My rendezvous with AI in AI

As I settled in my seat in the AI (Air India) flight from America to India glancing magazines pages, some carried legacy of JRD Tata on his birth anniversary as founder of the Air India before nationalization and AI current ambitious plan of transformation. However, almost all of them carried articles on how AI is and will change the world for sure (in this case Artificial Intelligence).??We are in historic moment, which most likely be remembered as the time when AI has moved out of R&D lab and research to using new AI tools every day and in everything we do. More like the same way as Air India ambitious to be one of the best global airline again. My mind ran through the years from my engineering undergrad days, and my rendezvous with AI time and again. I thought of penning this down as acknowledgement of wonderful technologists who helped me see the wide possibilities of AI newer ways of in human-machine collaboration.

My first rendezvous with AI in Undergrad Engineering: ?AI and Computer Vision/Graphics was 7-8th semester subject in our Computer Engineering at NIT Rourkela. As a theoretical paper subject, I had gone through the college-days motion of learning. However, that changed when our department head Prof RK Chhotaray suggested that we do a project on AI based pattern recognition. Our project team went through nights of not knowing how this will unfold. Finally, we did well through difficult questions in our assessment. In my opinion computer graphics project that Abu Bakar did was more interesting and provided a glimpse how morphing images could eventually lead to AI image generation, which now, we can creatively create with gen AI LLM models

Starting at Work - AI & Robotics Fascination: Chance drove me to the opportunity where I came close to work in Robotics, but more importantly how data/AI/computer vision can bring life to an non-operational robot in a steel plant mid-90s. Robotics project got scrapped due to financial viability ( or alternative was cheaper). My first realization that financial viability triumphs. It was not until 2005, in stock exchange surveillance initiative, where I started to explore patterns in trading data ( and AI) for detecting insider trading with FINRA

Information Management – Dipping my toe in AI with Unstructured Content: As I started to work with Information Services providers in mid-2000s extracting valuable information from unstructured content in pdf/images/digitized extracts, and answering questions became a significant opportunity. Could a machine go through reports of company earnings and answer human-like with contextual summary ? Can table notes be understood by machine and financial numbers accurately computed across diverse document and formats just like human experts do. We saw immense potential of such use cases like in construction information automation from 100s pages of bid document. Interesting work that I had the privilege of working with our scientists Lipika Dey to build early PoCs.?

Building platform solutions using AI: TCS Exegenix leaders famously code lined ?“No Document is unstructured, there is always a structure that human eye is able to identify.” ?Exegenix could ingest the documents and use models for entity extraction and linking, decades before Cloud made such computation possible for everyday business use cases. ?Exegenix platform helped legal information providers ingest legal documents/case-laws and use computer vision and early gen AI models/intelligent document parsers Charlie Halpern-Hamu . Later on with Manish Mandal Muhammad Shamim positioned SCMB as a cognitive smart content solution as they created cognitive services with open source AI models and tuning them for use cases

TCS Content and AI Lab – exploring AI use cases : By 2010s there was realization that AI will be a disruption in Natural Language processing with google showing the way and evolution of models like ELMo and BERT. It was important for TCS to bring together our contextual masters in media vertical and AI technologists together to solve our customers use cases using AI. Anjan Dutta and my Content AI team work helped us leverage cognitive capabilities with cloud compute to evolve our solutions in media industry vertical. Several use cases emerged in almost every sub-segment of media industry. ?Mostly we explored open source technologies but also worked with our technology partners. IBM Watson was fresh from chess win and Jeopardy fame. Our talented AI engineers from the lab with help from Shatadru to explore several interesting AI use cases along with Watson GTM team.

I contextualized the work to create Content Agility Framework and published two papers Content Agility 2.0 for Legal & Tax , Collaborative Agility ?for Scientific Information Services. Number of those capabilities in cognitive services finally made it to our patented TCS Smart Content Management Backbone.?

NLG – next gen AI based content writer (2010s): ?Some interesting cases started emerged like companies still take a lot of time in creating insights with human written reports. How could the machine just now analyze but also generate content i.e. reports. There were interesting AI use cases from pattern matching/fuzzy logic to AI based best fit mapping platform to fit the right taxonomy structure so that Nielsen can answer smarter. Significant opportunity for automation and elimination of repetitive manual work that needed SMEs.

Another set of use cases came out how do we have machine generate the marketing content for the website. Our cross-functional team of TCS AI technologists in our lab, copyeditors, and our Co-Innovation partners created a AI solution to generate the first draft website content. Little did I know that guidance to models which we needed to provide to refine models will be known as Prompt Engineering today.

Image and Video AI uses: Was fascinating to some interesting work in AI-based object detection in Video and images work was done by TCS scientists in our Media R&A teams led by Niranjan Pedanekar , and also our Co-Innovation partners. The PoVs / PoCs, eventually materialized to patents.

Conversational AI use cases brought in how user experience and engagement redefined, and opening up another set of use cases and capabilities. But all capabilities needed specialized models, and absence of general large language models was seen as a challenge. ?

Decade of 2020’s - Powerful AI Models galore: With rapid evolution of powerful computing chips, cloud platforms and General Purpose AI models, it is now truly the age of AI revolution. OpenAI released GPT3 (Generative Pretrained Transformer), a powerful general-purpose model in Spring 2020, which then evolved to ChatGPT3.5 and now multi-dimensional LLM ChatGPT4. ChatGPT has captured our imagination becoming the fast growing product, and all Cloud Platforms are in rush to bring their own LLM models and partnering. Google came out with BARD/PaLM, AWS released TITAN/Bedrock. All cloud platforms are doing onboarding popular LLMs and MLOps. ?We also are seeing debate over specific models for of use case category – some are better in text, some in chat and some in image or video; are in a specific industry use cases-themes combo like summarizer for legal or studio content video production. We have an unique opportunity to reimagine almost every process have AI co-pilot for everything, just like the flight has real pilots but can fly with automated co-pilots.

New concepts have emerged - Prompt Engineering, Diffusion, Co-Pilots. Lot to learn and reskill. Back to being student again and my rendezvous with AI continues.