Janvi Palan, a Senior Machine Learning Research Engineer at Samsung Research America, has been on an exciting journey, from her early passion for language and mathematics to her current work in hyper-realistic digital humans. As a technical lead at the Digital Humans division at Samsung, she has been inspired by the entrepreneurial energy and the opportunity to work on innovative research and cutting-edge technology, especially in behavior generation and conversational AI.
She started her journey with a degree in Electrical Engineering from BITS Pilani, Goa, and continued to study graduate school in Texas, where her love for language and logic led her to specialize in natural language processing and machine learning. Her projects include identifying echo chambers and hate speech in social media, understanding political discourse, and analyzing gender roles in literature because these are problems at the intersection of society and computing, with real effects on humankind. She believes AI can be used to mitigate these social issues by identifying disinformation, and bias, and helping users make informed decisions.
At Samsung’s Digital Humans division, a talented team of engineers and researchers are working to create digital human avatars that not only look real but also interact with users seamlessly, in real-time and with real emotive states. These digital humans are being designed with the ability to learn, empathize, and respond to human emotions. As AI-driven digital beings, they are set to revolutionize how we communicate and interact with technology. From customer service to virtual assistants and companions, the applications of these digital humans have the potential to reshape industries and redefine human-machine interactions. Janvi’s work in developing these lifelike digital humans at Samsung is helping push the boundaries of what is possible with artificial intelligence and creating a more connected and engaging future for all.
Creating lifelike digital humans comes with several challenges, such as overcoming the uncanny valley, tackling computational demands, and addressing user adoption. The problem of uncanny valley arises when we try to model “Realism” — we are used to seeing and interacting with human beings from the second that we’re born. Modeling this “human” behavior and “features” in a completely life-like way cannot allow for any errors, unlike typical AIML models, since we’re so adept at noticing errors in humans. Any mistake in the generative AI model would cause the digital humans to look unnatural and completely kill the experience. Similarly, in the language that these these digital humans speak, the quality of conversation, and the voice, has to be extremely human-like. A question-answering model like chatGPT simply won’t work – no human being talks like that. These are some of the fundamental challenges her team explores.
As the digital human-based interfaces advance, she envisions more applications in various industries, such as healthcare and sports. For e.g, there is a pilot program that introduces these AI-powered “digital humans” to assist customers in the San Diego Chevron ExtraMile stores. “Sam the Sommelier” is a hyper-realistic generative AI-powered virtual being that can interact with customers via voice or live chat through an app and provide recommendations for wine pairings with their food purchases. Customers can learn about each wine’s tasting notes, take a quiz to find their ideal bottle, and get immediate assistance to complete their transaction, all through a natural conversation with the virtual being.
With experience in integrating machine learning pipelines into the software development life cycle, Janvi has designed and built training platforms, data preprocessing tools, and APIs for various models. She sees the future of AI and machine learning in the context of social sciences as an opportunity to address bias, socio-technical harms, and privacy concerns.
As the role of AI in personalized recommendation systems continues to evolve, Janvi predicts a shift towards more sophisticated user-profile building, understanding unstructured behaviors, and increasing personalization capabilities. The future of AI is not only about technological advancements, but also about making a positive impact on society, and adapting the world to make the most of these changes!
