Will AI Destroy Life Insurance Before It Destroys Humanity?

There’s been a lot of talk of late about whether AI will take over the world and kill us all. If this comes to pass, it would be an awfully bad time to be in the life insurance business.

Yet, I haven’t heard much panic about what these predictions will do to life insurers. Met Life stock has been pretty flat over the last month. Its investors don’t seem too worried. Is that a more important signal than prediction from AI “experts”?

Should You Be Worried About The End of Times?

Before discussing in more detail what it means for life insurers if the doomers are right, let’s first think a little about how likely they are to be right.

Now, if you’re a long time reader, you probably expect me to say I am frightened and agree there is at least a 10% chance we get wiped out by machines, maybe even higher. After all, I’m the guy who spends way too much time worrying about remote tail risk events.

And yet, my objective is always to understand and, as best as I am capable of, estimate the probability of a tail risk. For example, while I worry a lot about Seattle quake, I am fairly dismissive of New Madrid recurring.

So, let’s do some thought exercises. The odds of a category 5 hurricane hitting Miami are <1%. What do you think is more likely? Another Hurricane Andrew that lands a bit further north or AI killing us all?

I feel pretty strongly a cat 5 in Miami is more likely.

How about another example? When the Patriots made their famous Super Bowl comeback against the Falcons, their estimated probability of winning when they were trailing 3-28 was under 1%.

What do you think is more likely? Tom Brady leading a historic comeback or Sam Altman killing us all? I’m going to go with Tom Brady on this one.

So, some pretty simple mental models can help us realize the odds of AI destroying us all are well below 1%.

If you prefer some other thought experiments, civilization has existed for over 5000 years and none of the many, many risks have eliminated us. While the past is no guarantee of the future, a Bayesian approach would suggest we are likely to keep overcoming risks.

Want more examples? The Black Death was certainly > a 1 in 1000 event. Yet, it didn’t kill us all. Hitler was probably a 1 in a billion event and he couldn’t even conquer Europe.

So, the doomers have a serious overestimation problem. If they made a more humble (but less dramatic) claim that AI could kill 10% of the population in a tail event, then we could have a more serious talk about the odds of that.

I’d still say they’re well over 1 in 1000. Remember, Covid killed only 1% of the population so even a 10% mortality prediction suggests AI is potentially 10X more deadly than a global pandemic.

How Would It Even Happen?

So we’ve established that the worrywarts have overestimated the likelihood of human annihilation. Now let’s move to step 2. How would AI even be able to attempt such a thing given that, you know, it doesn’t have a body?

The idea that AI can manufacture a bioweapon is ludicrous. Where would it get all the chemicals? How would it pay for them? Where would it assemble it? How would it convince humans to transport all the materials to the assembly site?

The only plausible path is an evil person uses AI to help execute their evil plans. Like if Saddam Hussein only had AI, he could have unleashed chemical weapons for real!

The flaw in this logic is, even if that evil person existed AND was able to get AI to comply with its wishes (without anyone at any of the AI labs noticing), the perpetrator would be human, not AI. AI would only be a tool, not the mastermind.

So the idea that AI will get a mind of its own and outsmart us doesn’t really fly. And the idea that the next Saddam will be an AI expert and evade all detection and not fear the retaliatory nuclear attack is farfetched, to say the least.

So, sorry, I’m not buying the premise that the next reboot of Dr. Evil will be a hacker chugging Monster Energy drinks attempting to convince AI to help him gain access to the nuclear codes (though it could be a fun plot for a movie).

Which brings us to the “AI will get control of the nukes” hypothesis. How? Again, it has no body. The nuclear codes aren’t stored on a AWS server. The activation sequence requires human participation. It is preposterous to suggest AI can commandeer nukes.

And even if it could, to what end? For fun? If it really ends up being smarter than all of us, then we are no threat so why destroy us all? There doesn’t seem to be a point to it. After all, who will build more data centers if humans aren’t around?

Back to Life Insurance

OK, now that we’ve established AI annihilation is highly unlikely, we can do a proper risk assessment. Should we ignore the risk because it is remote? No! That’s how we ended up unprepared for Covid.

Realistically though, what is a life insurer supposed to do differently knowing this AI risk exists? I guess you could raise your tail probabilities of calamity slightly but that’s not really going to result in any material change in behavior.

After all, if you hold 5% more capital for “the world ends” risk, that just means you lose 5% more capital when the world ends!

The more relevant concern would be could AI raise the odds of a Covid or Black Death type event. That you could at least buy reinsurance for. But it’s hard to see how AI is going to make disease more common or virulent.

If anything, AI may help us develop better defensive tools so we can slow the spread of a pandemic. That’s a win for life insurers.

What If The Doomers Have It Backwards?

There is an underlying assumption in all these fear based theories – that AI will be used more for evil than for good. I won’t dismiss that possibility, but I’m not sure it’s the most likely outcome.

For example, AI has great promise for cancer breakthroughs and other medical advances. You know what that means? There is a bull case where we all live far longer because of AI.

And if that happens, you know who one of the biggest beneficiaries is? That’s right, life insurers! Nobody is more levered to higher longevity than the life insurance industry.

So maybe now we understand why the life insurance stocks have been relatively unaffected by all the recent AI fears.

Perhaps the answer is the market judges that better longevity is more likely than higher mortality from rogue AI. In fact, given that markets hate uncertainty and worry more about negative outcomes than positive ones, the boredom of life insurance stocks suggests investors think it is far more likely that better longevity is the ultimate outcome.

If you’ve studied your doomer history, you know that there is always some type of scenario that enters the zeitgeist predicting eternal doom. AI is just the newest of these. It is not novel.

Climate change hasn’t (and won’t) kill us all. We didn’t all starve as global population increased (that was a big fear 50+ years ago). At various times, we were also going to die because the planet was cooling (not warming) too fast.

And obviously, it was all but a certainty there would be global nuclear war dooming us all. My generation was raised on that fear.

Yet, the reality that emerged was access to nukes prevented wars. The deterrence of the second strike made the world safer even if it felt more dangerous.

I suspect AI will be similar. Plenty of smart people will work on making AI safer. In fact, if you want the cynical explanation of all this sudden fearmongering coming out of Anthropic and OpenAI, it’s because they want to develop the solution to “save” the world and then take credit for a risk that never existed!

The takeaway is human beings are really good at finding solutions to seemingly impossible challenges. We will also solve how to co-exist with AI. It will be OK.

Where The Real Risk Is

Well, mostly OK. There are risks to be concerned about. They’re just not to our mortality. They are more a threat to our wealth and comfort. These are still remote risks but more probable than AI killing us all for fun.

Next time, I’ll discuss some of those as they are of much greater concern to P&C insurers.