Imagine you're 60 years old and take several prescriptions every day. Blue pill, red pill, a couple of white pills, each at a different hour. Patients, particularly older ones, often fail to take their meds--it's a serious health issue. Wouldn't it be amazing to just snap a picture and have each pill auto-magically identified (to make sure you're taking the right one) and logged (to make sure you're taking it at the right time)?
This is an idea we had at my startup, Iodine. More than an idea, in fact. We actually hired a computer-vision PhD and recruited a couple of other engineers to make this a reality. We spent about a year trying to work this out. But we committed the mistake of being too early. Though we made some progress--it's a really hard challenge--we couldn't get more than 80 percent of the way there. And in software, that last 20 percent takes 80 percent of your effort.
We'd been caught in a swoon--enraptured by a shiny new technology without understanding that, for our firm of just eight or nine people, it was beyond our capacity, tangential to our mission, and distracting to our team. It's a bit embarrassing to admit this, but I'm hardly alone. In Silicon Valley, swooning startups are about as common as Teslas. Here, if you're not in a swoon, you're not really trying.
Everyone is talking about virtual reality and augmented reality and artificial intelligence and deep learning (not using those actual words, but rather "VR" and "AR" and "AI" and "DL"). One accounting puts the number of AI startups in health care, where I compete, at more than 100. AI is everywhere else, too. It seems if you don't have a chatbot strategy in 2017, you're so flip phone.
Why are so many startups in an AI swoon? Timing. It's exceptionally hard to get the timing right, especially for a small company with a ticking clock and shrinking capital. A couple of years ago, all the cool kids were chasing VR--raising money, building prototypes, waiting for the billions to roll in. Today, many of those startups have gone into "cockroach mode" as they wait for some bona fide consumer demand for their toys.
Tech progress usually moves along a path from science to technology to industry to culture. It typically starts with a lab discovery: the semiconductor, an algorithm. From there, it gets turned into a technology--a tool that can be tested and developed. Next, it moves to industrial applications, and finally, once it has consumer utility, it reaches the culture at large. Computers took 40 years to move from lab to home; robotics, though, have so far had only industrial impact. (No, Roomba doesn't count.)
In theory, it's possible for a startup to capitalize on an innovation at any point along this arc, so long as you know what you're gunning for. But for startups chasing this dragon, it's just so easy to be outflanked and outmatched. In our case, we so loved the idea of offering photo identification that we didn't really think through the consequences. Given our size, we could either build a consumer website or a deep-learning company, but not both. And if we wanted to do the latter, we should've not only started with a different team but also chosen different investors, a different business model, and so on. Thankfully, we put the computer-vision project on hold--and then we shelved it altogether.
Shaking off the swoon, we focused on a proven technology called a website. Much less sexy, but it turns out that's where our market is: millions of people who just want better information about their medications, at the right time. A superintelligent, computer-vision technology with machine-learning capabilities? Sounds like a great idea. It's all yours.