AI/ML based solutions are becoming ubiquitous, solving many critical problems impacting people's lives. Banks and financial institutions are increasingly relying on AI based solutions to predict which individuals and entities represent the best risk to reward trade-offs. Insurance companies are relying on AI systems to identify potential fraudulent policy claims. However, machine learning algorithms that make this possible are black boxes, difficult for people to understand and trust. Moreover, since they are based on historical data, they may also perpetuate past biases and discrimination inherent in society. Considering this, there is a need to ensure that AI algorithms developed by organizations are transparent, ethically sound, and available for scrutiny. Mphasis Responsible AI is a unique framework that helps clients make their AI solutions more interpretable, non discriminatory, and available for scrutiny. The framework aids users to perform the following tasks:

1. Bias Detection: The solution employs several techniques are used to find if model has unwarranted bias towards any class. It can be used to establish fairness/unbiasedness in the model.
2. Bias Mitigation: Several human interventions and automated techniques are used to mitigate the biases identified using bias detection or pre-known. It can help in debiasing the data before initiating model building.
3. PII Protection: Several techniques are used to convert PII data into anonymous and not humanly decodable while keeping the mapping secure to use it for future purposes.
4. Model Debugging: Performance measures across a range of inputs are measured to find out if there are any anomalies. Model comparison with competing models or new scenarios can also be performed.
5. Highlight Key features: The global explainability techniques highlight the key features used by the model. It helps in explaining what the major patterns are which the model has identified in terms of features.
6. Decision Explanations: Using local explainability techniques decision taken by model are explained to understand the effect features had on them.
7. Governance: Methods which ensure traceability and integrity of audit trails are deployed to keep track of how data is used, or model is built for later scrutiny.

Assessments & Workshops

Our Assessments and Workshops help clients get started with responsible development of AI solutions. This engagement enables clients to plan and identify the relevant business problems which require Responsible AI. Assessments are a fast way of identifying how Responsible AI components could be leveraged for solving a business problem. In the workshops, the Mphasis AI/ML experts, SMEs, architects, and data scientists work closely with the client stakeholders to understand the system and data. The engagement includes activities such as stakeholder interviews, assessment and user journey workshop, identification of frameworks and tooling, gap analysis, understanding and assumptions, recommendation of solution, and report-out. The typical duration of the assessment and workshop exercise is spread across 3-4 weeks and the outcome of this exercise is a prioritized set of use cases along with recommended infrastructure to solve the problems, recommendations on frameworks and tooling, and reference architecture.

Algorithms Development and Implementation

Algorithms Development and Implementation is a 2–3 month long engagement, where we bring our Responsible AI IP as well as customizations to solve the specific client problems. Mphasis Responsible AI has been implemented in the solutions for our clients in insurance, banking and pharma domains. Our services are powered by our patented AI/Ml platforms and frameworks such as DeepInsights, PACE-ML, HyperGraf and InfraGraf. The Mphasis IP overcomes the limitations of systems not being able to work on varied input datasets and consists of 200+ pre-built and customizable Machine Learning, Deep Learning and Natural Language Processing algorithms. Our IP helps clients arrive at actionable recommendations faster and provide measurable results across scenarios. Our experience in AI/ML algorithm development and implementation are spread across several industries as well as use cases such as improved customer experience, recommendation of right products at the right time for the right customer, profit margin maximization, faster checkout processes, improved productivity, reduced disputes, reduce false positives, anomaly detection, predictive analytics recommendations, text categorization on large document corpus, natural language generators etc.