Many people wonder why I spend time on Twitter. For those wondering the value of Twitter, you have to look no further than this recent Twitter thread from Lisa Bari (@LisaBari). For those of you afraid of Twitter, I’ve collated her tweetstorm into a nice thread below for easy reading. Plus, Lisa’s thoughts on AI in healthcare were so spot on that I wanted to share them with the Healthcare IT Today community.
Enjoy this look into the future of AI in healthcare from Lisa Bari:
THREAD: This year, I spend a lot of time reading, thinking about, and discussing the future of artificial intelligence in health care. I want to share some of my impressions, recommendations for non-technical learning in the space, and hopes for the future. 1/
Quickly, why me: I studied cognitive science 20 years ago @UCBerkeley, which included survey courses on AI, but before important developments in deep learning/neural networks. Now I work in health policy, focusing on health IT, interoperability, and health data exchange. 2/
I’m going to cite @KaiFuLee’s excellent book, “AI Superpowers”, which explores progress in AI so far, areas in which China/US have the advantage, and future considerations–jobs, threats, and humanity itself. I strongly recommend it, even/especially for non-technical people. 3/
In brief, Lee says we are in the *implementation* phase of deep learning, a groundbreaking innovation in AI that allows computers to “think” and learn like humans, with simple rules and lots of training data. Deep learning originated less than a decade ago from @DeepMindAI. 4/
The technology and computing power already exists to implement deep learning across health care, quickly. Early successes in applications for medical imaging, like diagnosing diabetic retinopathy and identifying tumors have been published by @GoogleAI and others. 5/
One of the other immediate implementations of deep learning in health care is in natural language processing (NLP), which is being applied to unstructured data like clinical trial reports and EHR notes by @Amazon Comprehend Medical, and others.6/
It’s hard to discuss AI in health care without mentioning the hype curve. Health care is very susceptible to hubris, and it’s fair to say that AI is not yet up to the task of generalized clinical diagnosis. However, the limitation may simply be the types of data sets we have. 7/
What limits the types of data sets we have available? 1) Massive amounts of siloed data due to a lack of health IT interoperability; 2) Security and privacy laws that do not meet the needs of modern people and information systems 3) No national health care data/AI strategy. 8/
We are working on health IT interoperability, and even without a coherent national strategy, more data will become available. @KaiFuLee puts “doctor” in the category of jobs that will be affected by AI, but not fully replaced. 9/
This could be a good thing, if managed correctly. You don’t have to look far to find valid criticism of the current medical education system
Assessments like Step 1 are driving learning in the wrong direction, not only for our learners, but for our patients.
Our patients don’t need human computers. They need humans. Humans who have not been locked in a room memorizing useless facts for years so they can ace tests. –source
“Our patients don’t need human computers…” 10/
AI *will* match and eventually exceed humans in diagnosing disease, recommending personalized treatments, and curing illness before it starts. These advancements are coming soon. This is a good thing. 11/
There’s a lot more here, of course, including how the US can stay competitive in AI–we have an advantage, but it won’t last for long if we don’t take it seriously. I’d like to see an immediate effort to develop a national (health) data strategy, and a Chief AI Officer. 12/
Part of this effort must include a national conversation about the ethics of further developments in AI, including the increased collection and use of data, in health care, law enforcement, consumer behavior, and more. But ignoring it is no longer an option. 13/
Health policy leaders must incorporate interoperability, health data, and AI into their recommendations and decisions, starting yesterday. Health systems and insurers without an integrated AI strategy are already behind. 14/
And yet, quoting @KaiFuLee: “For all of AI’s astounding capabilities, the only thing humans can provide turns out to also be exactly what is most needed in our lives: love.” Doctors will need to find a path forward working *with* AI to deliver excellent care. 15/
Advancements in AI aren’t a panacea for all that ails us, including climate change, spiraling debt, division in our society – but they will help, if we approach it with intention, wisdom, and…love. Technology can, and has already, lead to a better life for many. 16/
I’m so grateful this year for the opportunity to learn from some of the world’s best thinkers, on the internet, which has changed my life. Happy (Gregorian) New Year, here’s to a transformative 2019, may we do our best, no more, no less. 17/fin
Great presentation by @math_rachel, cofounder of @fastdotai, which teaches deep learning to newbies. Dr. Thomas’ talk highlights the issues that a lack of participation and equity in AI research is causing, and how “unlikely” people must get involved.