Geetha Raju is an AI/ML researcher actively exploring the most recent advancements in NLP/NLU. 

Geetha is working as an AI/ML engineer at TNeGA. She has completed her PhD at Anna University in sensitivity analysis in online social media.

INDIAai recently spoke with Geetha about her research path, the challenges she had during her research, and how she overcame them.

How did you get started with AI? How is it doing thus far?

My AI adventure began when I opted to perform research on text analytics and online social media user behaviour in 2016. Since then, I've studied extensively online, worked with my PhD supervisor, Dr S. Karthika, and read countless blogs, articles, and research papers. And now, this fascinating quest of learning and implementing AI systems has landed me as an AI/ML Engineer in NLP at Tamil Nadu e-Governance Agency, a State Nodal Agency responsible for implementing AI initiatives in the State of Tamil Nadu, India.

How did you go from being an Android developer to an AI/ML engineer? What did you learn along the way?

My interest in AI/ML was by the fascinating decision to pursue PhD while working as an android application developer. My android projects involved psychological variables, sensor data to analyze human behaviour, decision-making, and mentoring. This project provided an overview of essential determinants in designing and developing a system to assist humans in decision-making; how external factors influence your system's decision-making process, resulting in convergence or divergence in meeting your ultimate goal. 

Beyond having a solid grasp of statistics, probability, and AI/ML models, I believe that a fundamental understanding of the applications and users of your AI product will be very beneficial in enhancing the real-time performance of the model. 

Can you describe your AI/ML engineer role on TNeGA?

My role as an AI/ML Engineer for NLP at TNeGA is typically R&D, where I set objectives and scope by the requirements and data accessibility for e-governance.

My focus areas are working on building NLP-based AI/ML models, data science projects, defining and evaluating scopes for AI project requirements received from various TN government departments, including government-to-government (G2G) and government-to-citizen (G2C) scopes. 

What challenges did you confront when you first entered the field? How did you triumph over them?

Initially, I faced numerous challenges working independently as a full-time AI/ML researcher. Therefore, during the PhD, I chose three out of four-course works related to AI/ML, big data technologies, text analytics, and information retrieval. I also attempted to build a use case relevant to my research work. This research was extremely helpful in understanding and overcoming the perplexing situation.

My research required a new dataset, so gathering, preparing, and standardizing data was a significant challenge. Considering AI as a black box and connecting the necessity to answer my research problem with AI proved challenging. To remain confident and unstoppable, I continued to learn, upskill, and follow the current trends and innovations in AI/ML.

Can you tell us about your research on online social media and some interesting challenges you faced during this phase?

My research on predicting the sensitivity of a social media post in personal, professional, and health domains. It entailed identifying "Principal Sensitive Entities" and "Sensitive Privacy Keywords" in each field, which aided in formulating a sensitivity scale for quantitatively estimating the level of "privacy risk/user regret" on OSM.

The greatest challenge was defining guidelines for discovering sensitive data and vulnerability in a post based on user context, global standards and cognitive science theories. This challenge led to the evolution of a ruleset known as Rules of Contextuality and Sensitivity.

Researchers often overlook the significance of transforming their research findings into a societal product/service. This approach piqued my curiosity, so I sought ways to leverage my research findings and productional them. Eventually, I decided to develop an AI-powered App that detects sensitive self-disclosures in text, notifies you, and allows you to review and edit, thereby sparing OSM users from the regrets and privacy concerns that arise with PII self-disclosure.

Since you've been in the AI and machine learning community for a long time, what are some of the most common myths you'd like to dispel?

Considering AI as a black box, I've seen that the industry and scientific communities have highlighted several myths and ramifications about AI, such as, 

1. 'AI is completely automated and will deliver autonomous systems/services.' Any AI system should be trained, tested, and verified continuously to enhance performance.

2. 'As AI advances, human resources and opportunities for the future workforce.' Truthfully, more sensible and intelligent humans would be required to develop an empowering AI system/service that could help humankind and the environment.

3. 'Yes…, this AI model is entirely accurate. Any AI system's accuracy fluctuates with time and data. It is a benchmark model if an AI model reaches perfect accuracy. Experience-based learning still exists, demonstrating that the performance of every benchmarked AI-model changes with data and other system characteristics.

4. 'AI models are data-dependent. Essentially, the performance of an AI model is by the data, algorithms, hardware, and human intelligence.

What, in your opinion, has been the significant advancement in AI over the past five years?

More specifically, remarkable progress has in Natural Language Processing, Understanding, and Generation, leading to the generation of coherent text, cross-lingual language services for translation, personal voice assistants, conversational systems, and question-answering systems. As a result, AI stakeholders are reaping more significant benefits from these emerging AI technologies in industries such as health care, e-commerce, agriculture, and education.

What advice would you provide to aspirants who wish to pursue a career in artificial intelligence?

To comprehend and extract potential contributions to your goals, I would recommend conducting as much research as possible, actively engaging and collaboratively working with AI expert groups in your area of interest and following standards and industry best practices. Since product development is a long-term process, you must pursue it right. 

Could you list some key research publications and books that influenced you?

The following research works inspired me,

How digital natives make their self-disclosure decisions: a cross-cultural comparison

Unpacking the Process of Privacy Management and Self-disclosure from the Perspectives of Regulatory Focus and Privacy Calculus

Semantic Space Theory: A Computational Approach to Emotion

The world-renowned physicist Stephen Hawking said, "AI is either the best or the worst thing that could ever happen to humanity. If we are not cautious, it could indicate the end of the human race". This approach sparked my interest and steered my research toward humanity and its well-being. As a result, I eventually began to follow everything about 'AI for Social Good.