In my most recent book, Flight Of The Unicorns, I touched upon the incredible capabilities of one of the most recent innovations in AI: Deep Learning. Unless you have been living in a basement, you are probably already aware of this.
Here’s an excerpt from the book:
Deep learning systems are great at performing three core tasks: classify, cluster and predict. Classification, for instance, can help identify spam emails and send them to spam folder. Prediction algorithms, on the other hand, can help e-commerce companies develop more accurate recommendation engines. Clustering, for example, can help identify similar photograph from massive datasets. Researchers and companies are now trying to build self-driving cars using the technology, for instance in solving the problem of image recognition in real-time so that the car is better aware and can navigate through its surroundings. Deep Learning has given Skype instant translation capabilities. When applied on a robot dog, it was found to walk very much like the living ones.
There is a big worry amongst a lot of thinkers that such a powerful system can replace thousands of jobs, especially the ones that large involve decision making based on pattern recognition (turns out, this is one of the primary abilities of our brain). This fear is not misplaced, as the efficiency improvements this technology brings itself would force companies to adopt it to stay competitive.
India’s terrible doctor-patient ratio
However, there is a big positive impact that this technology can offer: solving India’s perennial scarcity of medical staff. Consider this as a fact: India has one of the worst doctor-patient ratios in the world. Medical Council of India has pointed out, “The total number of doctors in India is much smaller than the official figure and we may have one doctor per 2,000 population, if not more.” A visit any of the well-known hospitals in India presents a scary picture — thousands of patients waiting for the doctor’s time.
A large part of the challenge is the lack of facilities for early warning, prevention and diagnosis. If we just focus on this problem to start with, it really boils down to creating systems that spot erratic patterns in lab test results, something that Deep Learning systems are showing incredible promise in. Furthermore, AI has the ability of processing far more data that learn from the latest research far better than any human doctor.
The solution
An example of a breakthrough in this front is IBM Watson powered cancer diagnostics system.
Human experts at the University of North Carolina School of Medicine tested Watson by having the AI analyze 1,000 cancer diagnoses. In 99 percent of the cases, Watson was able to recommend treatment plans that matched actual suggestions from oncologists. Not only that, but because it can read and digest thousands of documents in minutes, Watson found treatment options human doctors missed in 30 percent of the cases. The AI’s processing power allowed it to take into account all of the research papers or clinical trials that the human oncologists might not have read at the time of diagnosis.
(Read more here.)
For entrepreneurs interested in healthcare domain, there is no better time than now to catch the wave and create some of the early innovations in this space.
Forus Health is one such startup which managed to do so, and I have written about them in my book. Especially interesting is their approach towards creating a portable eye testing device that is battery operated, and can be carried to villages. In the future, if such devices could be attached to AI-driven pattern recognition systems, it could offer treatment advise, which the doctors optionally vet and offer to patients, thereby cutting their time and effort drastically, and also making more informed decision. This could impact lives of millions who go without any access to healthcare.
Another example in lines of IBM Watson is the recent research at the Stanford University and iRhythm Technologies, who have developed a model using CNNs (Convolutional Neural Networks) to detect irregular heart rhythms, and have shown that their model outperforms the cardiologist average. Read more about it here: Cardiologist-Level Arrhythmia Detection With Convolutional Neural Networks, by Pranav Rajpurkar*, Awni Hannun*, Masoumeh Haghpanahi, Codie Bourn, and Andrew Ng.
Is it a good problem to take on?
Consider the economics of such a solution. Typically, e-commerce startups in India have faced the challenge of growth once they penetrated the upper income strata. The growth plateaued, and with that came the realization that a large portion of the country is still below the poverty line, and an even bigger chunk only spends in essentials. Healthcare happens to be one such essential — if one can offer a healthcare solution at a low price point, it has a massive potential target market.
The challenge with such a startup, however, is that it requires investment into R&D, something that seed stage investors in India are generally averse to. The investment landscape in India is still very nascent: while VC firms like Kleiner, Perkins, Caufield & Byers (KPCB) had already started operating in the US from early 1970s, India saw investors enter only within the last ten years, and there is a long way to go before the ecosystem matures. While there are some rare few cases of such investment, generally it is quite challenging to raise funds when the startup doesn’t have a revenue potential for a while. When investment is this risk-averse, it is fatal for most big ideas.
Yet, as with most high-risk, high-return startups, if a startup does manage to harness the technology early on and actually prove that it works on the ground without too much dependency on early investment, the possibilities are endless and only sky is the limit for such a company. And the most important thing is, while making money, it would end up solving a deep rooted problem of the country.
Soum Paul 