The Need for AI Governance
Introducing IBM AI Governance to unlock full potential of AI investments while mitigating risks
This is Part 1 of a series of blogs on IBM AI Governance. See the bottom of the post for more details.
I am a big fan on Thomas Friedman. He is a New York Times columnist and has written groundbreaking books like The World is Flat and The Lexus and Olive Tree. He focuses on revolutionary changes that happen to humans globally. His latest book is Thank You for Being Late. In it, he discusses how Moore’s Law, originally used as a model for the semiconductor industry, can apply to technology in general.
I love this graphic… it can apply to many aspects of technology’s speed and how we interpret it as humans. It is why if someone is a few minutes late for a meeting, we appreciate the extra few minutes of unscheduled time. It explains why new developments in technology seem to be coming faster and faster. And if we apply it to Artificial Intelligence and Machine Learning, it shows the need for AI Governance. Here’s how I would apply it for AI Governance:
AI Governance is the essential set of software tools to unlock the full benefits of AI. It governs the AI Lifecycle to enable organizations to ride the exponential growth of AI, but does it in a way that brings humans into the loop at the right time. AI Governance bridges the gap between the potential of AI and the humans that are building, monitoring, controlling, explaining, and reporting on the AI systems.
This gap is large today, and it is growing quickly. It is driven by four key developments.
Drivers for AI Governance
The need for AI Governance is quickly increasing because of regulatory compliance, responsible AI, risk mitigation, and new stakeholders of AI.
- Regulations for AI are heating up in countries worldwide. These regulations have been accelerating over the past few years, and now they are becoming real and and forcing changes to how automated decisions are implemented. There’s the EU AI Act, New York City’s AI Hiring Law, the Canada Bill C27, and almost 2 dozen more tracked worldwide. Some of these have a real bite to them: the EU AI Act can impose fines of up to €30 million or 6% of a company’s global revenue.
- Responsible AI has been a popular topic recently, and for good reason: it helps protect customer data privacy and loyalty. It does this by expanding upon what dimensions a data scientist has to optimize: no longer just accuracy and model performance, but now we need to give equal weight to fairness and explainability. This practice of Responsible AI is a key pillar of AI Governance.
- Risks are inherent in the automated decisions that AI unlocks, and they must be tracked and mitigated throughout the entire model lifecycle. That cannot be simply checked when a model is placed in production, but must be monitored and evaluated as early as when the business use case behind AI is discussed. This can include deciding not to use a certain type of technology, using personal information only when needed, reducing delays of putting models into production, and monitoring over time for accuracy, quality, and for proxies of protected variables.
- Lastly, as AI has grown into a business imperative, the stakeholders involved in the AI policies has grown as well. More players internal to the organization have become involved because of regulations, to protect the organization’s brand, and to ensure efficient operations. Some of these personas are brand new to the AI Lifecycle, but some of them may have new reasons to be involved. This can be a great thing: elaboration between groups can result in more trustworthy models as different teams weigh in on what a model should be doing. But with such a large team, it is important to have a system to ensure the C-suite gets their dashboard view while those responsible for model quality get the right notifications, all with a system that scales from 1 model to thousands, from a small team to an enterprise wide organization.
The IBM AI Governance software solution addresses these four drivers so an organization can unlock the full ROI of implementing a AI solution while managing risks all without overburdening their AI team. It is available to work with any data science stack you may have today, works on any cloud, and is available now.
If you are ready to try a demo, read our eBook, or want to speak to an IBMer, click:
Or continue reading to learn more:
- Part 2: 3 Key Areas of AI Governance
- Part 3: Top 4 Applications of AI Governance
#IBM #AIGovernance #Explainable Ai #ResponsibleAI #CloudPakforData
