Photo: Fanatic Studio/Alamy Stock Photo
As an avid futurist, I enjoy the opportunity to share my thoughts about the potential impact of technology. But as an actuary, I must first qualify my predictions: Much of what I will describe in this article is going to happen, but some of my predictions will be wrong. Unfortunately, I have no confidence limits or meaningful probabilities that I (or anyone else) can apply to them. As many have said, “Predictions are difficult—especially about the future.” 1
I will offer some historical perspectives on the advances made in artificial intelligence (AI) generally, and machine learning (ML) more specifically, and then offer some predictive modeling perspectives to extrapolate these advances for health care and longevity.
Humans vs. AI
AI has been around for several decades, but the hype of the early 1980s was followed by a period of disappointment. Manually coded rule trees looked promising for small, simple applications, but they did not scale well as complexity increased.
I was the original architect and co-inventor of an underwriting expert system that has evaluated more than $100,000,000,000 of life insurance applications in several countries and languages, but now it has a staff of more than 100 to maintain and enhance it. Back then, the idea of dynamic self-modifications to meet changing needs seemed beyond the capabilities of AI. Then, in 1997, something dramatic happened that rocked our belief in the overarching superiority of humans over machines for strategic logic. Deep Blue, a program from IBM, beat Gary Kasporov in chess. He was the world’s champion—and he was soundly defeated by a program.
The excitement in the AI community was palpable, but the skeptics pointed out that it was just a game, and one with very logical rules. They asserted that true intelligence would require an ability to interpret human languages, such as English, with its vast ambiguities and puns. In 2007, another IBM program, Watson, beat the world’s best Jeopardy! players. The language barrier had been breached. However, Watson was not programmed even to know that it had won.
Fast-forward to 2017, and an AI program defeated the world’s 18-time champion in Go, a strategy game far more complex than chess. As a comparison, a chess player has about 20 moves available at any point during the game, while in Go there are about 200 possible moves at any given time. As the turns cascade, the different game possibilities become enormous. Yet, once again, the doubters could point out that Alpha Go (the AI program) was trained by inputting thousands of games played by human masters. A year later, that all changed! Alpha Go Zero emerged and soundly beat Alpha Go to become the undisputed Go champion. The training of Alpha Go Zero involved no human game histories. The program was fed only the rules of the game, and it learned by playing against itself (for 42 hours).
Unencumbered by the baggage of human knowledge, the AI ML was able to learn faster and better.
Limitations of AI
A year later, Alpha Zero, the generalized strategy game player AI, became the world’s best chess player (with only four hours of self-training). Still, naysayers and even AI experts point out that these are all applications of what they call weak AI, or narrow intelligence. It is very specific. Some of these experts say general intelligence is many decades away. For example, Alpha Zero can play chess, but it can’t teach you how to play chess. Watson had to rely on Wikipedia and other stored databases—it did not have the inferential, pattern matching and creative capabilities of even a human toddler. Mastery of numbers and words was still deemed to be a very small subset of human capabilities. And, surely, the recognition of faces and creation of abstract artwork were unattainable by AI—but not anymore.
Now, AI convolutional neural networks (CNNs) are routinely used to detect potential terrorists at airports and cancer nodules on X-rays. AI generative adversarial networks (GANs) have created paintings that have fetched hundreds of thousands of dollars at art auctions. Your smartphone, which possesses millions of times more storage and processing power than the onboard computer for our first moon landing, has progressed from a brick that only could make phone calls to a hand-held digital personal assistant that guides you through complex streets in a foreign city, keeps your appointments for you, chronicles your life via geo-tracking, shares photos and videos across continents, and remembers the phone numbers you have long forgotten.
In August 2019, yet another AI advance was announced when a program beat noted human experts in six-player Texas Hold ’em poker, a game involving multiplayer strategy and bluffing.2
I agree we have not reached the point of more general AI; however, I disagree that this milestone will be far in the future. In my opinion, we are entering an interim era of what I call Laminar AI. Thin layers of carbon fiber laminated together are far stronger than a comparable width of steel. The six layers, or laminations, of our human neocortex enable amazing pattern recognition very quickly despite the slow (about 5 millisecond) firing time between neurons. Likewise, I predict that the lamination, or binding together of many narrow AI systems, will accelerate progress toward wider and wider AI.
Impact of AI on Health Care and Longevity
As precision medicine advances at near exponential rates, we will be able to diagnose diseases earlier and actually prevent most. Actuaries, medical professionals and data scientists (some human, some AI-based) will tap into the microbiome’s interaction with our brains (primarily via the vagus nerve) and customize dietary treatments that will prevent future diseases for which we (as very specific individuals, since everyone has a unique combination of DNA and microbiome) have a genetic or environmental predisposition. They will mitigate or eliminate diseases already in place. They will be guided via both wearable and embeddable monitors and dosage devices. This will, in most cases, obviate the need for surgeons; the era of slash and burn to cure people will be viewed, in retrospect, as ignorant barbarism. Invasive biopsies will be replaced by the tricorder—a Star Trek product that has already been developed as a part of an Xprize contest3—and will be enhanced significantly in the next few years. Clinical studies have shown dogs are able to smell some cancers in humans. Imagine a bot with extrasensory capabilities for smell, sight (beyond the human visible spectrum), hearing and so on, coupled with AI, to assimilate all of this data and make inferences beyond human capabilities.
The smartphone, which has become ubiquitous today, will become a relic of the past as it is replaced by ear implants that replicate its features (perhaps communicating with our optic nerve for the camera input). It will check our vital signs many times per minute, communicate wirelessly via bandwidth far beyond current imagination, and serve as a universal translator (some hearing aids available today claim the ability to translate 27 languages in real time). The internet of things (IoT), specifically of health things (IoHT), will connect us to AI ML in ways barely imaginable now.
Sickness will become a distant memory of the pre-laminar AI times of the past. Longevity will extend accordingly—but it won’t just be a matter of living longer. We will also enjoy quality of life longer. Some longevity experts, such as biomedical gerontologist Aubrey de Grey,4 predict that during the next five years we will see medical breakthroughs that could extend life another five years. And during those next five years, we will have breakthroughs for another 10 years of quality life, and so on …
The possibility exists that some people alive today could become a-mortals.5 They would not be immortal, since having a building collapse on them will still create a puddle of a former person, but they need not fear dying from disease. Actuaries in the annuity business will need to be especially vigilant about selling to them.
But what other concerns should actuaries have about this new world? I predict health insurance actuaries will see an initial bubble of increased marketability as the explosion of new procedures, techniques and devices affect health insurance premiums in new ways. They will need to learn how to drink from the fire hose of increased data and apply predictive modeling techniques beyond current capabilities. This will spur the development of more and more AI-based automation, as models are developed and then refined by other models and ML becomes pervasive throughout our profession. Ultimately, ML will replace some actuaries and relegate others to niche roles, such as the few blacksmiths today who attend to show horses. Life actuaries will see dwindling opportunities as people live far longer and some don’t die (the a-mortals) from disease or system deterioration. Death by violence will be rare, as privacy will cease to exist, and crime will be immediately exposed and dealt with by monitoring far beyond the capabilities of Big Brother in Orwell’s novel, 1984.
The Pace of Change is Unimaginable
Some readers will assume I am going wildly blue-sky here with all of these predictions. But as futurist Peter Diamandis says, “The day before something is truly a breakthrough, it’s a crazy idea.”
Take, for example, the pace of improvement in data storage capability. In 1979, I bought my Apple II computer with the maximum 16K random access memory (RAM) available, but I needed a bit more RAM to program in Pascal, so I had to purchase an additional 16K add-in circuit board for $300. At that rate, a gigabyte (GB) of such storage (a hard drive was not yet an option) would cost more than $18,000,000. In today’s world, a GB of storage costs less than $0.02, or roughly a billionth of the former cost.
The rate of improvement is daunting. Who would have imagined in 2007, when the iPhone was introduced, that more than 2.2 billion of them would be sold by November 2018 (when Apple decided to stop reporting sales numbers)?
The city in which I live, Chesterfield, Missouri, hosts the world’s first virtual care center. The Mercy system opened it in October 2015. Its website shows virtual tours of the building, which houses very specialized medical professionals who remotely monitor patients, guide complex surgeries long distances from the operating theaters, and bring the collective expertise of its 43 hospitals to patients who are sometimes hundreds of miles away.
Looking forward from the amazing progress just in the last couple of years, I think of Donald Rumsfeld’s famous quote: “There are also unknown unknowns—the ones we don’t know we don’t know.” In 1884, Edwin Abbott published his revolutionary book, Flatland, where the inhabitants lived in a 2D world. When visited by a sphere from our 3D world, they could not comprehend or visualize its existence, just as we have great difficulty comprehending or visualizing string theory’s 11 or more dimensions.
Skills Needed to Survive
It is clear the skills that actuaries learned yesterday must be augmented with many new ones if we are to survive better than the blacksmiths, the slide rule manufacturers, the petroleum engineers and, frankly, most other specialty professions. The ability to look at a bunch of disparate facts and make an intelligent decision used to be the defining characteristic of an expert (doctors, lawyers, actuaries, etc.), and these individuals were paid very well accordingly. Now, we see AI ML challenging and often surpassing that type of expertise. The future will require cross-functional skills. Nurses will be in demand long after the work of most doctors has been automated. The skilled and creative home handywoman will earn far more than her white-collar counterparts. It will not be sufficient to look at a lot of disparate facts (body metrics, health histories, investment returns, tax regulations, etc.) and make premium or reserve projections—AI ML will do that better and faster and less expensively (good-fast-cheap is a powerful combination). We might need to ramp up our interpersonal skills to compete.
Insurance products will have to change as well. There will be a new or increased need to cover accidents; depression; avatars; virtual reality; liability insurance for your autonomous car; replacement insurance for your android companion or your robodog; nanobot sensors; the ravages of inflation as people outlive their savings; interruption of service insurance for your solar panel grid, wind turbines or geothermal heat pump; and anxiety insurance.
Accidental death insurance will move to the forefront as the a-mortals no longer die from cardiovascular diseases or cancers. Some a-mortals may think of themselves as immortal and take unnecessary risks (as some teenage drivers do today). The colonization of other planets will pose new types of accidental death risks and pricing methods.
Sadly, suicide will become an insurance product rather than an exclusion to life insurance policies. Deaths by suicide already exceed those of wars plus homicides combined. As the frenetic pace of change overwhelms some individuals, lyrics from a 1961 stage musical come to mind: “Stop the world, I want to get off.” Pricing suicide insurance will involve actuaries changing traditional thinking on moral hazard, and partnering with bioengineers and psychologists to embed nanobot sensors that detect dangerous mental anxiety and alert intervention agencies accordingly.
On the health side, malaise or disengagement with reality might be a major risk to be insured. Some individuals may find the virtual reality experience far more appealing than the banal reality of their corporeal lives. And, of course, they will wish to insure the avatars they have come to invest so much time with that life without them would be impaired.
Actuaries have had a good run in one of the best jobs for several decades. Are the good times about to end? Does AI ML represent the barbarians at the gate? Not necessarily!
We can survive by reinventing ourselves to avoid commoditization. Predictive analytics should be an important set of skills for us in the future, but it must be accompanied by other skills as well. We can leverage our reputation for ethical behavior and improve our communication skills. We can combine our business knowledge, analytics and compassion for families to embrace a more holistic approach to risk management.
Early humans discovered the horse was a faster runner, and we learned to ride it. We observed fire as a powerful source of heat, and we learned to control it. AI and ML can be powerful tools for good or for evil. Let’s master them, enhance the profession, and continue to serve humanity and make the world a better place. Let’s reinvent ourselves.
- 1. Variants of this quote have been attributed to Nostradamus, Yogi Berra, Mark Twain, Niels Bohr, Samuel Goldwyn and many others. ↩
- 2. Brown, Noam, and Tuomas Sandholm. 2019. Superhuman AI for Multiplayer Poker. Science Magazine, August 30: 885–890. ↩
- 3. Visit xprize.org for a detailed discussion of the tricorder contest. ↩
- 4. Aubrey de Grey is a well-known speaker on longevity at actuarial conferences. Here is a link to a video where he describes the likelihood of living to 1,000 years old. ↩
- 5. Yuval Harari discusses a-mortals in his excellent book, Sapiens—A Brief History of Humankind (Harper Collins Publishers, 2015). But don’t stop reading there! His sequel, Homo Deus—A Brief History of Tomorrow (2017), carries the idea even further. ↩