Today AI permeates every aspect of business function. Whether it be financial services, employee hiring, customer service management or healthcare administration, AI is increasingly powering critical workflows across all industries.
But with greater AI adoption comes greater challenges. In the marketplace we have seen numerous missteps involving inaccurate outcomes, unfair recommendations, and other unwanted consequences. This has created concerns among both private and public organizations as to whether AI is being used responsibly. Add navigating complex compliance regulations and standards to the mix, and the need for a solid and trustworthy AI strategy becomes clear.
To scale use of AI in a responsible manner requires AI governance, the process of defining policies and establishing accountability throughout the AI lifecycle. This in turn requires an AI ethics policy, as only by embedding ethical principles into AI applications and processes can we build systems based on trust.
IBM Research has been developing trustworthy AI tools since 2012. When IBM launched its AI Ethics Board in 2018, AI ethics was not a hot topic in the press, nor was it top-of-mind among business leaders. But as AI has become essential, touching on so many aspects of daily life, the interest in AI ethics has grown exponentially.
In a 2021 study by the IBM Institute of Business Value, nearly 75% of executives ranked AI ethics as important, a jump from less than 50% in 2018. What’s more, suggests the study, those organizations who implement a broad AI ethics strategy, interwoven throughout business units, may have a competitive advantage moving forward.
The principles of AI ethics
At IBM we believe building trustworthy AI requires a multidisciplinary, multidimensional approach based on the following three ethical principles:
- The purpose of AI is to augment human intelligence, not replace it.
At IBM, we believe AI should be designed and built to enhance and extend human capability and potential.
- Data and insights belong to their creator.
IBM clients’ data is their data, and their insights are their insights. We believe that data policies should be fair and equitable and prioritize openness.
- Technology must be transparent and explainable.
Companies must be clear about who trains their AI systems, what data was used in training and, most importantly, what went into their algorithms’ recommendations.
When thinking about what it takes to really earn trust in decisions made by AI, leaders should ask themselves five human-centric questions: Is it easy to understand? Is it fair? Did anyone tamper with it? Is it accountable? Does it safeguard data? These questions translate into five fundamental principles for trustworthy AI: fairness, robustness, privacy, explainability and transparency.
AI governance: From principles to actions
When discussing AI governance, it’s important to be conscious of two distinct aspects coming together:
Organizational AI governance encompasses deciding and driving AI strategy for an organization. This includes establishing AI policies for the organization based on AI principles, regulations and laws.
AI model governance introduces technology to implement guardrails at each stage of the AI/ML lifecycle. This includes data collection, instrumenting processes and transparent reporting to make needed information available for all the stakeholders.
Often, organizations looking for trustworthy solutions in the form of AI governance require guidance on one or both of these fronts.
Scaling trustworthy AI
Recently an American multinational financial institution came to IBM with several challenges, including deploying machine learning models in the hundreds that were built using multiple data science stacks comprised of open source and third-party tools. The chief data officer saw that it was essential for the company to have a holistic framework, which would work with the models built across the company, using all these diverse tools.
In this case IBM Expert Labs collaborated with the financial institution to create a technology-led solution using IBM Cloud Pak for Data. The result was an AI governance hub built at enterprise scale, which allows the CDO to track and govern hundreds of AI models for compliance across the bank, irrespective of the machine learning tools used.
Sometimes an organization’s need is more tied to organizational AI governance. For instance, a multinational healthcare organization wanted to expand an AI model that was being used to infer technical skills to now infer soft/foundational skills. The company brought in members of IBM Consulting to train the organization’s team of data scientists on how to use frameworks for systemic empathy, well before code is written, to consider intent and safeguard rails for models.
After the success of this session, the client saw the need for broader AI governance. With help from IBM Consulting, the company established its first AI ethics board, a center of excellence and an AI literacy program.
In many instances, enterprise-level organizations need a hybrid approach to AI governance. Recently a French banking group was faced with new compliance measures. The company did not have enough organizational processes and automated AI model monitoring in place to address AI governance at scale. The team also wanted to establish a culture to responsibly curate AI. They needed both an organizational AI governance and AI model governance solution.
IBM Consulting worked with the client to establish a set of AI principles and an ethics board to address the many upcoming regulations. This effort ran together with IBM Expert Labs services that implemented the technical solution components, such as an enterprise AI workflow, monitors for bias, performance and drift, and generating fact sheets for the AI models to promote transparency across the broader organization.
Establishing both organizational and AI model governance to operationalize AI ethics requires a holistic approach. IBM offers unique, industry-leading capabilities for your AI governance journey:
- Expert Labs for a technology solution that provides guardrails across all stages of the AI lifecycle
- IBM Consulting for a holistic approach to socio-technological challenges
The post From principles to actions: building a holistic approach to AI governance appeared first on Journey to AI Blog.