How gamified AI is being used to teach us about our own tastes.
When we think of artificial intelligence or machine learning, we often think of a robot or algorithm automating a repetitive task that is menial for humans. One of my favorite examples is object detection: identifying pre-trained classes of objects (like people or cars) within an image or video. This application of AI is often seen as competing with human beings; AI which can detect mistakes in manufactured goods can replace the worker who used to perform quality control.
AI as Human Augmenting
Of course there are other applications of AI that augment human experience directly, in ways that were impossible before. The algorithm behind Spotify’s Discover Weekly playlist is one example: the algorithm exposes you to music you’d otherwise never have listened to, enriching your experience and curating your music tastes.
And now there’s an app that’s breaking new ground by using AI to improve human skill, in a domain most of us could use a little help with: cooking.
Chef League, which just launched on the app store, teaches users to cook better by pitting them against AI chefs as they attempt to choose the best ingredient to fix a bad recipe — specifically a recipe scraped from the internet with one star reviews.
As in most successful artificial intelligence implementations, the AI chefs in the Chef League app rely on vast amounts of data, in this case millions of online recipe reviews. The engineers behind the app used natural language processing, sentiment analysis, and clustering algorithms in order to develop eight distinct personas that describe the most common flavor palates.
When I reached out to the app’s founder, Harvard Engineering Professor Beth Altringer, she described the process of building the AI chef personas, comparing it to the strategy behind Spotify’s Discover Weekly: “It took me three years to build the dataset through the Flavor Genome Project, and get to the point of identifying distinctive and predictive reviewer palates. What I mean by this is similar to how a company like The Echo Nest was able to use real review data matched up to song data to identify new ‘genres’ that were entirely data-driven. We have been doing something similar by matching review data to recipes and discovering data-driven palates. Eventually, a subset of these became the AI Chef Coaches that live inside of the Chef League game.” (The Echo Nest is the company Spotify bought to power their Discover Weekly Algorithm.)
Cooking Up a New Way to Learn
Notice how in both Spotify and Chef League, artificial intelligence doesn’t replace human understanding, but rather enables humans to learn in a way that was impossible before. When we think about the way food tastes — the flavors we enjoy, or the unpleasant ones (too salty, too bitter) — we often think of it as a highly subjective experience. However, by processing and clustering the subjective preferences encoded in millions of recipe reviews, Chef League is able to produce new insights (in the form of AI chefs) which would have been impossible for humans to generate on their own. Even if you sat down and read through millions of reviews, there would be no way for you to summarize or use that information in a compelling way. AI has unlocked this information and made it possible to improve in our all too human task of making a tasty meal.
Altringer firmly believes that there is huge potential in applying AI to education, particularly in fields where we may struggle to adequately describe our experience: “I draw a lot of inspiration from what Spotify and Pandora both did, in different ways, for music. They let users find music they love even if they lack the expert vocabulary to find it. You can find something you’ll enjoy via their algorithms even if you don’t know the artist’s name, the song name, the beats per minute, the mood, the album name, the genre, etc. I love that.”
Chef League differs from Spotify and Pandora, however, in that it will actually help users understand their own food tastes better in a way Spotify and Pandora have yet to do explicitly, as Altringer writes, “I wish there were an educational layer to what they built that also helped me understand my music tastes better. Flavor has a similar challenge. People consume the same things over and over, not because they lack a sense of adventure, but because they are limited by the language they know for searching for flavor experiences.”
The Brighter Future of AI
While AI’s economic disruption deserves a healthy skepticism — manual tasks will continue to be automated while new jobs will be created — AI’s ability to unlock previously unattainable insights and augment human potential is incredibly exciting. Even with Chef League, Altringer already has plans to iterate on the app using more data: for example, using food waste data in order to encourage users to become more efficient cooks.
It’s easy to see how this sort of AI training might be applied across other disciplines as well: perhaps a fitness app could determine whether you should focus more on cardio or strength building (of course, it’s probably cardio), or a spending app like Mint could train you how to better manage your finances. Ultimately, though, it will depend first on a pleasant user experience. Talking about her goals for the app, Altringer wrote: “What I hope to do with Chef League is first and foremost to simply make it fun. If that fails, the whole thing fails. If that succeeds, there is so much more a game like this can do.” While Chef League is a first step, it offers a window into a future where gamified AI teaches us more than we ever knew about ourselves.