle of integrating this information to actually produce valuable predictions.
This has become even more difficult with the advent of non-traditional data sources which span geospatial boundaries, such as weather, social media and visual and satellite data. Building on a deep AI foundation, our researchers have successfully built AI models to accurately and rapidly capture and learn from this data, enabling companies to take new factors into account when determining the risk of an investment: such as whether or not it is likely to be impacted by fluctuating climate conditions or extreme weather.
As advances in AI and data encryption continue to take root and evolve, our teams are continuing to grow the accessibility and feasibility of these tools. If you’re attending SIBOS 2019 in London this year, we’d welcome you to have a conversation with us about the work we’re doing, and how we envision AI continuing to transform and accelerate the digital shift in the finance industry and foster new ecosystems.
- Towards Deep Neural Network Training on Encrypted Data: http://openaccess.thecvf.com/content_CVPRW_2019/papers/CV-COPS/Nandakumar_Towards_Deep_Neural_Network_Training_on_Encrypted_Data_CVPRW_2019_paper.pdf
- Towards a Homomorphic Machine Learning Big Data Pipeline for the Financial Services Sector, Oliver Masters and Hamish Hunt and Enrico Steffinlongo and Jack Crawford and Flavio Bergamaschi