Behind the code: Meet Saloni Potdar

Saloni Potdar, Manager and Senior Data Scientist, Watson Assistant
Finding Partnership in AI with IBM Watson

Saloni Potdar has made significant contributions to the Artificial Intelligence algorithms used in Watson Assistant, particularly in Natural Language Processing. After receiving a Master’s degree in Intelligence Information Systems from Carnegie Mellon, Saloni came to IBM Watson where she now leads and manages the team responsible for developing Watson Assistant Algorithms, and improving its capabilities to interact with humans.

We talk to Saloni about her passion for technology, how she uncovers ideas that shape the next generation of technology – and her advice for anyone interested in pursuing a career in AI.

How did you become interested in Artificial Intelligence?

In high school, I took a programming class to better understand how computers worked and how I could use them to build amazing things. I was amazed to find that programming helped me write simple pieces of code that could make a computer do fairly complex things. In college, I learned newer technologies – including machine learning and AI. By incorporating AI into solutions, I saw a dramatic increase in the capabilities of my programs.

For example, I built an AI solution that would identify and analyze any tweets offering legitimate advice and recommendations on the stock market: what to trade, what was trending, what to avoid. The system analyzed recommendations from a diverse group of experts with different backgrounds and delivered a collective recommendation. To our surprise, this system was more accurate than leading stock experts!

I decided to come to the US to study at Carnegie Mellon and gain more knowledge in the field of natural language processing and machine learning. At IBM, my interest in AI continues to grow as we work on cutting-edge algorithms and developing the latest chatbot technologies to assist human interaction.

What is the most interesting part of your job at IBM Watson?

One of the most rewarding parts of my job at a global company like IBM is when I get to see the impact of our work around the world. IBM has a diverse set of strengths in technology and it has been a rewarding experience to learn from leaders in the field in AI technologies and beyond.

After my internship at IBM, I started full time where I incorporated Deep Learning into our Watson Natural Language Classifier. While Deep Learning has been in existence for quite a while, today we finally have the infrastructure and computing power in place to see its full benefits. I now work on Watson Assistant where we develop conversational AI technologies that power chatbots to reduce costs, improve efficiency and provide a delightful experience to end-users with faster resolution times.

It is challenging to lead and manage a team focused on building out AI technology which is constantly changing and evolving. We need to work closely within our team to analyze the ideas to solve the major customer pain-points algorithmically, and in collaboration with the larger engineering team. Since AI is a rapidly evolving field, we also need to be aware of the latest developments and maintain close relations with research teams across IBM research. We also file several patents and publish research papers in leading scientific conferences like ACL, EMNLP, NAACL, etc.

What can we expect next from you? Do you have any exciting projects you can share?

We have been focused on improving the performance of Watson Assistant with any kind of data that the enterprise bring – noisy, imbalanced and domain-specific. We develop algorithms that power the intent identification, irrelevant detection, entity recognition, and spell-correction through deep learning, natural language processing and traditional machine learning approaches across many languages. A major challenge of using AI in the real world is incorporating various real-world constraints of the enterprises. For example, Data Privacy is of utmost importance which means that we have to tune the algorithms without having access to the data.

Our most recent project has been that around irrelevant topic detection. We often see a lot of virtual assistants responding to irrelevant queries as “Sorry, I don’t understand you”. Understanding such irrelevant queries is often pre-programmed into the virtual assistant. Enterprises build their chatbots using the IBM Watson Assistant cloud-based product. This means that the definition of “irrelevant” queries changes with the enterprise’s use-case.

The topic “banking” is out-of-domain for a “weather-bot”, whereas it is not out-of-domain for a “banking bot”. We have tried to solve this problem with our latest algorithm which has recently been launched as a beta feature in Watson Assistant.

What advice do you have for people pursuing a career in this field?

While I truly believe in the transformative capabilities of technology, and AI in particular, I think everyone developing such technologies should think of the bigger impact they have. For AI practitioners, this means understanding your data first and particularly its biases. A biased algorithm or an algorithm trained on poor data can do more harm than good. It is okay to be in awe of the predictive power of machine learning, but one must also be mindful of its limitations and generalization abilities.

I also think that making AI robust and applicable in the real world is an open problem. Building AI solutions is not just about the most accurate algorithms – there is a lot of engineering, UX design, and optimization effort required to build a usable solution.

Interested in interning at IBM? Check out the openings here. You can connect with Saloni on LinkedIn and read more about IBM Watson Assistant capabilities at https://www.ibm.com/cloud/watson-assistant/