AI Risk Management Programme

To establish the foundational principles of responsible AI, you need to go back to the basics of differentiating the various parts of AI ethics from Ethical AI, Responsible AI, and AI governance. 

Using a blindfold grocery analogy, participants will learn the philosophical and foundational moral principles used to guide the fair, safe, human-centred and sustainable development of AI.

We’ll then move to an interactive session of "Unmasking the Rotten Oranges”. Attendees will be tasked with identifying "rotten oranges" (hidden biases), "broken eggs" (privacy leaks), and the completely replaced items (unintended toxic features) in a selected AI model and domain. 

The class will close with a simple risk assessment of their selected AI model using a tiered structure from most critical to least critical. 

Key Takeaway: AI systems need to be vetted with risk assessments before any AI audits can be carried out.

There are several important aspects of the ML/AI development lifecycle you need to consider, from procurement or development to deployment, use, responsible AI, and AI governance processes.

We’ll deconstruct the 8-pillar responsible AI framework, learning the various areas of safe AI during AI use and adoption. We’ll also review case studies highlighting the good, bad and ugly sides of AI, with a key emphasis on the “good”. 

Understanding the different guardrails and their limitations will be our next focus, reviewing these from the 8-pillar framework.

We’ll round off with a trivia quiz identifying the different parts of AI between the good, bad and downright dirty ugly. 

Key takeaway: Following AI risk assessments, mitigation strategies and guardrails need to be applied. 

AI governance establishes the precise rules, oversight mechanisms, and internal controls required to keep systems legally compliant and socially safe. 

Looking at Moore’s law, we’ll evaluate if AI’s growth velocity still mirrors Moore’s law and the need for regulatory and ethical safeguards. We’ll have a quick overview of existing AI regulations across the world.

A high-performance governance framework relies on five key areas. Participants will role-play an “AI Crisis Room”, and work on an executive compliance board to deploy the 5 key areas of governance. 

The executive board will review 3 different high-stakes AI applications assessing risk tiers and compliance with regulatory frameworks across the world. 

Key takeaway: AI governance is a critical part of AI adoption and requires ongoing risk assessment and compliance frameworks. 

Fairness in AI demands the equitable treatment of individuals and groups, protecting marginalized or historically disadvantaged populations from digital discrimination. 

Fairness in Generative AI requires a non-invasive treatment of human users to prevent cognitive decline, surrender and overreliance on AI applications.

We’ll review the various bias entry-points in AI development and the impact of AI use on the human psyche. 

Setting up 2 live scenarios, participants will fix gender bias in a loan financing company and design a human-AI framework to preserve human cognition and agency. 

Key takeaway: Fairness in AI systems mitigates bias and diverse teams are part and parcel of bias mitigations. To preserve human agency, Human-AI frameworks should be adopted. 

To ensure AI is deployed and used safely without any unintended systemic risks or downstream failures, there are four safety principles that must be applied. 

Using trivia quizzes, we’ll have an interactive session understanding the workings of a complex automated decision, and how an organisation can apply transparency and explainability principles to break down the decision making process from the AI system.

The primary goal is to understand the opaqueness of generative AI and AI systems in general, and learn the various ways to explain the results of each system to a user. 

Participants will then role-play different scenarios explaining the different variables to different stakeholders in plain language.

Key Takeaway: All companies using AI systems must be accountable and have transparency measures in place, which will improve AI safety. 

The growing costs of hyperscaling, water / energy consumption and the impact supercomputer datacentres are having on communities are compounding.

Top of mind for most people is how to measure AI energy consumption. 

Participants will learn how to evaluate water and energy consumption when using generative AI. 

There’ll be an interactive session with an eco-AI Dragon’s Den, where new eco-friendly AI business solutions will be pitched to sustainable investors - the group. 

We will wrap up with an overview of the various existing tools that measure the energy consumption of AI models. 

Key Takeaway: It is possible to reduce water and energy consumption from data centres and AI use.

We’ll wrap up the course by bringing together all the different aspects of responsible AI and AI governance. 

We’ll break into groups and build an AI governance process for your organisations and a responsible AI framework for your technical teams. 

A human-AI framework where your teams are front and centre of all business processes will also be developed, with a focus on the various stakeholders from your customers, internal teams, policy makers, and communities. 

Participants will leave with a solid AI governance and responsible AI template and most importantly, a thorough understanding of how to reduce AI risks in an organisation and users.

Bring your questions, and live challenges for a 1-hour unstructured session, where Toju will answer your questions and real life challenges that have been brewing from the first 7 sessions.

Participants are encouraged to come with:

• A governance, ethics or compliance question you’ve been wrestling with
• A specific AI tool or application that raises ethical questions
• A project that’s full of ambiguity and uncertainty on the risks
• A policy that is narrow towards one geographical region
• Anything that needs further clarification or analysis.