The next generation of digital health offerings will use machine learning to treat patients more effectively, but it's still a decade away, according to Vinod Khosla, founder of Khosla Ventures, who spoke at Partners HealthCare's Connected Health Symposium in Boston this week.

Khosla said no one can deny medicine has improved over the past 50 years, but it could be much better. One area ripe for improvement is error rates.

"The error rates in medicine, if you look at the Institute of Medicine studies, are about the same as if Google’s self driving car was allowed to drive if it only had one accident per week," Khosla said. "I’m not talking about the best doctors, I’m talking about the median doctor or the below median doctors -- the bottom 50 percent of doctors."

The technology being implemented today, according to Khosla, is simplistic, but serves a very useful purpose.

"There’s no technology in telemedicine, or very little, from my point of view, because there’s no new data," Khosla said. "It’s the same doctors making same subjective judgement. Instead of you sitting in their office, they’re doing it remotely."

He explained that a symptom navigator tool as it exists today helps patients who have the common cold or another similar illness, rule out something like strep throat, but the next generation of this tool should be able to address more serious concerns.

"You are diagnosing a condition, you are talking to a patient, you ask a bunch of questions, very typical," he said. "Now imagine a rare condition that is very serious... If you don’t ask two questions to exclude that rare condition -- and you may not have seen it because it happens one out of 10,000 times -- you are not going to remember to ask as the doctor. So the first generation of systems might actually say, 'Here’s additional questions to exclude this other thing.'"

Clinical decision support tools, Khosla argues, will allow healthcare professionals to do more than they could before and take on different roles when caring for the patients. For example, the next generation of digital health tools may let a skilled nurse do what a primary care physician does and then perhaps allow a primary care physician to do what a cancer specialist or cardiologist does.

"I think the first generation of systems in the next decade will be about assistance," Khosla said. "If you are a heart patient and something happens at home, you have a little AliveCor device that tells you, 'Hey, this is actually atrial fibrillation happening, not just heart burn.' And you won’t have to decide in a blind way whether you should rush to emergency or not. Those kinds of systems will happen."

Over time, Khosla said the systems will start answering more advanced questions, like ‘Do I need to rush to [the emergency room] or is there something else wrong?’.  But to get there, digital health offerings need a lot of data.

"AliveCor has an algorithm, and because they collect 1 million EKGs a quarter, they now have FDA approval to diagnose atrial fibrillation when you take an ECG..." he said. "Machine learning comes only after you have a lot of data, you have lot of data only after you do a lot of monitoring and use of these devices."

And in the future, Khosla believes doctors will be able to move from the practice of medicine, which he said is based on cumulative experience and research, to what he calls the science of medicine.

"My hope is in 10 or 20 years we go from the practice of medicine, which doctors practice today to the science of medicine," he said. "[A patient] might... read an ECG the same way that a doctor would today and then telemedicine would present that remotely for a doctor to look at or to diagnose it. But if you can find new kinds of data that says you're about to have a heart attack in six hours, and predict it, you have invented new science and that’s much more exciting."