It may well turn out, as technologists are already suspecting, that AI makes everything better. But plenty of startup founders are still in the experimental phase of figuring out whether — or maybe how much — machine learning can improve an existing app category.
The label that will probably end up being prefixed here is ‘smart’. Here’s one example: FitGenie, an iOS app whose ex-Georgia Tech co-founders bill it as a “smart calorie counter” — on account of applying machine learning algorithms to simplify nutrition planning for people wanting to achieve a certain weight or fitness goal.
“Our self-adjusting diet algorithm is based on a model we created that maps and forecasts the progress of an individual user and makes intelligent weekly adjustments based on the data we gather,” says co-founder Keith Osayande, explaining how it’s applying AI to calorie counting.
“This model is based on both [original co-founders’] our combined experiences as nutrition coaches and the most current scientific evidence we have available. Rather than just create a simple calorie calculator, we also process all of your data and take all aspects of dieting into account including, body composition, adherence, weight change rate trends, hunger, fatigue, and a handful of other metrics.”
“The key thing to note is that we are not just gathering this data. Our algorithm is constantly assessing how these data points trend with each other and making informed adjustments,” he adds.
Now it’s worth emphasizing that FitGenie’s app is still at an early stage, with the bootstrapping team starting work on their idea in fall 2015, when they were still students at Georgia Tech, and touting an impending 3.0 update — coming in the next “one to two months” — that aims to improve the polish and usability of their freemium offering.
Last year the team took in $20,000 in funding from the Georgia Tech Create-X Startup Summer Program, and they now have some 21,000 users, a year after launch — generating early revenue from those choosing to pay the subscription to access premium features.
“Currently… we’re focusing on the nutritional aspect of people’s journey into fitness,” says Osayande. “So it’s mainly focused on helping people lose weight, gain muscle mass or kind of maintain their current body with whatever physical activities that they’re doing.”
Though he also notes there are obvious opportunities to integrate with fitness trackers down to line to automatically pull in relevant fitness data (vs asking people to manually input physical activity as the app currently does).
FitGenie is aiming to stand out in a crowded space through “simplicity and automation”, says Osayande. So really its idea is that AI will be able to help users adhere to a diet and stay on track with their fitness goals by acting as a de facto pocket nutrition coach.
“A lot of the [existing calorie counting] apps, they take in your data and they just give you graphs or pictures to look at — but they don’t do anything with your data,” he argues. “And that put a lot of strain on the user to actually have to understand their data and make adjustments on their own. But most people don’t have the training to do that and quite frankly don’t have the time or energy to put the effort in to learn that stuff — so we wanted to pretty much do all that for people.”
“The top counters in the market are cluttered, ad-heavy, and difficult to use,” he continues. “Our algorithms emulate the major functions of a personal nutritionist to eliminate much of this guesswork, thus making dieting simpler and less time consuming.
“So [users] can just enter their data, they don’t have to understand it, they don’t have to really know what they’re entering — just put what you’re eating, put your weight and then FitGenie does all the number crunching for you.”
Users of FitGenie do need to do some leg work, however — including inputting their current weight (ideally weekly), and logging the foods they’re eating and any activity they undertake (ideally doing so at least twice per week, says Osayande; more being better).
The app then generates custom daily nutrition targets, based on whatever a user’s desired fitness goal is — such as losing weight, building lean muscle and so on.
“All of that is done by an [AI] model that we created in-house based off of [co-founder and chief adviser] Eric Helms’ research and then testing the algorithm when we were at Georgia Tech when we were at the barbell club,” he adds.
The paid version of the app include additional features, such as meal planning — with FitGenie currently drawing on a database of five million foods to generate its suggestions.
Here it’s applying a genetic algorithm to “create an optimal meal plan that meets the user’s nutrition targets as well as their [dietary] preferences”, as Osayande put it.
Although he also concedes that, currently, the app’s food database skews towards the co-founders own nutritional expertise — which means its assumption is users are meat eaters. So the team has a fair bit of work to do building out the database to ensure the meal planning feature can support users who are vegan or vegetarian, for instance, or who are trying to stick to a specific meat-eating diet such as paleo.
Another feature of the app aims to simplify sticking to a nutrition plan by keeping track of particular foodstuffs users have to hand, via a virtual in-app pantry, with the algorithm then adjusting its meal suggestions to take account of what they do or don’t have in the cupboard.
At this (still early) stage it’s fair to say there’s a lot more potential than perfect execution here — even as the team works on the touted polish coming down the pipe with v3.0.
One area where it can spectacularly fail to alleviate nutrition-related tedium right now is in its approach to food logging — with an over-reliance on scanning barcodes to log what you’re eating.
So while that might work fine if you’re always eating US pre-packaged meals and foodstuffs, anyone who falls outside that — say people who prefer to prepare their own meals using fresh ingredients, or users outside the US — will find they have to manually create food items in the app, including inputting exact levels of nutrients (grams of protein, carbs, fat etc).
This gets very tedious very fast, with apparently no ability to search and select basic foodstuffs from a database to add their nutrient info. (Though there is at least a ‘quick add’ option which lets you create a food and just add a handful of the core nutrients vs the extensive list you have to fill in if you select to ‘create food’).
Finding a way to improve or automate food logging would make for a much more compelling proposition. And indeed Osayande says the team is working on some ideas using either smartphone photos and even potentially Apple’s ARKit to automate this in a future version.
“That’s still a little ways out and we’re still developing to see if we can either take pictures or scan people’s food to measure it that way — but we just got our hands on ARKit so we’re looking into that to see if that’s a possibility,” he adds.
In terms of rivals, existing players in the calorie counting space include apps like MyFitnessPal, Lifesum, LoseIt and Noom but Osayande names the closest competitor as another AI-applying automatic meal planner — called Eat This Much.
“They also do meal plan recommendations through artificial intelligence but their web interface is better than their mobile app interface and they’re more of just a meal planner for a day — while ours is weekly, and they don’t really have an effective calorie counting aspect in their app,” he adds.