Ways to engender trust in AI

--

Rajesh Varrier

From personalised buying recommendations to drone-assisted farming to fraud detection in insurance claims and predictive maintenance of aircraft, artificial intelligence (AI) and large language models (LLM) are already enabling several exciting real-world operations.

While there is a general sense of urgency towards AI integration to optimise costs and vendor ecosystems, this excitement is tempered by multiple bottlenecks, especially concerns over data privacy and trustworthiness of LLM outputs.

Before getting into the challenges, let’s consider the opportunities. McKinsey has projected a long-term AI opportunity in excess of $4 trillion as a result of possible productivity gains through enterprise AI. But while AI has the potential to help rejig business and operations, co-create products and services, and redefine workforce dynamics, organisations are unable to tap into its full potential.

Realising the potential of AI through the three-vector strategy

A true transformation across engineering and business processes in the evolving enterprise AI landscape requires organisations to explore a three-vector strategy that spans hyperproductivity, scale, and agentification.

Improved workforce productivity is one of the big promises of generative AI. Studies suggest that individuals who work in customer support, software development or consulting have seen average productivity gains ranging from 5% to over 25%. This hyperproductivity can potentially drive major efficiency gains, such as using AI to generate code. We are seeing companies across the globe who are gaining enterprise agility, cost savings, and a productivity boost by automating manual bottlenecks while integrating generative AI.

Next, industrialising AI at scale is necessary for businesses to see sustainable returns from their investment. Modernising cloud and data platforms will pave the way for scalable, differentiated AI adoption. This also means establishing the last-mile infrastructure – for instance, domain-specific LLMs trained around specialised data sets and customised reasoning frameworks.

In addition, the agentification of enterprises can transform service delivery through embedding AI agents into the workforce. Unlike generative AI models, agentic AI systems are proactive and collaborative, can interpret and handle complex tasks, take multiple actionable steps, talk to external tools, and learn on the go – for the most part, with very limited human oversight. Thus, agentification can, in a way, augment creation of completely new service pools.

This not only promises to significantly uplift core workflows but also adds the exciting prospect of AI agents and Humans working alongside – each injecting their own specialised knowledge and unique strengths to open up untapped service pools.

Responsible AI is non-negotiable

Undeniably, AI is a driver of growth and innovation. However, it must be coupled with another imperative – to uphold ethical standards for protecting individuals, communities, and institutions – for the true impact of AI to play out in the long term.

Despite its popular appeal, generative AI does not intrinsically inspire universal trust because of potential biases and errors that remain shrouded in its black-box approach.

Agentification, too, has its share of deterrents – future workforces are expected to be a hybrid mix of human and AI resources that must be optimised for accuracy, costs, productivity, and new growth possibilities. But how can humans learn to trust their digital colleagues to carry out their share of tasks with unimpeachable precision and consistency?

Ways to offset distrust around AI

The current lack of trust and other concerns around AI can be offset through leadership, governance, and ecosystem collaboration. This means building transparency around how AI systems are developed and deployed and how these are designed around values and outcomes that matter to the stakeholders – employees, customers, and wider society.

Measures to uphold technical guardrails such as prompt engineering, output filtering, and safety classifiers as well as regulatory requirements are critical. For example, trust metrics can be embedded into the app, enhancing traceability and explainability around the model’s decision-making process.

Incorporating source references, continuous feedback and improvements to the AI systems will help users grow more confident and less sceptical of the mysterious nature of AI-generated output and decisions. This confidence will grow further by putting up safeguards to drastically reduce the possibility of systemic biases, and errors.

Role of human overseers remains vital

Above all, governance is key as the role of human overseers remains vital to ensure timely hand-offs from AI systems and maintain non-biased, ethical practices all the way. The leadership can prove instrumental in building an organisational culture that thrives on accountability while being closely aligned with what their stakeholders expect.

Responsible AI is not a constraint on innovation. On the contrary, it can act as a catalyst for sustainable growth and long-term value creation by putting to rest some of the long-standing worries clouding over AI, such as trust and transparency.

Rajesh Varrier is President – Global Operations and Chairman & Managing Director, Cognizant India

Views are personal and do not represent the stand of this organisation