Proper AI governance requires everyday action, say experts

And companies which are adopting AI at scale need to demonstrate clear visibility in three critical dimensions – value, effectiveness and oversight – in order for artificial intelligence integration to be successful throughout the lifecycle of the system.

The issue was highlighted in a report published earlier this year by operations consultancy Mozaic and Pinsent Masons that looks into the governance challenges being faced by companies amid the current AI boom.

Simon Colvin, co-author of the white paper and a technology expert at Pinsent Masons, said organisations which have successfully scaled AI do not treat governance as merely a policy document.

“AI governance needs to function as a part of any adopting organisation’s everyday decision-making, technology management, and risk oversight processes,” he said.

“If this integration is incomplete or it fails, governance frameworks will quickly become disconnected from how AI systems are actually built and used.”

The report explored the challenges governance of automated decision making and AI processes pose – and how much the process requires continual oversight rather than just crisis management.

For organisations, this means linking up governance processes effectively to provide clear accountability at board or executive level for AI deployment, how AI is deployed across the organisation, clear definition of roles in the oversight process to ensure accountability, and embedding these processes into day-to-day operations.

This means organisations looking at how best to align and structure their data, rights and legal oversight processes, and ensuring proper AI literacy across the whole workforce. It also means organisation operational oversight of models in production.

It may require external advice and oversight to ensure a level of expertise which may not currently exist in organisations, with the example of SAP being cited – following their launch of an AI Ethics Advisory Panel - drawn from a selection of external advisors across academia, politics and industry – who provide support and input to a steering group.

The SAP example highlights the challenges organisations face in integrating Ai effectively, and the need to bring in experience in both developing operating and governance structures, and in mitigating against regulatory and legal challenges.

“Effective AI governance is not simply a policy problem,” added Colvin.

“It is an operating model design challenge.”