How health care startups can protect their AI edge
摘要：Artificial intelligence is an emerging solution for some of the most important problems facing health care today, including medical imaging, clinical decision support, pharmaceuticals, and more. There is clearly significant value in applying AI to medicine, but there are many challenges as well. Who will reap the benefits? How can health care startups make sure their proprietary AI technology is protected? And how do they maintain their AI edge?
Artificial intelligence is an emerging solution for some of the most important problems facing health care today, including medical imaging, clinical decision support, pharmaceuticals, and more. There is clearly significant value in applying AI to medicine, but there are many challenges as well. Who will reap the benefits? How can health care startups make sure their proprietary AI technology is protected? And how do they maintain their AI edge?
The AI space is increasingly crowded, and while that does mean more people to drive innovation forward, it also presents a challenge. Over the past two years alone, AI — deep learning, specifically — has taken a giant leap forward towards democratization. AI algorithms and infrastructure have been opened to the public, and anyone can now audit the best college courses (such as Stanford’s CS231n) covering all the latest AI innovations. With such a swiftly evolving market, unique advantages in AI technology will clearly not last very long.
An innovative AI algorithm nowadays will become obsolete in just a year or two, and employees can easily jump ship with AI secrets (as seems to have happened in the autonomous vehicle industry). That’s why AI-powered companies must continually innovate to stay relevant. Health care startups in the AI domain face not only the challenge of creating valuable products or services for their sector, but of fending off fierce competition in AI too.
Here are four ways health care startups can maintain their AI edge.
1. Gather good data, and lots of it
The basis of all machine learning is copious amounts of quality data. Getting access to the types and amounts of data required to train machine learning algorithms requires close cooperation with the health care providers that generate the data (very few companies currently generate and own medical-grade data, as collecting this data directly from patients has proven extremely difficult). While data sharing is increasingly commonplace, as providers learn both to accumulate data and provide access to outsiders, getting access to data released in the proper manner is still a challenge.
A second enduring problem is access to sufficient annotated data, which is required for most types of machine learning algorithms today. Thus, finding ways to produce copious, cheap, labeled data can represent a huge competitive advantage for AI startups.
2. Find and keep the best people
By far the most valuable resource for creating successful AI-based innovation is people. A strong technical team will make or break a startup’s ability not only to create initial innovation, but to stay ahead of the pack.
In fact, the ubiquitous adoption of AI across industries led to an acute shortage of talent: We simply cannot train data scientists and software engineers quickly enough to meet demand. Add to that the fact that startups generally look for experienced algorithm specialists (as opposed to freshly minted university grads), and it is clear why talent acquisition is a key challenge in developing AI-based products. Good data scientists are scarce, and companies that can bring real talent to the table will see that it makes all the difference.
3. Build out a strong infrastructure
Even if you have access to all the data you need, it still takes time and resources to develop AI for a specific domain. The time to market for a new application is governed not only by the capabilities of your team but, as importantly, by the strength of your AI infrastructure. Having a high-quality computational infrastructure translates directly into velocity, as it enables a company to simultaneously run many machine learning experiments quickly, fostering the release of new applications.
This doesn’t simply mean raw GPU power — which is available to rent on the cloud fairly easily today — but rather, configuring a system that can execute the specific types of calculations required in a specific domain in a super-efficient manner. This still requires a significant amount of know-how, and building such a configuration is no small feat. In short, startups that can boast high-velocity infrastructure capabilities unique to their problem domain can establish a formidable competitive advantage.
4. Negotiate the regulation process
One of the key ways AI changes the game is scale. Good algorithms make it much easier, in turn, to develop next-gen algorithms, exponentially accelerating the machine learning development process. The problem is, regulatory authorities can’t keep up with such a pace. The FDA, for example, insists on re-evaluating any new algorithm that is substantially different than the previous ones. For most applications, this means additional clinical trials for any new clinical indication.
This is a major hurdle for health care companies building future algorithms based on existing underlying technologies. Google Translate, for example, built an engine to translate Spanish to English, but still needed to build an engine to translate Russian. Imagine that between those releases there was a six-month, $300,000 gap. Would we still be able to produce Russian translation applications if that were the case? Health care companies that discover a smart regulatory approach wouldn’t just hold an advantage in terms of go-to-market strategy for current solutions, but a lasting advantage further bolstered by every new solution for which they achieve regulatory approval.
Though health care AI is an exciting space with endless opportunities for improving real lives, the key to startup survival is to discover real, lasting advantages that will guarantee a competitive edge for now, and for the foreseeable future.
Elad Walach is the CEO of Aidoc, a smart radiology company utilizing artificial intelligence to augment the radiology workflow and improve diagnosis efficiency.
Dr. Yoni Goldwasser is principal at Goldmed VC, a venture capital firm.