The 3 Ts Of AI Governance: Transparent, Trustworthy And Traceable Data

The 3 Ts Of AI Governance: Transparent, Trustworthy And Traceable Data

The first AI Safety Summit, attended by the likes of Sam Altman and Elon Musk in November 2023, highlighted both the opportunities presented by ‘Frontier AI’, and the potential societal harms and risks posed by its misuse and loss of control. In light of this, AI firms are set to face increasing governmental scrutiny. Both the Biden-led US administration and UK Prime Minister Rishi Sunak have announced that leading AI firms must now give ‘early access’ to their models, to allow their governments time to test them, before launching them to the public. Google DeepMind, Meta and OpenAI have already agreed to this, with more firms set to join the roster imminently.

These recent developments have also spurred AI analytics providers to help their customers unveil the inner workings of the models they are using. One such provider is IBM, which has recently launched watsonx.governance, a new module in watsonx, the firm’s flagship AI and data platform. watsonx.governance will assist clients in understanding how and why their AI models produce certain results – and, more importantly, to comprehend the data going into these models.

Data quality is a central piece in the puzzle of AI governance: AI is merely a tool, and training it with incomplete, incorrect, outdated or biased data can only result in outcomes with the same faults. In the context of industrial firms, ‘bad’ data comes in the form of data recorded from faulty sensors, data silos preventing the contextualization of data, duplicated data, and more. To mitigate this, firms will need to adopt robust data management practices, using technologies such as DataOps solutions, which can keep a record of accurate, complete and anomaly-free data, and vitally, can trace the source of data and understand if they have been altered in any way. In fact, 78% and 71%, respectively, of the 304 respondents in the Verdantix 2023 operational excellence global survey recognize that data quality and establishing a single source of truth (SSOT) are either ‘very significant’ or ‘significant’ data management challenges for their business. Software vendors such as C3 AI, Cognite and Hitachi Vantara all offer SSOT and data lineage features as part of their data management platforms.

Despite many governments taking a ‘wait and see' approach to AI governance, industrial firms can set themselves up for success by investing in robust data management software with strong data governance features. This will not only serve as preparation for any upcoming regulations, but will also prevent losses from AI models hallucinating undesirable outcomes.

For more information on industrial AI analytics, generative AI and data management, read the following Verdantix reports: Market Overview: Industrial AI Analytics Solutions, Smart Innovators: Industrial Data Management Solutions, Market Insight: Ten Applications of Large Language Models For Industry, and Strategic Focus: Why Industrial Firms Need DataOps Platforms For Asset Management Digitization.