Although every technology product released today seems as if it is “powered by artificial intelligence,” the actual AI revolution is ahead of us. When it arrives, it will be on par with the industrial revolution in changing our lives, especially in the world of medicine.
John McCarthy, a legendary computer scientist, coined the term “AI” in 1955. He defined it as “the science and engineering of making intelligent machines, especially intelligent computer programs. It is related to the similar task of using computers to understand human intelligence, but AI does not have to confine itself to methods that are biologically observable.”
There are two broad types of AI: general and narrow.
Think of general AI as replacing humans (something that is a long way off—think science fiction such as Westworld or Ex Machina), while narrow AI automates some of our tasks (think Siri or Cortana), something that has been happening for years and is set to change entire industries such transportation.
Machine learning and healthcare
Machine learning is a subset of artificial intelligence. A program is said to have “learned” if it can improve its performance on a task from past experience or data without being explicitly programmed to do so. The standard computer program has hard-coded conditions and responses. For example, a tic-tac-toe program might be programmed to place its X in front of its opponent’s two Os, or in the center spot if it’s available. However, a tic-tac-toe algorithm that uses machine learning may start off with just knowing the basic rules of tic-tac-toe, and then determine its own strategies to win.
Machine learning is poised to revolutionize medicine and healthcare delivery. Crowdsourcing healthcare data will also accelerate the incorporation of machine learning into healthcare, even improving the speed at which research is done. One example is Apple’s ResearchKit, which is collecting data on Parkinson’s disease patients’ assessment of their condition. In a study conducted by Sage Bionetworks and Rochester Medical Center, participants downloaded the app and completed tasks that measured their performance in balance, speech, memory and dexterity over time. Information was collected through their iPhones’ data sensors. This is a vast improvement over the typical model of patients coming into a facility, getting hooked up to expensive and complicated machinery and performing tasks in an artificial environment. These data are already being used to measure the association between medication use and symptom improvement through machine-learning algorithms.
AI in the lab
Pharmaceuticals are now reflecting the effects of machine learning. Today, the drug-discovery process takes many years. Millions of compounds are synthesized and tested against a target to find viable drugs. Often this process is conducted through automated high-throughput screening, which uses robots to sample each compound and a reagent in microplates. This process is time-consuming and very expensive.
Work is now being done to enable virtual drug screening using machine-learning algorithms called neural networks. Using previously obtained drug-screening data from a particular disease, a neural network can predict which compounds will act on the target of interest. In the paper “Massively Multitask Networks for Drug Discovery,” researchers at Stanford University and Google discuss using data from a variety of sources to accelerate this process even more. By combining drug-screening data for multiple diseases with a variety of biological processes, a neural network has even more data available to help it make predictions about which compounds are viable for further development. Not only will this lower the cost of drug development, it will accelerate getting drugs to trial and lower the barriers to entry, allowing smaller companies to enter the pharmaceutical industry.
Paging “Doctor Data”?
The possibility of your physicians being replaced by a machine is many, many years away. Instead, I believe the future of machine learning and medicine is a merging of algorithms, technology companies and healthcare providers. You can see the stage being set for this by big-data collections and partnership formations.
For instance, Memorial Sloan Kettering (MSK) has partnered with IBM’s Watson to treat cancer patients. MSK shares information about individual patients with Watson, which will match that information with data from similar patients and use published literature to optimize treatment. No individual physician can match Watson’s speed in combing through the massive amounts of literature released each year. However, it’s also clear that no algorithm can currently take a good history and physical, assess a patient’s physical exam and incorporate the entire clinical picture into an automated regimen for patient care.
A physician knows that the human ability to understand quickly and empathetically a patient’s facial expressions, family dynamics and goals for care has a major role in decisions about a course of treatment. Being hands-on allows a level of interaction—looking forward and backward—not yet possible via AI. Machine learning requires previously gathered data to make predictions, but there will always be some combination of history and physical conditions that is unique to each patient—not seen in any previous patient dataset. This is where the art of medicine becomes intertwined with science and technology.
