Those who run regularly or who have experienced the endorphin euphoria known as "runner's high", can experience the same heady feeling reading Joanna Goodman's "Robots in Law: How Artificial Intelligence is Transforming Legal Services" (Ark, 2016). The hope was that like Nike running apps, "Robots" would provide her with the tools and insights she needed to understand the AI legal tech hype, and intelligently speak to the topic with fellow colleagues in legal innovation. While the Twitterati debate gets granular rather quickly with varying definitions of the terms artificial intelligence, lawyer and robot, the takeaway is consistent with the "and" versus "or" conundrum – none of, robot (traditional AI), lawyer (human intelligence), or a robot lawyer (augmented AI) provide a perfect solution or path forward. Goodman showcases LISA as an example of AI augmentation since the tool leaves more complex issues to human lawyers, but misses the opportunity to explore further with insights on the critical limitations of the App.
So, imagine a runner, let's call her Ann, who has previously run the London marathon in a time of 4 hours 13 minutes (253 minutes), with a given pacing profile; that is, a given set of split-paces. This is a common machine learning technique, based on the intuition that, to solve some new problem (predicting a PB time for Ann) we should look to similar problems in the past (those runners who ran similar non-PB times to Ann) and use their solutions (subsequent PB times) as the basis of a prediction for Ann. In the case of our marathon PB prediction task, if we have lots and lots of cases, covering male and female runners, of all ages and experience levels, and representing a wide range of finish-times, then we have a good chance of being able to find similar runners to act as a basis of a prediction for any given runner whose PB we wish to predict. A key idea in case-based reasoning concerns how we determine the similarity between a new situation (Ann's recent marathon race) and a past problem in some case (Sarah's non-PB race).
I had been testing out Vi, a set of $249 Bluetooth running headphones with its own built-in AI assistant and biometric tracking features. After a convoluted series of events in which I was offered a potentially illegal entry to the Brooklyn Half Marathon a week before the race, I found my adventure: I decided to run my own 13.1 miles in the Prospect Park Loop with nothing but the AI headphones to guide me, using Vi for a crash training course to prep in less than a week. Vi doesn't offer much more than other running apps I've used: It tracks the distance you run, measures your heart rate, and offers some realtime coaching direction to fine-tune your step rate to find your ideal pace, which it calls your "Comfort Zone," -- but it leaves much to be desired as a next-gen personal trainer. It currently has no dedicated feature to set specific goals, so users prepping for races like me have no guide to train for big events or set more defined goals than just fine-tuning their running style.
"It's a really versatile tool," said George Robusti, senior design director of global running at Adidas, of the ARAMIS system. As we sat and talked inside Adidas' headquarters in Portland, Ore., I asked Robusti how AlphaBounce compares to the Ultra Boost and NMD, two of the company's most popular runner lines. During a demo of the sneaker, the team behind AlphaBounce compared its blend of materials and design techniques to Apple's signature approach: seamless integration between hardware and software. "In the past, we've always used off-the-shelf materials or processes that have existed," said Andy Barr, Adidas' category director of global running.