The message is clear: AI and machine learning are going to change the world, becoming even more ingrained into our daily lives. But so many AI projects are limited in their boundary-pushing and only aim to deliver a one-time solution to an ongoing problem. That kind of patchwork solution might work in the short-term—like how duct tape can temporarily fix a leaky pipe—but it’s typically at the cost of long-term results. As a result, AI success often falls short, with 87 percent of organizations reporting struggles with long deployment times.
I’m setting out to show what applied AI can do to help people in consistent, ever-evolving ways. The goal is to turn AI into a creativity maximizer that’s fair, unbiased, and trustworthy. When you do that, there’s no limit to what you can solve with AI.
The AI Adventures Project
The inspiration for this project came from thinking about what people stress over. I believe that projects should always start with a selfish need—particularly when there’s an audience selfishly interested in that same need. The past couple of years have introduced plenty of stressors into our lives, but there’s a recurring challenge for many of us: getting more physically fit.
The march toward fitness is a common refrain. The most popular New Year’s resolution is to exercise more, which means there’s a wide audience that wants to work toward this goal.
There’s another aim to this project, too. I also want to break through the brittle AI mindset that’s been permeating through this industry.
Why We Need to Stop Embracing Brittle AI
Brittle AI is approaching AI with a narrow mindset. A brittle AI mindset sets out to solve one problem without thinking of how that problem—or the solution—might shift over time. Through a lack of creativity, it hinders AI projects instead of exploring limitless possibilities, and that can prove quite toxic to an organization.
There’s an episode of Rick and Morty where Rick, the genius/mad scientist, quickly builds a robot whose sole job is to pass butter to him for his pancakes. The robot asks what its purpose is, and Rick bluntly tells him, “you pass butter.”
The robot looks at its hands and sighs, despondent over the situation. While this is funny commentary, it highlights a real issue across the tech industry. Far too often, that brittle AI mindset makes its way into our work.
For example, we use AI to scan a person’s brain or lungs and predict cancer merely by what the image shows. That’s not how the healthcare industry operates, though—we need to move beyond that narrow focus and look at all the context around what we see on that image. How is the person feeling? How do they look? What other health issues have they had in the past? What risks are present in their daily lives? We need that context to go beyond solving for a solo, specific task.
Now, that’s not to say we should never have a focus during artificial intelligence projects. While before, our limitations for applications of AI were based on the machinery, now our biggest limitation is our creativity. Experimental AI should be celebrated if it’s on the path to applied AI. After all, 85 percent of AI and machine learning projects fail. But if we only took on projects that had 100 percent success rates, we’d never fully realize the capabilities of AI.
The differentiator is that we need to be using applied AI to think the same way humans think. In fact, I prefer to refer to AI as “augmented intelligence,” because we’re not merely trying to solve one problem. Rather, we’re creating generalized models that can evolve and adapt—just like humans do.
The Defense Advanced Research Projects Agency (DARPA) is overseeing a “Machine Common Sense” program that aims to break through that brittle AI mindset. It’s doing so by studying the learning patterns of infants between zero and 18 months old.
While we probably shouldn’t let the learning patterns of our babies guide us in doing a few dumbbell curls, DARPA is heading in the right direction. It’s seeking to apply AI in a way that’s constantly growing and learning, rather than providing one solution and calling it a day.
Why Use AI for Fitness
No one should expect to reach peak physical fitness through either diet or exercise alone—both need to play a part. But fitness is always frustrating if it takes work. If I have to enter everything I ate in a given day into an app, or if I need to track every rep I did during my workout, I’ll quickly lose steam.
That’s where AI comes in, offering substantial benefits.
Instead of needing a fitness instructor to yell at you, AI will comprehend everything going on around your home at all times, and alert you when it’s time to start moving.
(And if you really do crave that yelling, you can program AI to shout when you’re due for a fitness break.)
Better Security, Better Privacy
Installing 50 cameras in your home? That’s going to cause an uneasy sense of dread, like you’re always being watched. But AI can do all of this work on the edge. You’ll never feel like you’re having your privacy or security invaded.
Going the opposite direction of brittle AI, we’re able to take what we learn from the early days of our AI fitness program and expand upon it. As you begin working out, AI can offer new and improved workout routines, exercises, dietary recommendations, and other suggestions. Those types of actionable insights make it more likely you’ll stick to your fitness goals.
Tapping Into the Human Mindset
When we approach AI as augmented intelligence, we’re producing more generalized models that account for changes in activity and mindset over time. There’s no AI project that can be successful if we don’t put the human brain on trial.
One area I’m intrigued by: the threat of punishment can incentivize us to do activities, particularly if those punishments are applied quickly and regularly.
For example, my kids eat all over our house, which I hate. I’m constantly stepping on Goldfish and Cheerios. That’s a great situation to implement a punishment. If they eat in the TV room or leave a mess in there? No TV. I can program AI to turn it off.
For me, I want to incorporate exercise into a part of my day. In my case, a punishment that both embarrasses me and gives my audience anxiety has the most impact. Something like a distracting noise that would be noticeable on a call, or my Internet connection being wobbly.
Fill in your own consequences with what works for you, but implementing a punishment can be a strategic way to drive toward your ultimate success.
We also do better at sticking to broad, ethereal goals if we set actionable objectives that are higher or better than what we’re currently doing, but not by too much. Don’t say you’re going to complete Mark Wahlberg’s workout routine by the end of the week; walking 500 more steps than you did yesterday is more attainable.
How It All Works
Now that we know why we’re doing all this, how can we make it happen?
Typically, you’d use a webcam and write some code, but this project uses a GoPro because you can do cool stuff with it, it’s waterproof, and I really like the brand. I get a 180-degree wide-angle view, and in real-time I can set the Python code images that are 224×224. The full resolution comes in and there’s a pre-processing step.
We use a video buffer—a cube of images. For instance, if I’ve got 30 seconds of time I can take that and build an image, which represents spatial and temporal motion.
One of the things I loathe is that anytime someone’s doing something cool with video, they always digress to frame-level analysis. That approach falls into the brittle AI mindset for a couple of reasons:
– If I simply lay down and breathe for five minutes with my elbows in a somewhat elevated position, a computer will think I’m doing situps. However, I’m not actually doing any work. I could be playing on my phone, but AI wouldn’t be able to tell that from a frame-level analysis.
– Frame-level analysis also doesn’t take into account other context that we as humans might know to look for. Rarely is an issue so cut and dry; you may move into new exercises, or maybe you’re incorporating more stretching into a workout. Even wearing different clothing could set alerts off if we’re approaching AI with a narrow mindset.
With temporal flow, AI will actually know I’m burning calories. I can’t fool it. That means I have to put in the work—and I’m more likely to get results.
Again, the key to this project is measuring that additional context, something a brittle AI mindset doesn’t consider. Bring in those additional sources like time of day, activity level, and how much of a challenge the workout was. Use the data and apply it to keep the AI firing.
The Path to Applied AI
I can’t wait to unleash AI to achieve my fitness goals and to keep evolving this project over time. Check out a video of AI Everyday in action below and follow along with me on LinkedIn.
Chief AI Evangelist
AI AdventuresAI EverydayAI EverydayarticleArtifical IntelligenceArtificial IntelligenceDataRobotfitnesshealthmachine intelligenceMachine Intelligence