By Samantha Walravens
Archie Agrawal, Product Manager at Amazon, at Carnegie Mellon
Agrawal
In 2018, Elon Musk tried to replace humans with robots in his Tesla factories, to disastrous results. In his apology, he tweeted, “Excessive automation at Tesla was a mistake. To be precise, my mistake. Humans are underrated.”
Truth be told, we are far from a time when machines will replace humans. But, we are in the thick of a world where humans and machines are work together to solve problems. For example, self-driving cars get smarter the more they observe the actions of human drivers; smart devices get smarter as they hear more voice commands; and search engines get more accurate results when they observe which sites people actually click on for each search term.
In this interview, we talk to Amazon Product Lead Archie Agrawal, who is leading the charge in “Human-in-the-loop” machine learning — a process that optimizes the interaction between machine algorithms and the humans who create the data to power those algorithms.
Samantha Walravens: What is your role at Amazon?
Archie Agrawal: As a Product Lead at Amazon, I work on making Amazon’s online store experience safe, delightful and in-line with customers’ expectations – our top priority. This involves handling user-generated content in our online store via Human-in-the-loop machine learning.
Walravens: Can you explain Human-in-the-loop (HITL) machine learning?
Agrawal: Machine learning (ML), as we all know it, relies on huge amounts of quality data and learns from human feedback. The Human-in-the-loop concept leverages both human and machine intelligence in tandem to manage the shortcomings of traditional ML. For example, in the case of scarce data, HITL expedites the labeling of tricky or novel data that a machine can’t process with confidence, thus reducing the potential for data-related errors or biases.
Walravens: How does bias creep into AI (Artificial Intelligence) systems?
Agrawal: Neither humans nor machines are completely free of biases. It’s a well known fact that machine learning models emulate the bias of the creator. In spite of the best of the intentions on behalf of the creators, bias inadvertently creeps in. Sometimes the biases arise due to second order effects, which are even hard to catch and measure.
To build a fair system, we need to tackle biases in both AI and humans. Humans and machines need to work in tandem and augment each other to put this problem behind us. Just as humans are needed to manage bias in AI systems, AI can also work with humans on reducing their biases. AI systems can remove subjectivity arising due to irrelevant inputs, like gender, ethnicity and income. Humans can ensure that diverse viewpoints are represented, provide judgment and societal context, and help formulate the right public policies around technology and AI.
Walravens: Will there come a time when machines and Artificial Intelligence replace humans?
Agrawal: As of 2021, pure AI is science fiction. For the foreseeable future, humans are going to be involved in different steps of machine learning, including dataset creation, algorithm development and final decision making in scenarios where the complexity of the decisions exceeds statistical encoding the inputs or when domain expertise is required. As we start to narrow down the domain of the problems at hand, humans will slowly start to feel comfortable delegating additional responsibilities to AI/ML systems. Pragmatic AI has already taken over our life in various ways to varying degrees. We see it in autopilot systems that control everything from takeoff to landing in modern aircraft. Every single credit card transaction around the world goes through comprehensive trust and risk models. In addition, you can now walk into autonomous checkout stores, grab a can of soda and walk out.
Walravens: What is the key to Amazon’s success?
Agrawal: The key to our success at Amazon is that we continue to innovate on behalf of our customers and build solutions that meet their needs— whether it’s delivering packages in two hours, enabling entrepreneurs, small businesses or content creators to grow their businesses, or creating great shows to watch on a weekend. Every idea starts from the customer and Amazonians work backwards from that to create that reality. The working backwards process is how we ensure customer obsession and build the right thing for them. Amazon also embraces what we call the “Day 1” culture, which means our approach remains the same as it was on Amazon’s very first day – striving to stay curious, lean and nimble, all while focusing on our customers.
Walravens: What is the key to your success?
Agrawal: The single most important factor in my success has been the gift of being curious and having a deep sense of perseverance and grit. My Bachelor’s was from a no-name engineering college in India which didn’t attract the famous names in technology for recruitment. At the time, getting an interview call from the top technology companies felt like a far fetched dream. I figured that a Master’s degree could help cover the knowledge and opportunity gap. I was accepted into a Master’s program at Carnegie Mellon, and I’ve never looked back ever since.
Disclaimer: The thoughts expressed are those of the subject and not of the employer.
