Is AI Antifragile? The Lindy Effect and 3 Reasons Why it Matters

I had the great privilege of having Nicholas Nassim Taleb as a colleague many years ago. It was a pleasure to learn from him, as he was open, approachable, and the kind of guy that you’d enjoy having squid ink pasta with. I’ve been following his writings and twitter posts for years and admire his brand of no-nonsense anti-intellectualism. His recent book “Antifragile: Things That Gain from Disorder (Incerto)” got me thinking about AI’s role in our future. Would AI pass his test of being antifragile, and if not what does this mean for AI?  

According to Taleb, "Actors hanging out at Lindy’s Deli in NYC gossiping about other actors discovered that Broadway shows that lasted, say one hundred days, had a future life expectancy of a hundred more. For those that lasted two hundred days, two hundred more." This became known as the Lindy Effect, a way of acknowledging something that’s been around a while will probably stay around for a while longer.

The Lindy Effect works for non-perishable things like technology, ideas and ideologies, not with perishable things like people. For technology the longer it lasts the longer it can be expected to live as a direct result of it being antifragile, or not sensitive to disorder or the passage of time. So something that is antifragile actually gains longevity with the passage of time and gets better the more it is subjected to stress and shocks. It simply ages in reverse.

From Wired: The Surprising Truth: Technology is Aging in Reverse, Nassim N. Taleb

So let's look at some practical examples. Books made of paper haven’t been replaced by digital versions as they’ve been around for hundreds of years and will survive for many more. They are antifragile and the technology demonstrates the Lindy effect. A book written by Justin Bieber will not survive a long time, as his celebrity is passing, so his book is not Lindy compatible, unlike say, The Bible, which is. 

So is AI Lindy compatible? If we look at AI tech from a narrow perspective, it has been with us in production form for all of about 5 years, so according to the Lindy Effect AI can only be assured of another 5 years survival. Not too shabby for new tech, but a short period to amortize an expensive investment. If we look at AI in the bigger picture we can make the argument that AI is culmination of over 50 years of digital computing experience and will have a longer predicted lifespan. So which time scale is the right one for AI is it 5 or 50?

So which time scale is the right one for the Lindy Effect and AI, is it 5 or 50 years?

My bet is that it’s closer to 5 years than we’d like. Something that is “antifragile” actually gains from disorder and or the passage of time, it actually gets better when subjected to shocks or stress. Now this is where AI starts to get into real trouble. AI is really not very intelligent and the emphasis on AI has to be on the word “artificial” more so than intelligent. AI cannot decide to alter itself, or self-adapt, when presented with disorder. It hates any changes to either the job it does, the information it processes, the questions people ask, or changes in the systems it runs on. Shocks and stress generally cause it to fail. Take Microsoft’s AI powered chatbot “Tay.ai” for example, which morphed into something very sinister and was shut down when subjected to the shock of real world chat inputs from the general public. None of our AI’s can be guaranteed to work with the passage of time since none are really survivable without significant maintenance and modifications. AI is the model of fragility.  

So why does any of this matter? 

1. Do not succumb to the hype. AI is a stepping stone technology it is not antifragile. Do not expect what you buy today to be around tomorrow and make sure you calculate AI system payback over a shorter period of time than you initially thought. Your new AI will be transformational, until the day that newer technology arrives and replaces it and it isn’t. Think about the last time you reached for your Blackberry? It was the transformational device and then it wasn’t. AI is no different; it will do great things for us, no doubt, until “intelligent” systems are designed with increased survivability. If you believe the hype you’ll think that the systems of today are an “investment.” That just might be true with specific, definable use cases, but remember that today’s AI is very much, a transitional technology. 

Do not succumb to hype or horror stories.  

2. The horror stories of AI replacing human jobs en-masse are unfounded. I’m happy to tell you that the jobs done by butchers, bakers, lawyers and even customer service clerks are Lindy compatible. These jobs have existed for quite a while and are sufficiently antifragile that the Lindy effect shows they will be around for a while more. So take the horror stories of massive, imminent job losses with a grain of salt. Yes AI will displace jobs, it's happening now, but there are certainly many Lindy compatible jobs for humanity to occupy within most companies and new jobs coming that were never envisioned prior to the launch of AI systems. 

Customer service front liners are antifragile, your AI is not.

3. Control your enthusiasm for the elimination of the aforementioned customer service front liners, and others whom your managers deem expendable, they are antifragile your AI is not. If firing all of your customer service people is what pays for the AI, think again. Retaining them will build redundancy into your business operations that make your organization more antifragile and better equipped to deliver quality service. Your customer services team will survive the disorder of market crashes, cyber attack and blackouts better than your AI. How long will it take to program your customer service chatbot to give the right answers after a financial crash, or after its been hacked? I emphasize this because elimination of customer service clerks seems to be the first use envisioned for any AI system without a thought as to whether this would make the institution more fragile. 

So the Lindy effect is not a law, it’s a “rule of thumb” or estimator and sometimes estimates are wrong. It’s hard to ascertain what the Lindy life span of our current AI products will be and I accept this critique. What is equally important and less debatable is that it gives us a way of thinking about new technologies to consider whether they increase fragility of our systems. In all of the AI discussions I’ve had this -never- comes up in conversation

I work in AI and have real skin in the game, as do the AI system partners I work with. No one wants AI to survive and prosper by making your company more antifragile than me. Still, AI will have to earn its Lindy Effect chops just like all of the other technologies that came before it. No amount of hype proclaiming a Star Trek like future can do that. In the meantime, I’ll be counting every day and month and adding it to AI’s life expectancy. 

You might also enjoy my other articles on AI: 

3 Ways to Make Your AI Standout in a Crowd

Insurtech's Perfect Storm

Risk will be Assimilated By AI, Robos First, Resistance is Futile

Robos Get Your Hands Dirty in the Field of Dreams

You Can Run but You Cannot Hide, Wealthtech and AI

WealthTechs Read My Mind! Please!