“AI has massive benefits for automating and streamlining more tedious, redundant, and time consuming tasks. If we can offload these kinds of tasks to AI, I think humans can then work on more complex things that they are more suited to, which involves critical thinking, and creativity. What this will allow humans to do is to solve problems that they’ve never been able to solve before.”
That’s one of the thoughts from Dr. Vasi Philomin, vice president of generative AI at Amazon Web Services (AWS), on this episode of Shift AI, a show that explores what it takes to thrive and adapt to the changing workplace in the digital age of remote work and AI.
We discuss his experience as a visionary in machine learning and AI, his insights on applying generative AI at Amazon/AWS, and his thoughts on the ethical dimensions and future potential of AI technologies.
The conversation comes in advance of AWS re:Invent next week in Las Vegas, where many of the topics will be in the spotlight.
Listen below, and continue reading for highlights from his comments, edited for context and clarity. Subscribe to Shift AI and hear more episodes at ShiftAIPodcast.com.
Early work experiences: In the 90’s I pursued my PhD in computer science, where I focused on machine learning, when no one was talking about ML or AI. A lot of my family and friends thought I was crazy back then but they think I’m a genius now. I initially started my career as a scientist at Philips Research and then quickly shifted to managing R&D teams to build various innovative products for their consumer lifestyle business. I experienced my first big breakthrough, working on the analog to digital lighting transition for Philips Lighting where I introduced the concept of connected lighting, taking our lighting fixtures and putting cellular chips in them for tracking connectivity and providing a SaaS app for people to remotely manage these fixtures from anywhere. The system was called City Touch and the backend was built on AWS. It allowed cities to control and monitor all of the lighting across their grid. We launched in Los Angeles and scaled globally with a lot of adoption and success.
First paying job : While I was a PhD student at Maryland, I started a summer internship in Germany and went to intern at Daimler Benz where I worked in computer vision. Daimler Benz was interested in autonomous navigation, trying to understand objects that you see in the environment, such as traffic signs and pedestrians. Eventually, I published a research paper that focused on how we can detect these objects and we were early in demonstrating how this could play out live in vehicles.
Family life: My family is full of doctors. My parents are both medical doctors and my brother is a dentist. A lot of our dinner conversations were about my family wanting me to study to become a surgeon, but I was more into things like math and probabilities. It was quite different from the things that my parents and brother would talk and read about, but they were still very supportive of me and I knew I wanted to go into an aspect of computer science that was math focused. AI eventually became one of the things that was on my radar.
Amazon & Generative AI: Before any artificial intelligence integration happened at Amazon, we talked to a lot of our customers about the challenges they face with applying generative AI in their business. Through this discovery we came back with three overarching themes.
- There isn’t a one size fits all solution when it comes to AI models. There are different families of models, and different models work better for different use cases.
- It is important to consider how enterprises differentiate themselves from competitors when AI resources become available to everyone.
- There are use cases where the cost may be much higher than the value Gen-AI provides. The key is how do you take artificial intelligence and cost effectively apply it at scale to real world problems.
Amazon Bedrock: In order to take these technologies and apply it at scale to real world business problems, you have to think end-to-end, applying across the technology stack. This is exactly what we were thinking when we launched Amazon Bedrock.
Three main elements of Amazon Bedrock:
- Amazon Bedrock offers the best foundation models from AI providers such as: AI21 Labs, Amazon, Anthropic, Cohere, Meta and Stability AI.
- Amazon Bedrock has private customization that allows enterprises to differentiate themselves from competitors and securely use their own data and IP.
- For the last few years Amazon has invested in two chips called AWS Trainium and AWS Inferentia, one for training, and one for predictions. These chips are specifically meant for generative AI and offer a better price performance than what is currently out there.
Human and AI congruence: AI has massive benefits for automating and streamlining our work, however you definitely want humans to continue to verify AI processes and check what AI produces. In general, I think more tedious, redundant and time-consuming tasks should be offloaded to AI so that humans can work on more complex tasks, like critical thinking and creativity. It’s all about augmenting human productivity, in terms of what humans are able to do now and what they were not able to do before.
AI for Software Developers: As a developer, my experience with front-end tech highlighted the challenges in keeping up with new programming languages and frameworks. I find myself getting a ton of value from tools we have developed like Amazon CodeWhisperer, which is designed to enhance developer productivity. Through a company-wide productivity challenge, we found that Amazon CodeWhisperer significantly increased efficiency and task completion rates. Recognizing the need for customization, we recently updated Amazon CodeWhisperer to adapt to internal code bases, further boosting its effectiveness in improving developer productivity.
AI Regulation: Amazon is very committed to the responsible development of all AI products being built internally. We’ve introduced a responsible AI framework that covers issues like fairness, explainability, robustness and governance. We also believe in a people-centric approach, making sure there is enough education around AI and ML. It’s very important to define processes within your organization and be transparent about communicating information about an AI system so individuals can make informed choices about their use of AI tools.
Mentors: Azriel Rosenfeld was my advisor during grad school at University of Maryland. He was a rabbi and is widely considered the father of computer vision. What I learned from him was discipline; he was the first one in the office at 7 a.m. and the last one to leave. The second person is Jeffrey Cassis, who was my boss during my time at Philips Lighting. Jeff taught me how to build better relationships. When somebody asked me something, my early instinct would be to say no, “get out of my way and let me do my stuff.” Jeff taught me the notion of “yes, but.” It’s a more inviting way to develop relationships, and to get people together, participating with what you’re doing.
Future of work: Generative AI is about to unleash transformational productivity on a scale we’ve never seen before. We’re truly just scratching the surface and we are still very early. It’s going to be one of the most transformational technologies of our generation, with a chance to surpass the internet in terms of impact. This is going to help us tackle some of humanity’s most challenging problems, augmenting human performance, and maximizing productivity at scale.
Listen to the full episode of Shift AI with Vasi Philomin here.
