I remember well how over-hyped AI was back in the early 1980s when I worked with Applied Expert Systems, a startup founded by some MIT professors that aspired to use expert systems to transform the world of personal financial planning.
Sadly for those who invested time and money in this company, its product never found much of a market and it folded.
And nearly 40 years later, it looks to me as though the promise of AI way ahead of what it will deliver.
Here are three reasons I've reached that conclusion.
1. Many CEOs Are Being Scared Into Caring Too Much About AI
Consulting firms have the power to scare companies into paying for projects designed to alleviate the fear they create. A case in point is pwc, which published a 2019 survey of CEOs which found that "80% of those surveyed believed that AI will significantly change the way they do business in the next five years."
Are all CEOs taking the lead in driving AI as their first strategic priority? Not really.
As Kartik Hosanagar, Prof. of Operations, Information and Decisions, at Wharton told me in a February 8 interview, "There are three kinds of CEOs when it comes to AI. The first category are the blind followers -- they don't understand AI but they've heard it's a 'thing' and they trust it. The second are the ignorant skeptics -- they don't understand AI and don't trust it. And the third are savvy managers who are integrating AI into their business. I have not done formal research but there are more skeptics than blind followers."
2. There Are Very Few Examples of High Payoff AI Applications
Experts I talked to cite Google as the source of at least one successful use of AI. While I do not know how much Google spent on this application, I am convinced that its payoff is economically significant.
Despite the paucity of compelling examples of high-payoff AI applications, the market forecasts for AI are pretty large. For example, IDC estimated that spending on cognitive and AI systems for 2018 totaled $24 billion and would grow at a 37,3% compound annual rate to $77.6 billion in 2022.
AI expected most of the 2018 spending to go to four application areas: largest automated customer service agents ($2.9 billion), automated threat intelligence and prevention systems ($1.9 billion), sales process recommendation and automation ($1.7 billion) and automated preventive maintenance ($1.7 billion).
Over the longer term. IDC expects the fastest growth in other areas. Specifically, the following will receive the fastest growth in five year average investment through 2022: pharmaceutical research and discovery (46.8%), expert shopping advisors & product recommendations (46.5%), digital assistants for enterprise knowledge workers (45.1%), and intelligent processing automation (43.6%).
There are some startups that are using AI. As Adam Pah, clinical assistant professor of management and organization at Northwestern's Kellogg School, told me in a February 12 interview,
There is a startup in China that uses AI to make consumer micro-loans with a higher chance of being paid back. Other examples include Amplero, which is getting 100% growth through more effective ad targeting. Hello Fresh has boosted revenues 4% just from pushing its AI-developed recommendations to households. And Lemonade sells renter's insurance using AI instead of insurance agents."
A fairly typical story about a company that's trying AI is a system called Philyra -- intended to help invent new types of perfume -- built by Symrise, a large perfume maker.
There is no compelling payoff from this system after two years of trying. According to MIT Technology Review, Philyra was developed in partnership with IBM, has taken two years to get working, and is just being used by a handful of Symrise 70 fragrance designers.
Let's take a deeper look at the Google example. As Kartik Hosanagar, Prof. of Operations, Information and Decisions, at Wharton told me in a February 8 interview, "Google decided that AI was going to be the next big thing so it moved from operating a centralized AI group to lending them out to the product teams for three to six months."
An initial result of this effort was to improve the quality of Google's search function. "Using machine learning, Google was able to track which search results users actually clicked on most frequently. Often Google's algorithm listed the most frequently clicked result third on the list. Using AI, Google improved its search algorithm so that 95% of the time, the most frequently clicked link made it to the top of the list," said Hosanagar.
I am guessing that Google has found a way to use this improvement in its algorithm to boost its revenues.
Meanwhile, Google has also used AI to reduce its data center costs. As he explained, "Google used machine learning to predict its electricity costs every hour. By making accurate predictions of how much electricity the company would need, Google was able to reduce it electricity costs in its data centers by 40%."
3. Very Few Companies Can Afford or Find Good Uses For AI
AI engineers are expensive -- their total compensation packages can go into the millions of dollars. It does not seem likely that large companies with limited AI capabilities will be willing or able to attract and retain such talent.
Moreover, even if they could, at this point it is unclear how companies could implement AI applications that would enable them to earn a high return on their investment.
Hosanagar thinks that the cost of building AI applications will drop as they did for iPhone apps. As he said, "When the iPhone first came out, it cost $500,000 to $1 million to build an app -- now the cost is $25,000 to $30,000. The costs of AI applications will drop -- in part with help from open source technology."
He also suggests that companies start small. "Companies should not try to build AI applications that will boost revenue or reduce costs right away. Instead they should set a goal in the first 12 to 24 months of increasing their organizational learning and expect to build high payoff AI applications over five years," said Hosanagar.
Having lived through a previous wave of enthusiasm for AI in business, I am conditioned to be skeptical about whether reality will live up to the hype.