I'm an enthusiastic realist when it comes to technology. What motivates me is not the machine, but what it can do to help us to solve problems. More sophisticated machines allow us to tackle more complex problems. And the more complex the problem, the more that interests me.
You may have heard that AI is going to solve all our problems. Or did you hear that it's going to steal our jobs, enslave and then destroy us? Not a day goes by without thrilling or terrifying media stories about the consequences of our AI progress. There is good reason to be energized by the potential, but we're being bamboozled by hype.
Melodramatic reporting muddles sci-fi and doomsday scenarios, leaving us with a narrow scope of AI. It leads us to accept puffed-up claims at face value. What a relief when the respected analyst firm Gartner declared (via City A.M.) AI to be at the "peak of inflated expectation."
Perhaps, given my field of work, I should regard AI's impending slide into the "trough of disillusionment" with some trepidation. In fact, I welcome greater rigor in how we assess AI. Its value is not intrinsic, and the bigger the problems we can solve, the more valuable AI becomes.
Truly difficult problems are rarely binary. To get where we want to go in life, we frequently need to improve or alter multiple factors and control the order and timeframe in which change occurs. In scientific terms, these are called multi-objective optimization problems, or MOOPs. For me, they've been something of a life's work.
I first discovered MOOPs in the unlikely surroundings of rural West Virginia. Raised by my great-grandparents, my family was materially poor but rich in the values that are characteristic of "the greatest generation." In 1994, when I was 13, something unexpected happened. A relative I barely knew decided to buy me a computer. It was a typical office machine, but this kind of tech was pretty rare in my neck of the woods. Before long, I found my way to the DOS terminal and became obsessed with programming. As I progressed through the labyrinth of logic, I found I increasingly evaluated life's various trade-offs in more analytical and systematic ways. I didn't have it fully cemented, but I was thinking about life as a MOOP.
I realized that my personal MOOP was escaping the cultural inertia around me. The values of grit, determination and resilience that had been etched into my psyche combined with my newfound appreciation of logic, organization and computing. These attributes equipped me to be accepted at the Air Force Academy, which served as a springboard to grad school. My chosen field of study was -- you guessed it -- MOOPs.
The most important thing I learned about MOOPs is that to have a chance of solving them, you need software that learns. Even our most powerful computers aren't fast enough to process the variables and data volumes inherent to a MOOP. When problems get really complex, you need AI.
Health Care Is A Massive MOOP
Shortly after I completed grad school, my wife began to suffer problems with her vision. The cause was a cyst in her brain. The operation to remove it was successful, but the vision problems remained. These were misdiagnosed and mistreated, leaving her with severe health problems that continue to this day.
The challenges of coping with and attempting to restore my wife's health have been a MOOP my family lives with every day. What I have learned through years of this experience (investigating clinical procedures, interacting with hospital systems, etc.) is that health care is arguably the biggest and most meaningful MOOP we face.
Health care is an insanely complex problem. It involves every human on the planet, it's life-and-death important, there are almost innumerable variables to consider and everything happens at huge scale. Even with my cautious evaluation of AI, I believe that machines that learn will be the most disruptive force to affect health care in my lifetime.
AI Is Addressing Health Care MOOPs Today
Fortunately, we don't need to wait a lifetime to see the start of this disruption. AI is already making an impact, and the most compelling use cases are those that address one of health care's many MOOPs.
Analyzing human communication is deceptively prosaic. In health care, an accurate and speedy understanding of clinical records and reports directly impacts patient outcomes. Ordinary computers struggle with this task due to the variability of human language, where meanings change due to time, context, nuance and other factors. My company, Digital Reasoning, has developed an AI platform that tackles the MOOP of language analytics. One application is machine reading of pathology reports to intelligently automate the triage of cancer patients. This frees doctors from vital yet time-consuming rote work, increasing their efficiency while allowing them to spend more time with patients.
Or consider personalized genomics. Sequencing the human genome was a monumental project that took 13 years. Now, as we seek to understand the fine detail of what genes do and their complex interactions, AI is helping us to move much faster.
A final example, which is dear to me, is stem cell therapy. Making sense of the complex interactions between stem cells and damaged tissues, and figuring out who makes a good candidate for a therapy based on chemical signatures in their body, is another MOOP. It's no surprise that the bioinformatics community is leveraging AI to understand the complex interactions between stem cells and damaged tissues. I'm happy to say that my wife was determined to be a candidate for one such therapy. It's too early to know if it will be the solution to our MOOP, but I'm hopeful and optimistic.
Hype is a distraction, but certain developments show that the opportunities for leveraging AI to improve health care MOOPs and patient outcomes are very real. We should scrutinize hyperbolic claims, but companies capable of explaining how their AI aligns to a MOOP deserve a hearing.