PRACTICAL SOLUTIONS
For organizations to implement trustworthy AI into their corporate strategy , necessary policy adjustments are needed to ensure success . Assessing and adapting policies can ensure that appropriate checks and balances are in place to monitor and regulate AI systems used within the organization . This step does not require a complete overhaul of existing policies ; instead , a fine-tuning of the current policies to align with the principles and strategy for trustworthy AI . For example , an organization can embrace transparency as part of its trustworthy AI strategy , which will require clear , explainable communication about the model . Complex model components are often difficult to understand for non-data scientists . Embracing transparency and implementing tools for transparency can help people understand the decision-making process . “ Organizations should communicate models ’ intended scope and outcomes , identify potential biases and foster user trust .” 10
To establish AI governance , strategy and enforcement , consider these three essential steps : 11
• Establish principles for trustworthy AI that align with your organizational “ why .”
• Develop a trustworthy AI strategy that complements your overall business strategy .
• Adapt and enforce governance and accountability policies to accommodate the new strategy .
After the oversight framework has been created , leadership should curate operational procedures that align with this framework . These procedures should be practical , actionable and easy to implement . It is essential to embed the procedures into the deployment and operations of AI . One procedure an organization could implement using the example of transparency is around data lineage . Data lineage tracks data as it moves through the data ecosystem and allows for visibility into how the data is stored , processed and analyzed . Data lineage also helps organizations understand the suitability of their AI system for different uses . 12 Organizations should implement procedures to understand which data assets are relevant and appropriate for modeling activities by keeping robust documentation of the inputs , transformations and outputs . Another important part of transparency is model explainability . Model explainability helps users translate the results of AI models with techniques like partial dependence plots , individual conditional expectation plots , local interpretable model-agnostic explanations and HyperSHAP . 13 Establishing procedures around model explainability utilizing these ( and other ) techniques helps users understand how a decision was made by the AI system .
As the organization ’ s AI journey evolves , multiple tools , software , vendors and other factors will come into play . Organizations should assess their AI vendors for consistency with their own established principles . Consistency across its entire ecosystem will help the organization maintain a cohesive approach to trustworthy AI .
AI-powered decisions and systems are increasingly being embedded into AML processes . Organizations need to be aware of the potential negative consequences of AI-driven decisions . The humans behind the algorithm should be considered throughout the design , development and deployment of an AI system . As the algorithms utilized become more advanced , the MRM framework needs to become more comprehensive than ever before .
Mason Wheeless , systems engineer , SAS Fraud and Security Intelligence Practice , Mason . Wheeless @ sas . com
Kristi Boyd , Trustworthy Artificial Intelligence specialist , SAS Data Ethics Practice , Kristi . Boyd @ sas . com
1
“ How AI and Machine Learning Are Redefining Anti-Money Laundering ,” SAS , https :// www . sas . com / content / dam / SAS / documents / marketing-whitepapersebooks / sas-whitepapers / en / ai-machine-learning-redefining-aml-110762 . pdf
2
“ 5 Ways to Harness AI to Accelerate Your KYC Process ,” FileInvite , March 8 , 2023 , https :// www . fileinvite . com / blog / 5-ways-to-harness-ai-to-accelerateyour-kyc-process
3
“ 2023 Edelman Trust Barometer Global Report ,” Edelman Trust Barometer , https :// www . edelman . com / sites / g / files / aatuss191 / files / 2023-03 / 2023 % 20 Edelman % 20Trust % 20Barometer % 20Global % 20Report % 20FINAL . pdf
4
“ Why Addressing Ethical Questions in AI Will Benefit Organisations ,” Capgemini , https :// www . capgemini . com / gb-en / insights / research-library / why-addressing-ethical-questions-in-ai-will-benefit-organisations /
5
Ibid .
6
Ibid .
7
“ Europe fit for the Digital Age : Commission proposes new rules and actions for excellence and trust in Artificial Intelligence ,” European Commission , April 21 , 2021 , https :// ec . europa . eu / commission / presscorner / detail / en / ip _ 21 _ 1682
8
Ibid .
9
“ Oral Statement of Kevin Greenfield , Deputy Comptroller for Operational Risk Policy before the Task Force on Artificial Intelligence , Committee on Financial Services , U . S . House of Representatives ,” U . S . Office of the Comptroller , May 13 , 2022 , https :// www . occ . gov / news-issuances / congressional-testimony / 2022 / ct-occ-2022-52-oral . pdf
10
Vrushali Sawant , “ The ethics of responsible innovation : Why transparency is key ,” SAS , March 30 , 2023 , https :// blogs . sas . com / content / sascom / 2023 / 03 / 30 / the-ethics-of-responsible-innovation-why-transparency-is-key /
11
“ A Comprehensive Approach to Trustworthy AI Governance ,” SAS , https :// www . sas . com / sas / whitepapers / a-comprehensive-approach-to-trustworthy-aigovernance . html # formsuccess
12
Vrushali Sawant , “ The ethics of responsible innovation : Why transparency is key ,” SAS , March 30 , 2023 , https :// blogs . sas . com / content / sascom / 2023 / 03 / 30 / the-ethics-of-responsible-innovation-why-transparency-is-key /
13
Ibid .
70 acamstoday . org