When it Comes to AI, the Fun Part is Over
By Beaumont Vance, Director of the Analytics Center of Excellence, TD Ameritrade
That’s the first question I usually ask when I’m pulled into a discussion about the dangers of Artificial Intelligence (AI). Only once has anyone answered it correctly. Typically I get puzzled looks that say ‘what does this have to do with AI-driven human extinction?’
Within the first five minutes, conversations about AI and innovation typically turn to discussions of killer robots (or similar dangers). The mere mention of AI tends to incite hand-wringing conversation about a dystopian future in which humanity is unemployed and battling our robot overlords.
This discussion conjures up the image of Nikola Tesla in my mind, because I am sure when he invented the first radio controlled boat, he too hoped the world would change radically overnight. I imagine him showing the world his incredible logic defying remote control boat there at Madison Square Garden in the very improbable year of 1898. (1898! Talk about being ahead of the curve.) And then waiting, decade after decade, for someone to implement this quantum leap in technological capability and change the world.
We tend to think of the great innovations as being suddenly game changing. Someone invents the better mousetrap, the world beats a path to his door, and the inventor does an IPO and spends the rest of his life as a billionaire giving Ted talks about the dangers of highly effective mousetraps.
But this isn’t reality. Tesla’s fantastic invention, incredible as it was, was not turned into a viable product until the 1960’s. The problem is that innovation itself does not equal value; rather, it is just one part of the value chain. Many could appreciate his invention, but none could find a path to implementation.
So the likelihood of some great new AI algorithm suddenly transforming the world as we know it is slim. Much more likely is that someone inside of an organization has a breakthrough, and spends the next ten years trying to get someone to appreciate it, allocate enough resources to create an enterprise version and get it into a production environment.
We need to innovate on how we work together, execute and implement within the context of a sea of well-worn stable, change-resistant processes
Brilliant, innovative concepts are not enough. Before we can imagine a world where AI is ubiquitous we must first tackle the far more mundane task of implementing AI into everyday business such that it produces measurable value. The implementation process, along with the compliance, legal, security, risk and insurance components, it turns out, is the one thing we need to innovate more than any other.
If you want to understand the value of implementation over pure innovation, consider Henry Ford. He did not invent the automobile. That honor goes to Karl Benz who created a gasoline-powered vehicle in 1885. Ford did not even invent the assembly line for which he receives so much credit; Adam Smith had outlined the value of the assembly line in 1776.
The true invention that Ford introduced was how to implement that brilliant idea from 1776 as it applied to the mechanical innovations of the late 19th century. He figured how to practically and profitably move innovation from merely an idea to something you could buy and drive home.
We stand, as far as AI goes, at a point in time much like right before Ford put together his assembly line. We have great AI capabilities already discovered and reported on. We have many visions of what the future could look like should these capabilities be applied. In other words, the fun part of inventing is largely done.
What has yet to happen for AI is the shift from invention and possibility, to the present, more arduous task of implementation. This is at least as large a challenge as coding.
Clayton Christensen underscores the devilish difficulty of employing innovations in his book, The Innovator's Dilemma. Companies, he says, can know that they need to innovate, create sound ideas, and still be unable to execute on them. His valuable insight is that stable, profitable companies are built of many stable, repeatable, efficient processes. The very definition of a stable process is that it resists disruption. And so the very idea of changing a conglomeration of stable processes is anathema to their very raison d'être, i.e., companies are designed to resist change, and innovation is by definition radical change.
Our gravitation towards a “killer robot” discussion around AI is evidence of the built-in systemic resistance to innovation. AI can’t pour a glass of water and already we are coming up with reasons to stick to our old tried and true practices. We will not be disrupted!
We need to innovate the entire system, or at least innovate around it. We need to develop people and processes that can handle disruptive change. We need more than a new way of thinking. We need a new way of doing.
So, the challenge of AI today is less about neural networks, and more about employee networking. Less about deep learning and more about re-learning. We need to innovate on how we work together, execute and implement within the context of a sea of well-worn stable, change-resistant processes.
Until we approach this challenge with the same energy, passion and determination that we have approached developing AI, we will remain something like Nikola Tesla in 1898. Having just shown the world our breakthrough innovations we wait, wondering why someone else doesn't implement our great idea.