‘Trust and governance must be at the core of enterprise AI’: IBM’s Sriram Raghavan

The IBM executive described three key aspects of this approach: AI ethics, trustworthy AI, and AI governance. “AI ethics refers to what an organisation believes is the right thing to do, which is a societal decision, not a technological one. Trustworthy AI involves technology that allows enterprises to measure models, detect bias, and make models more robust,” Raghavan said. 

AI governance is more about processes, “ensuring you know what data went into your model, what tests were performed, and that you can monitor its behaviour post-deployment. IBM offers trusted models and a platform, Watsonx.governance, that lets businesses implement governance, ensuring that they can continuously monitor and adapt AI as it is rolled out,” he said.

Challenges in AI adoption

When asked about the main challenges that organisations face when integrating or scaling AI, Raghavan enumerated four aspects – skills, cost, data, and trust. He said that IBM’s strategy to address these issues has multiple approaches: “We are focused on offering fit for purpose, customisable, small models, so that enterprises can start and do not have to spend for the biggest, largest, most powerful models.”

On data readiness, the IBM executive said that Watsonx has been designed as a “data and AI platform” to allow customers to prepare their data for AI applications. In order to foster trust, IBM offers its proprietary Granite models, which the company firmly stands behind and indemnifies clients for their use. When it comes to skill development, IBM works through its consulting arm and has made a global commitment to skill two million people in AI by 2030, which is aimed at bridging the talent gap. 

How is IBM helping organisations scale AI responsibly?

Raghavan explained that IBM’s approach to helping businesses scale AI responsibly involves, ethics, trustworthy AI technology and governance processes. The executive emphasised on the significance of use case-based risk assessment. “You can’t look at risks only at the model level. Risks are at the use case level,” he said. 

To illustrate his point, he compared two scenarios: “Imagine a model making a recommendation for the next shoe you should buy, versus a model recommending cancer treatment. These are clearly not at the same risk level.” He added that risk assessment is a continual process.

He said that the company also offers InstructLab that allows for incrementally adding new skills and knowledge to an AI model. This, according to Raghavan, allows companies to customise AI models for their specific data and use cases. 

IBM’s approach to foundational models

When asked to elaborate on IBM’s approach to building foundational models, Raghavan said that IBM’s approach has two key elements – open sourcing and focus on useful model sizes. With open-source, he said that IBM is committing to open innovation in AI. “We joined forces with Meta to incubate the AI Alliance.” On model sizes, Raghavan said that the company has restricted most of its models to the 7 to 30 billion parametre range which makes them more usable and accessible. 

“We offer these models with a standard Apache 2 licence, giving developers, partners, and clients full freedom of action. For enterprise customers, we provide continued improvements, revisions, and security patches for our Granite models,” he said. 

On IBM’s safety guardrails for its AI models, Raghavan said that the company offers full-fledged guardrail capability through Watsonx.governance. These capabilities include pre-packaged detectors for bias, hate speech, etc, and capabilities allowing customers to define additional policies. The VP also stressed on the importance of configurability, “we believe configurability is important, and we always use the term use case based risk and regulation, not model based risk and regulation.”

AI landscape in India

When asked about envisioning India as a global leader in AI adoption, Raghavan said that India has significant potential owing to its vast talent pool and innovation capabilities. “India has a lot going for it in terms of talent and innovation. IBM is deploying Watsonx in CDAC’s Airavat platform and partnering with L&T Semiconductors to build AI capabilities at the chip level, aiming to address the industrial and automotive sectors.”

When asked how AI will contribute to job creation and skill development in India, Raghavan admitted that although some jobs may be impacted, new ones will be created. He asserted that the key is to focus on skilling initiatives. “The more we are able to enable our nation’s citizens to pick up the skills that are going to be required, you’re going to be able to straddle the transition.” 

He noted that the rapid pace of AI development comes with challenges and opportunities. “One of the big differences between this kind of technology and past transformation is it’s happening at a rapid pace. So the time window available to us as humans to sort of adapt, it requires us to be even more agile.”