Advantages for investigators and institutions
Consider how the above-discussed uses of GenAI and large language models in the financial crime detection process can come together to enhance the investigation process and help control compliance costs . An appropriately programmed detection system with these AI language models can even evolve to :
• Empower investigators to identify and counter malicious actors more rapidly and effectively than ever before , reducing investigative time
• Identify suspicious AML typologies present in an entity ’ s transactional behaviors , such as structuring , high-velocity payments and more , with a simple investigator ’ s request
• Present the investigator with a detailed narrative , including why an alert might have been raised and the factors contributing to the risk
• Allow the investigator to submit his or her own naturallanguage questions in a back-and-forth Q & A session for deeper investigation and making additional queries to create a comprehensive analysis
• Generate an easily understood narrative describing the risk posed by a suspicious entity , acting as a basis for SARs and other reports
What about the risks ?
With all the advantages GenAI and large language models bring to the table , it is crucial to acknowledge that there are potential risks associated with this rapidly evolving technology . Some of these potential risks include data privacy and security , ethical considerations , bias from training data , and legal as well as regulatory compliance . These are indeed valid concerns . Whether conducted privately or through government regulation , oversight will be required to mitigate these risks as this technology continues to evolve .
In fact , at their May meeting , G7 officials stated , “ We recognize the need to immediately take stock of the opportunities and challenges of [ GenAI ], which is increasingly prominent across countries and sectors .” 3 Their summit included meetings to consider issues , such as intellectual property protection , disinformation and how technology should be governed . In addition , the European Union Parliament has already voted in favor of greater transparency for GenAI models .
Proper use of GenAI can help mitigate risks in business to business applications
GenAI models themselves can be designed to help address privacy and security concerns in business settings . For example , GenAI can :
• Generate synthetic data that looks real but does not contain any personal or sensitive information , allowing the data to be used for analysis or AI model training without risking privacy
• Allow for federated learning from data without actually seeing it ; differential privacy adds random noise to the data . Both help keep individual data private
• Help make data anonymous , still keeping the overall information private but useful for analysis
• Work with encrypted data sources to keep the original information protected
To dramatically reduce risks , institutions can also :
• Carefully curate and diversify the training data
• Continually monitor the models ’ outputs to eliminate bias
• Invest in techniques to flag misleading generated content
• Invest in robust data anonymization techniques
• Develop comprehensive legal guidelines specific to these AI models
Conclusion
Almost all advancements involve some level of risk . How we address that risk is paramount to a program ’ s success . With GenAI and large language models , confronting risk requires an interdisciplinary approach , including policy and regulation , as well as public awareness .
While large language learning and GenAI models will never replace human financial crime investigators , they will most certainly make their jobs more efficient and less stressful . The sooner we can identify ways of integrating GenAI into our everyday spaces , the more efficient our lives can become .
Deleep Nair , head of Solution Engineering , NAM , SymphonyAI Sensa , Dnair @ symphonysensa . com
1
“ The state of AI adoption in AML ,” SymphonyAI and ACFCS , August 11 , 2023 , https :// www . netreveal . ai / resources / white-papers / ai-adoption-in-aml /
2
“ 2022 True Cost of Financial Crime Compliance Study — Global Summary ,” LexisNexis , https :// risk . lexisnexis . com / insights-resources / research / true-cost-of-financialcrime-compliance-study-global-report
3
“ G7 Hiroshima Leaders ’ Communiqué ,” The White House , May 20 , 2023 , https :// www . whitehouse . gov / briefing-room / statements-releases / 2023 / 05 / 20 / g7-hiroshima-leaders-communique /
ACAMS Today September – November 2023 93