One of the necessary conditions for AI to go mainstream was a wealth of training data. With the advent of mainstream Internet adoption, and a subsequent explosion of publicly available data, this condition was met about a decade ago — and enabled huge advances in ANNs, computer vision, and predictive analytics. However, there are still fields and use cases where AI companies are struggling to find enough data to train their algorithms. One such area is health, where patient privacy laws restrict the access that tech companies have to training data, thereby making advances in the field difficult to achieve.
NewtonX conducted a two part survey, the first part with 100 executives at the health verticals of AI/tech companies, and the second with 100 patient data privacy experts, including the heads of patient data at large health insurance companies and at major US hospitals. The data and insights in this article are informed by the results of this survey.
Privacy Hacks: How Tech Companies Are Getting Access to Patient Data
Recently, the US Department of Veteran Affairs (VA) partnered with DeepMind, offering the company access to 700,000 medical records from US veterans over a 10-year period. The partnership was developed with the goal of being able to predict those who are most at risk of acute kidney injury using AI. The VA encrypted and sanitized the records, and maintains full access control to them.
This partnership is a promising one for DeepMind and for advances in AI as a predictive healthcare tool. The VA has millions of health records, and represents one of the most comprehensive health data repositories in the US.
Thus far, Google/DeepMind has struggled to gain access to such quantities of patient data. In November of 2018, DeepMind turned over its work on a health care app for hospital staff, called Streams, to its parent company, Google. The move sparked outrage from the public over concerns that Google, a profit-driven advertising company, could use sensitive patient data, including HIV status, to deliver targeted advertising. Additionally, Google/DeepMind has faced barriers in the EU, where GDPR limits what data the company can access, and how they can use it without user consent.
However, if DeepMind is able to solidify trust over its data usage through partnerships such as the one with the VA it may come to own the data vertical of healthcare tech. This would be a massive triumph for the company.
So far, while the largest tech companies, including Apple, Microsoft, and Amazon have all made healthcare plays, each of them has played to their strengths: Apple has focused on its consumer products, Amazon on its e-commerce platform, and Microsoft on its storage and analytics services. Google has focused on data, the hardest vertical of healthcare to access. The partnership with the VA is a promising sign for the company, but they will likely face significant barriers for any medical advances relating to highly sensitive patient data.
The Privacy Problem: Why Protecting Patient Data is Bad for AI
China’s formidable rise in AI development has been largely the product of unregulated privacy policies. Because the country has ample data for machine-learning applications, Chinese companies have been able to create highly sophisticated computer vision and analytics AI systems.
In the U.S. and Europe, where privacy laws are much more stringent, gaining access to sensitive data and integrating it with other potentially identifying data is a risky proposition. Indeed, Google was recently slammed with a $57M GDPR fine — and it’s likely that this won’t be the last time the company gets in trouble with European regulations.
That said, many patient privacy laws were written before the advent of machine learning and AI. The potential for this technology to lower healthcare costs and effectively leverage preventative care may propel lawmakers to revise and update regulation and ethics standards.