Ashish Vaswani, Co-founder, Essential AI Image: Courtesy Essential AI
For as long as he can remember, Ashish Vaswani has been fascinated by science and mathematics. The son of an architect and a doctor, he grew up in Oman before his family shifted to Nagpur when he was 15. “I am not able to recall if it was then or later, but you pick up your heroes. I was a big fan of a lot of Indian scientists like Bose (Jagdish Chandra Bose) and CV Raman,” Vaswani says. Equally fascinating to him, while in school, was the story of how Bill Gates and Paul Allen founded Microsoft. That, in many ways, led Vaswani into the world of computers.
After he finished school in Nagpur, Vaswani went on to study at the Birla Institute of Technology, Mesra, where he graduated in computer science and engineering. A life in the IT sector as a computer engineer followed, and he spent a few years in Delhi before realising that his true calling was in the world of science.
“I enjoyed algorithms, early theory of computation and was also interested in data structures,” Vaswani says. That prompted him to leave his job and move to the University of Southern California where he completed his master’s before going on to do his PhD in computer sciences. “I was doing my PhD in natural language,” Vaswani says. “My lab was not doing deep learning, but I could sense this was the one. Second, I thought this was going to be where breakthroughs are going to come from.”
His work on natural language models took him to Google where he worked for more than six years on the Google Brain project. Google Brain was a deep learning artificial intelligence (AI) research team that served as the sole AI branch of Google before being merged with Google DeepMind. It was at Google that Vaswani and his then-colleague, Niki Parmar, became the first and third authors of Google’s 2017 research paper ‘Attention Is All You Need’, which introduced the “transformer” deep learning architecture. That architecture was primarily responsible for the chatbot, ChatGPT, and in many ways reshaped the AI race globally and remains a foundational paper in the AI boom.
“At Google, I could both contribute to and see universal computational principles, which I think I enjoy, and the transformer is very much like that because its impact has transcendent research and it’s emerging as a universal way of even learning how software operates,” Vaswani says.
By 2021, Vaswani left Google to found Adept, an enterprise AI tool that focussed on training a neural network to perform tasks. Vaswani was joined by former Google director David Luan and Parmar. So far, Adept has raised $415 million from investors, including General Catalyst. Vaswani and Parmar left Adept in 2023 to set up Essential AI.
Essential AI, Vaswani says, focuses on solving humanity’s enormous challenges through powerful AIs that can mimic and amplify the ability to solve unseen problems. “We are making fundamental advancements in research by building on capabilities across model training, cluster management, scaling laws, evals and data pipelines,” Vaswani says. “The transformations in AI have been referred to as an industrial revolution. I think it might even be an enlightenment.”
Essential AI has so far raised funding of $64.5 million over two rounds and includes investors such as Thrive Capital, Google, Franklin Templeton Investments and Nvidia. “AI is the most important tool that humanity is going to use to solve our hardest problems, and I believe that this tool should be accessible to everyone that wants to use it to solve their problems,” Vaswani says. “So, it should be open.”
Essential AI, Vaswani says, is building an open science frontier model, a research approach that focusses on sharing and openness in all stages of the scientific process. “We want to build frontier models in certain areas and certain application areas that we care about and, by revealing this process, I believe others will be able to pick it and contribute more breakthroughs.”
Today, Vaswani says that most of the leading advances in the world of AI happen behind closed doors, and Essential AI, he reckons, will help build an ecosystem that will also partner with companies while solving problems for others. Already, a recent report by the United Nations says that AI is on course to become a $4.8 trillion global market by 2033, but requires urgent action and intervention to ensure that the benefits do not remain in the hands of a privileged few.
“There are groups or pockets among the AI community that see open science as a danger and a risk,” Vaswani says. “I see it as an opportunity to identify, expose and remedy some of the vulnerabilities which will help us understand better. I’m an optimist and I believe that ultimately people will appreciate good work.”