Infosys is building core IP around specialised AI models: CTO Rafee Tarafdar

As of today, among the enterprise customers, the most common use cases observed are in eight broad areas, namely customer and advisory services, sales and marketing, business operations, IT operations, software engineering, search and knowledge management, learning and enablement, and employee experience.

Clients are currently experimenting and working on these technologies and use cases to improve productivity and efficiency, drive cost reduction, and improve customer and employee experience. As clients gain more experience with the technology and start reimagining critical business processes, the revenue generating opportunities to their business will emerge. This will happen through a combination of digital, cloud, and AI technologies.

Q. How are you addressing the issue of ensuring robustness, reliability, and security (without ‘hallucinations,’ for example) of your generative AI solutions so that they are as much enterprise-grade as any other application/IP you build for your customers?

One of the core principles of our ‘AI First’ approach is to be ‘responsible by design’. To implement this, we have enhanced our existing frameworks to cover AI-specific critical areas. These include regulatory compliance for AI, reproducibility and reliability, explainability and interpretability, fairness with managing harmful bias, privacy preservation, security, safety, IP protection and traceability for auditing.

We are using the framework to evaluate all AI use cases and ensure that risks and mitigation strategies are planned and discussed with all stakeholders. In addition, we have started working on codifying the policies into the AI engineering lifecycle to ensure automation and compliance.

We are making these available to our clients and have started applying to client projects through Infosys Topaz (suite of AI solutions). This area is very fast evolving and requires continuous tracking, planning, monitoring and is an ongoing area of focus for us.

Also read:  Infosys falls after outlook whiplash: 5 Takeaways from Q1 FY24

Q. Give us a sense of Infosys’s efforts to build its own fundamental AI models and technologies which may only see commercial applications 5-10 years from now, but which you feel are critical to your company’s long-term future. Tell us about what kind of problems you’re addressing in these efforts.

Infosys is transforming itself to be AI-first and as part of this journey we are looking at AI as a way to amplify the potential of our employees, unlock value from organisational knowledge and create client value.

We have started rolling out in phases, AI assistants for coding, learning, knowledge management and for client facing teams. These AI assistants are powered by a digital brain that brings all our structured and unstructured data together. To power the models and digital brain, we are using both closed access models and open-source models.

Using a narrow transformer approach, we have built specialised AI models based on open-source large language models, or LLMs. We will build more such specialised models that will become our core IP, and we plan to use these and AI builder playbooks to accelerate our clients’ AI programs through Topaz.

To drive automation and speed in implementation of AI projects, we have built and implemented an applied AI platform that enables automation and poly-AI architecture supporting multiple AI providers, multiple models and embeds responsible AI policies in the pipeline. As we scale our AI efforts, we will infuse it into all our current products and platforms over a period of time to make them AI-first.

Q. What is the experience so far if any ‘AI anxiety’ is showing up among employees, from the combination of worries about being replaced, the need to continually upskill and now benchmark against an AI twin as well?

At Infosys, the AI-first approach includes our talent strategy as an important pillar, with focus on three levels of enablement and upskilling. Level 1 is called AI Aware, wherein we are working on making everyone aware of generative AI technologies and how AI assistants can help them be more productive and relevant to clients.

Level 2 refers to AI builders who can reimagine experience and processes to build industry-specific AI-led solutions. Level 3 refers to masters who understand the under-the-hood workings of ML (machine learning), DL (deep learning) and LLMs.

They are working on harder problems like fine tuning, pre-training, runtime optimisation and responsible AI. With the right investments in training and enablement, we are able to drive adoption and employees are excited about the potential of technology, as it will help amplify their potential.

In order to build skills of future, we are also identifying Horizon 1, 2 and 3 skills (Note: The Three Horizons growth planning model was originally developed by the consultancy McKinsey, and it is now widely used.)

It helps us to structurally build deeper capabilities on emerging tech, plan talent transformation for legacy skills and do upskilling on AI for mainstream skills talent.