The Bleeding Edge of AI Applications: Amazon re:MARS Conference Recap
AI is the future, and this is what that future looks like according to the experts
There is no denying there is a lot of hype and confusion surrounding the development of Artificial Intelligence (AI), with many players rushing into the space and, in many cases, turning it into a buzzword to make their products sound cool. As a result, most brand marketers, even those working closely with innovation teams, are often confused about what exactly AI means and what the real use cases are.
To cut through the noise, it is important to understand what AI entails and keep up with the real-life use cases that are being developed. For the first part, you can check out our 101 guide to AI for marketers to brush up on terminology and relevant applications. For the second part, we attended Amazon’s re:MARS conference in Las Vegas last week to hear what the brilliant minds working on AI and related subjects have to share. (Check out highlights from our Instagram stories here.)
Amazon has been hosting a private, invite-only conference for a few years to gather the industry insiders and academics working on MARS, which stands for Machine learning, Automation, Robotics, and Space, four interlinked areas of technological innovation that will play a major role in shaping the future of humanity. This year, in true Amazon fashion, the company scaled the event and opened it to the public, resulting in a 4-day forum full of eye-opening presentations and insightful discussions.
Machine learning, a dominant application of AI today, provides the software infrastructure that powers many other innovations, both in the enterprise and consumer-facing domains. “AI is the new electricity,” proclaimed Dr. Andrew Ng, CEO of deeplearning.ai, during the Day 2 keynote session, and he was absolutely right. A dedicated “tech showcase” ballroom was set up in tandem with the event, and every vendor (most of them Amazon partners) showcasing their work at re:MARS, from the smart home devices powered by Alexa to the automatic lawn mower demoed by iRobot, was using AI to power their products and services.
Automation is an end result of machine learning, which, by definition, enables machines and algorithms to learn from data by themselves and, with the right setup, to take actions based on those learnings. In the same vein, robotics is an important application of AI, and represents a tangible manifestation of software-based automation, allowing “dumb” machines to come to life with various sensors as their eyes and ears and algorithmic brains allowing them to react to the environment.
Space, then, represents the ultimate frontier for AI and robotic applications. As the most sophisticated and difficult area to apply our learnings on AI, space tech is a highly aspirational territory that inspires the public and attracts the crème de la crème of AI talent. Thus, it should be no surprise that Amazon is including space science as a key theme of the event.
Being the sole organizer, the presence of Amazon was understandably strong throughout the event. Some presentations were clearly aimed at educating developers on how to use Amazon’s machine learning tools, and quite a few individual sessions were dedicated to showcasing Alexa and Amazon Go’s automated checkout system. Thankfully, there were just as many sessions that set their sights beyond Amazon to ensure this event is not just one giant sales pitch for AWS. Nevertheless, it is interesting to hear just how influential Amazon has become in the MARS fields and how they are quickly becoming the infrastructure owner for the cutting-edge AI applications.
Sitting through a wide variety of presentations — ranging from the razzle-dazzle of an opening keynote where executives from Boston Dynamics and Walt Disney Imagineering showcased their exciting achievements in robotics before bringing A-list star Robert Downey Jr. on stage to talk about his robots-powered environmental imitative, to the more intimate breakout sessions where developers and academics got down to the technical details of Amazon’s AI tools or debate the ethics of AI — I learned about the latest and newest in AI applications in various domains. Here are some of the key takeaways from each of the key domains.
Machine Learning: Going Beyond Software
If AI is the engine that powers the fourth industrial revolution, then machine learning is the 8-cylinder that most companies will use to apply AI to their businesses today. In other words, AI is moving beyond software and is being applied to various facets of businesses, be it managing logistics, projecting product demands, or building tools to serve developers and customers. From tools like AWS SageMaker, which allows developers to build, train, and deploy machine learning models quickly, to voice assistants like Alexa or Google Assistant that puts natural language processing (NLP) and automated speech recognition (ASR), two crucial features powered by machine learning, front and center, machine learning is a tool that all companies need to become accustomed to and deploy effectively both in the front and back end of their businesses in order to remain competitive.
In order to do so, companies should consider following the three steps that Dr. Andrew Ng suggested. One, companies should start small. AI and machine learning have a wide range of applications, so the first step is to narrow down what are the specific issues that your company needs to address, whether it’s a more accurate sales projection or an automated customer service tool, and start from there. Two, companies should think about automating tasks, not jobs. Each job consists of numerous smaller tasks, and it’s far more productive to think about using machine learning to handle some of the manual, repetitive tasks that could be automated, rather than trying to find a wholesale AI solution that would just do a job entirely by itself — the technology simply is not there yet. Last, companies should combine machine learning with their subject matter expertise, where they have the most data, to pinpoint the unique intersection that they can improve by applying machine learning tools.
In other words, AI tools should be built to empower humans, both as employees and customers, not to replace them — a sentiment echoed by many speakers at various breakout sessions of the event. Daphne Koller, CEO of insitro, used her stage time to discuss how they ethically collect and apply a range of very large data sets to train AI models to help address key problems in the drug discovery and development. Similarly, Andrew Lo, Director of the MIT Laboratory for Financial Engineering, took attendees through the complicated process of his team using machine learning to visualize investment portfolios and explore new ways to raise the “sharpe ratio” for all. Throughout the event, machine learning was evoked in almost every session, big or small, as the fundamental source of innovations across various industry verticals and academic domains. To most re:MARS attendees, AI and machine learning are not just mere buzzwords, and they have come together to demystify them through real-life applications.
Automation: Make AI Do It
Automation is a central application of AI that is already decades in development. By teaching the machine to spot patterns and learn by itself, we are on a path towards automation that will allow AI to do the work for us. For example, CEO and co-founder of June, Matt Van Horn led an interesting hour-long session showcasing the design of the Alexa-enabled smart oven his company created. Walking attendees through the trials and errors his team had to go through in order to make an oven that can automatically detect the food item you put in and intelligently suggest the right way to cook it, the June oven is nothing short of a kitchen revolution in the making. But because different people have different preferences for cooking the same ingredients — for example, thin, crispy bacon or thick, chewy bacon — personalization is also taken into account to ensure that the oven will only serve up the methods and recipes that cater to individual users.
Beyond smart home devices, which are proliferating across millions of U.S. households and automating many mundane tasks in our daily lives, most of the discussion on automation inevitably centered around the impending arrival of autonomous vehicles (AV). In the tech showcase space, Amazon set up a mini-race track to showcase AWS DeepRacer, a fully autonomous 1/18th scale race car built with Amazon’s machine learning tools and backed by a global racing league. During separate keynotes, executives from Aurora and Zoox, both AV startups that Amazon partners with, presented their vision of leveraging the latest advancements in AI to make more precise detection of road conditions, make more accurate predictions of agent movements, and even create more realistic simulated testing environments. Designing AV from the ground up for AI, and not for human drivers, may be the key to building the future of transportation, but there is still a lot to explore and learn in terms of the longstanding impacts of self-driving cars on our future.
In that regard, a session led by Matthew Johnson-Roberson, an Associate Professor from the University of Michigan (a leading institution in AV research and development and the site of MCity) acutely pinpointed the many key issues in the AV industry today. While most automakers are eagerly partnering up with startups to outfitting their vehicles with lots of sensors to solve the long tail of random challenges caused by unpredictable actions of human drivers and pedestrians, it is just as important to remember that AV development does not happen in a vacuum. Automakers and AV startups need to work together with legislators and urban planners to work out the various issues surrounding AV deployment and minimize its disruptive impact on the economy and our urban environment.
Robotics: From Labs To Factories & Homes
To a certain degree, self-driving vehicles are basically robots on wheels, even though most people don’t usually think of autonomous vehicles as robotics. That being said, the exciting field of robotics on display at re:MARS extended far beyond robo-cars and shows a wide variety of commercial and non-commercial applications. Even though most of them function more like show ponies today with a long way to go towards mass production and monetization, there is little doubt that what was on show will become everyday helpers that coexist with humans.
Thanks to advancements in AI-powered automation, the robotics field has seen tremendous growth in the past couple of years. Beyond the dazzling demos that Boston Dynamics and Walt Disney Imagineering showcased during the opening keynote, there were also many smaller robotic startups that came to showcase their latest achievements. On the enterprise front, Plus One Robotics applied the collaborative principle to their design of industrial robots to create co-bots that are meant to work alongside each other and human worker. Ken Goldberg from Ambidextrous Robotics also showcased the latest developments in robotic grasping, which, despite being an easy task for humans, is in fact a rather difficult thing for robots to master. RSE robotics presented their remote-controlled underwater robot designed to detect lionfish — an invasive species causing havoc in the Atlantic ocean — and capture them to turn them into seafood ingredients to be sold at Whole Foods.
Increasingly, robots are starting to become viable consumer products, as some companies gleefully demonstrated at re:MARS. For example, iRobot, maker of the Roomba, demonstrated their latest home robots including an automated lawn mower and explained their approach to system intelligence in smart home. RRC Robotics explained how they used AWS RoboMaker, Amazon’s cloud robotics service, to test and improve LEA, an automated walker they created for the elderly, and how they use RoboMaker to deploy intelligent applications at scale instead of building those algorithms by themselves.
As robots start to come out of factories and warehouses and start to enter the consumer market, another dimension of the discourse around robotics centers around the future of human-robot relationships. In her section of the Day 1 keynote, Kate Darling, a researcher at the MIT Media Lab, examined our evolving attitude towards service robots and presented her team’s research. According to their findings, our innate instinct to anthropomorphize any moving objects will likely prompt most people that come into close and frequent contact with robots to develop empathy and even emotional bond with robots. This shift in our relationships to robots may just be the key catalyst we need for mass consumer robots deployment.
Throughout the event, it was clear that development in robotics is highly dependent on the latest breakthroughs in AI and machine learning, as much as in the supply chain of advanced sensors. As Erik Nieves, Founder of Plus One Robotics candidly put it, most robots companies today are simply too small to scale, and therefore have to wait to ride the coat-tails of other AI developments such as computer vision and cloud robotics services, usually backed by the big tech firms. It is interesting to consider the brand opportunities for other deep-pocketed companies to invest in robotics startups to help them scale and develop industry-specific robots.
Space: The Aspirational Frontier
Looking beyond the automated tools and funny robots, space remains the ultimate frontier for AI applications. While this may be a bit “out there” for most attendees at re:MARS, it is easy to see why Amazon singled out “space” as a separate theme for the event. As an awe-inspiring subject and the cutting edge of science and engineering, space technology is something that many scientists and developers currently working in AI would find fascinating and exciting. Therefore, by including space technologies, especially those built for Mars exploration, Amazon found a great way to emphasize the potential impact of AI applications beyond our planet and, from a marketing perspective, lends a sheen of coolness to the event itself.
Through the conference, the space-themed sessions are usually the ones that have the most enthusiastic audience, eagerly soaking up all the grand theories and interesting tidbits presented by space tech companies. Blue Origin, a space rocket company owned by Bezos, elaborated on why they are working on reducing the cost of space exploration by developing reusable launch rockets and demonstrated their latest testing progress. Jet Propulsion Laboratory explained their efforts to democratize Mars and space exploration by crowdsourcing projects such as building the next-gen Mars Rover. The Aerospace Corporation unveiled their roadmap for 4 generations of modular robotic satellites that will autonomously explore beyond the solar system, while Lockheed Martin Space plans to build a network of hive-like cloud servers in space for processing the AI tasks in space exploration, instead of relying on servers on the ground to process the data transferred back from space.
Beyond the application of AI and robotics, another through-line connecting all the space-themed sessions is a shared sense of anxiety regarding the worsening ecological conditions on earth and pressing environmental issues such as climate change, which injects a newfound urgency into space exploration as we look up to the sky for a potential backup habitat for the future of humanity. Ironically, some AI tools being developed for space satellites will also be helpful to track weather patterns and sea level changes, giving environmental scientists more accurate data to conduct their research. At the end of the day, putting AI to use in space may just end up helping us better monitor and understand the earth.
It is abundantly clear from the re:MARS event that AI will be a disruptive force for all industries as real-life applications start to move beyond software and start to manifest in physical, tangible ways, whether its from automated devices and vehicles or robots. Artificial intelligence will soon be involved in nearly every aspect of business in one form or another. In response, all brands will need to develop an AI strategy that identifies the opportunities and challenges that various AI applications will bring.
But as important as it is to ponder all the changes and innovations that AI will unlock for us, it is just as crucial for brands to understand the things that will not change any time soon for their customers. When Jeff Bezos asked to share his ten-year predictions on tech innovations during his time on stage, the Amazon CEO namechecked robotic grasping and biotechnology before pivoting to talk about the need to focus on the things that will not change in a ten-year time frame. He suggested brands to “look at what’s stable in time, and continue to focus there. Identify those ideas, they’re usually customer needs.” Devising a brand strategy centered around solving those relatively stable and fundamental customer needs, be it fast delivery, low cost, or good customer service, can create opportunities to develop flywheels around those questions and apply the suitable AI tools to solve them.