The way Indian enterprises are embracing AI is noteworthy: Geeta Gurnani, IBM

Since the emergence of artificial intelligence along with the advancements of generative AI, IBM has been a key player in research and development for the niche. Its Watson system is originally designed for question-answering tasks and fulfilling the requirements of a perfect AI chatbot. 

In a conversation with TechCircle, IBM's Technology CTO & Technical Sales Leader for India & South Asia Geeta Gurnani shared insights into how Indian businesses are leveraging AI for innovation, along with the key focus areas and trends shaping the industry. Edited Excerpts: 

What are the key trends shaping AI in India right now? 

The IBM Global AI Adoption Index 2023 report reveals that 74% of Indian enterprises are already using AI. Indian businesses are investing in workforce reskilling and R&D to enhance their AI capabilities. 

After months of evaluation, it's clear that many enterprises are now focused on how to scale and industrialise AI. A key trend is the operationalisation of AI, where businesses consider factors like cost versus return of investment (ROI) and the suitability of different models for specific problems. 

Another emerging trend is, understanding that Gen AI alone isn't enough. Businesses need a combination of Gen AI, traditional AI, and automation to be effective. Automation is crucial for process optimisation, and ensuring data integrity, proper process automation, and security are all vital. 

Finally, the most critical discussion revolves around trustworthy AI. Businesses have numerous concerns like whether their AI models are explainable, robust, transparent, fair, and privacy-compliant. They are also scaling AI ensuring the security posture of their organisation as a top priority.

How do you think AI will be adopted by Indian businesses in the next five years? 

Predicting how technology will evolve in five years is challenging. However, the way Indian enterprises are embracing AI is noteworthy. AI is becoming integral to digital workforce functions and can be seamlessly applied to Gen AI use cases like IT support, customer care, and legal support. 

Another key area is knowledge workers. Businesses can leverage individual expertise to solve problems and speed up time to market. For instance, an auto manufacturer,  I spoke to, was struggling to keep track of original component configurations due to numerous design changes. They employed over a hundred people to manually track these changes. Thereafter, we implemented a solution to automate this process, allowing them to detect even the smallest changes throughout the design stages. 

Another significant trend is code generation. With India being a major developer market, we will see AI transforming how developers generate code, accelerating the software development lifecycle. 

Additionally, AI will become more pervasive in enterprise tools, including infrastructure management, security, and finance tools. More products will incorporate Gen AI features. 

Looking ahead to the next five years, IBM has identified several key areas for Gen AI is language, code generation, time series, and geospatial models. Beyond large language models, IBM is exploring large models for time series data, geospatial data, and code generation. 

How crucial is infrastructure for scaling AI? What unique challenges do Indian enterprises encounter in this aspect? 

The availability of infrastructure remains a challenge, prompting enterprises to consider which specific models can address their specific problems. For instance, a business shouldn't use a 30 billion parameter model to solve a minor HR issue or a particular customer care problem. Companies will become more mindful of selecting the right model for each problem to optimise infrastructure use. 

Significant advancements are being made in training models more efficiently, reducing infrastructure demands. At IBM, we developed our Granite language model with a unique training and inference method that uses less infrastructure than traditional approaches. Many open-source communities are also working on this, leading to an increase in available models. 

At IBM’s Think event in Boston recently, we released several models, ranging from 7 billion to 30 billion parameters, into the open-source domain for experimentation. Careful model selection will be crucial as infrastructure remains limited, and there is growing interest in targeted, smaller models due to their efficiency in training and use.

What is IBM's approach to AI governance? How does the company ensure the ethical use of AI?    

IBM has long been a leader in responsible AI. It's clear that AI governance applies to both traditional AI and Gen AI. IBM's approach has always prioritised responsible AI from the start, not as an afterthought. 

Previously, with traditional AI, you had full control over your models, including the data and algorithms used. This ensured confidence in data handling and governance. With Gen AI, however, your base model isn't trained on your own data, making AI governance more critical. 

IBM addresses Gen AI governance through three key areas. The first is lifecycle governance, which involves managing the entire inventory of models used across your enterprise, including those from various vendors and open-source communities. The second is compliance, which requires adhering to regulations, such as those from the EU and US, necessitating robust tooling to ensure compliance for every model and project. The third is risk management, which involves continuously monitoring and mitigating risks like model drift, bias, and fairness. 

IBM supports enterprises with its platform, IBM watsonx, to tackle these challenges of lifecycle governance, compliance, and risk management. Recently, IBM partnered with Tech Mahindra to integrate watsonx.governance into their AI suite, enhancing overall governance. 

Additionally, IBM shares its internal best practices with clients, drawing on its extensive use of Gen AI and oversight by its AI ethics board.

What are the upcoming features or innovations that we can expect in watsonx?    

We've open-sourced the IBM Granite model, making it available to the open-source community. This move promotes open innovation. Next, we have InstructLab, a new approach to training models. Previously, you could only fine-tune parts of a model, not incorporate the full knowledge and skills of your organisation. InstructLab changes that by dividing the model into three parts: data, knowledge, and skills. 

For example, in the banking sector, knowledge is a home loan offer that a bank provides. Skill, on the other hand, involves knowing how to use this offer and what it means for you personally. 

InstructLab helps enterprises train models with their own data in two domains, skill and language, while keeping the base data model unchanged. This innovative approach is a major advancement in model training. 

Are there any specific areas of AI or any other technology that your company is currently investing heavily in, especially in context of the Indian market?

India presents a unique challenge for businesses due to its vast consumer base across various sectors like banking, telecom, and retail. Solving India-specific issues, especially for large enterprises, requires leveraging innovative solutions like Gen AI. The key question becomes: How Gen AI can be used to effectively manage the intricate infrastructure of these enterprises? 

According to IDC, by 2028, there will be 1 billion virtual apps globally, posing a significant challenge in managing the backend infrastructure as organisations modernise their platforms. This involves various layers such as security, IT, network, and observability.

IBM has introduced IBM Concert, a tool powered by Gen AI, to address this complexity. It aggregates signals from the entire infrastructure and provides a single dashboard for comprehensive oversight. Additionally, it allows users to communicate with the infrastructure management system directly. IBM Concert is poised to streamline the management of complex infrastructures for businesses.