Worries over data quality undermine trust in AI projects

A survey of 400 data professionals across the US and Europe by Dataiku has revealed significant challenges in building trust in enterprise AI projects and perceptions across roles within the organisation.

Monday, 23rd March 2020 Posted 4 years ago in by Phil Alsop

Only 52 per cent of respondents said that their organisations have processes in place to ensure data projects are built using quality, trusted data. With topics like trust, explainability, responsibility, and ethics at the forefront of discussions in AI uptake, Dataiku asked respondents about how their organisations are managing these challenges.

When asked whether their organisation had processes in place to ensure data science, machine learning and AI are leveraged responsibly and ethically, fifty-seven per cent said no, or don’t know, despite thirty-five per cent saying they were working on it.

The perception of the impact of AI on people’s roles seems to differ greatly from CEO to non-management, suggesting AI projects struggle with inclusivity within organisations. Managers and C-suite executives were significantly more likely to respond that AI would “completely” (i.e., a 5 on the scale) change their company than non-managers.

On the other hand, despite the fact that non-managers in non-technical roles (business professionals in marketing, risk, operations, etc.) should see - or at least see the potential for - AI impact in their jobs, in practice, only 11 percent of non-managers in non-technical roles responded that they thought AI would “completely” change their role - a much lower percentage than the other more senior roles.

“Trust in AI projects will continue to present significant challenges if we are still tackling fundamental issues such as data quality, as well as more complex problems associated with ethics,” said Florian Douetteau, CEO at Dataiku. “Building internal trust will provide the foundation for external trust; this starts with trust in the data itself that is being used in AI systems. Data quality is one of the most basic but most important hurdles to overcome in the path to building sustainable AI that will bring business value, not risk.”

Inclusive AI encompasses the idea that the more people are involved in AI processes, the better the outcome (both internally and externally) because of a diversification of skills, points of view, or use cases. Practically within a business, it means not restricting the use of data or AI systems to specific teams or roles, but rather equipping and empowering everyone at the company to make day-to-day decisions, as well as larger process changes, with data at the core. The model today for traditional businesses leveraging AI seems to lean more toward data democratisation, or inclusive AI, for its larger potential to scale.

“It goes without saying that AI will impact individual roles, enterprises and industries, yet there are clearly some questions around trust, responsibility and inclusivity which need addressing before AI can have the optimal result,” added Douetteau.