By Josh Krist, Staff Writer, Workday
Artificial Intelligence. In the world of business buzzwords, this one seems to be at the top of the list. Everyone’s talking about it, but how many people actually understand it? Ajay Agrawal has made incredible strides in this area. The 2018 book, “Prediction Machines: The Simple Economics of Artificial Intelligence,” authored by Agrawal, Joshua Gans, and Avi Goldfarb, provides business leaders with actionable advice on how to realize the value of artificial intelligence (AI).
Much like cheaper light and power thanks to electricity or cheaper calculations from computing, the book’s authors explain that better, faster, and most importantly, cheaper predictions will get rid of uncertainty in business models and processes, and will lead to reimaging whole industries. Bottom line, AI promises to be a transformative, general use technology with manifold ramifications.
We met with Agrawal on a foggy afternoon in Half Moon Bay, Calif., during a break at 2019 Forbes CIO Summit, and chatted about some of the key learnings from his book. Excerpts from our conversation are below.
Most business leaders don’t have computer science backgrounds. What’s your best advice for them on how they can leverage artificial intelligence?
My best advice is for them is to simply think of AI as a drop in the cost of prediction. When prediction, or anything else, becomes cheaper, we use more of it, and we start to use it in more ingenious ways.
When I’m meeting with CEOs and CIOs, they often say things like,”We employ 25,000 people and we’re in this line of business. Where should we start with AI?” The answer is usually quite straightforward—your data science group. That team has already identified your company’s prediction problems, has data in a digital format, and has integrated those predictions into the workflow. Now, all they have to do is drop in some new statistical techniques that, if there’s enough data, will generate better, faster, and cheaper predictions.
AI is a substitute for only one thing, and that’s human prediction.
Another way to look at AI is by recasting problems we didn’t used to think of as prediction problems in a way we can tackle them with AI. For example, we have quite a few HR leaders from large organizations that come to our Creative Destruction Lab in Toronto, and they’ll say, “We want to figure out what skills to look for, who we should hire, and how to upskill our existing people.” Then in a separate conversation, we’ll hear business leaders say, “We need AI in sales, marketing, and manufacturing. We need AI in most places—except for HR.”
Most people think that because HR is very human and requires a lot of emotional intelligence, it doesn’t need AI. That’s a mistake. People can leverage AI by converting HR functions, such as certain aspects of recruitment and skills development, into a series of predictions, where humans can then apply their judgment.
AI is also viewed with some fear, especially when it comes to the future of jobs. What would you say to that?
This is really important because a lot of people feel very threatened by it. But, AI is a substitute for only one thing, and that’s human prediction.
All human prediction is susceptible to being replaced by machines. But, there are many other valuable things that humans do that are complements, not substitutes, to prediction. One area, as we just talked about, is judgment—humans have it, AI doesn’t. Everywhere humans are deploying judgment, the value of that judgment goes up because we get to apply it to increasingly higher-fidelity predictions.
Here’s an analogy: Imagine two accountants interviewing for the same job before spreadsheets existed. One of them is very good at adding up numbers in his head very fast and accurately. And the other accountant has good judgment; he’s good at asking smart questions like, “What happens to our business when interest rates go up by a quarter point?”
An interviewer may have said to the first accountant, “You have a valuable skill of adding up numbers quickly and accurately in your head, so you’re going to save a lot of time.” With the second person, the interviewer may have said, “It’s great that you’ve got this judgment skill, but every time you would come up with one of your clever questions, it would take us three days to answer it. That’s an interesting skill, but it’s limited in how valuable it is.”
Then along come spreadsheets, and the first guy’s value falls because the machine equalizes. It doesn’t matter if you’re fast or slow at adding up numbers in your head. The machine is faster and more accurate than any human.
Yet the second accountant still has good judgment. His value goes up because now every time he’s got a clever question, instead of taking three days to answer the question, he can just change one cell on the spreadsheet and find the answer.
But can’t judgment itself be quantified? In other words, making a judgment is just a prediction of a different type; a second- or third-order prediction, right?
Yes, judgment can be quantified. If the machine gets to observe enough of us taking actions in response to its prediction, then it can start inferring what our judgment is and predicting that. But, it’s always going to be an arms race. As AI gets to observe many judgments, and it turns those into predictions, then we can apply judgments to those predictions. Judgment will remain a human’s game.
You talk about “AI moments” in your book. What’s a recent AI moment you’ve had?
I was recently in Tokyo and was amazed at how many big international companies conduct all of their business in Japanese. Even their senior executives didn’t speak English, so many times there had to be an interpreter present during meetings. And I thought, “Wow, we really are quite separated linguistically. When we finally achieve a commercial grade translator that can translate instantaneously, these barriers are going to fall.”
To learn more about AI and machine learning, be sure to check back for an upcoming Workday Podcast, where we sit down with Ajay and Workday machine learning expert Sayan Chakraborty to dive deeper into how machine learning is affecting business today and in the future.
This story was originally published on the Workday blog. For more stories like this, click here.
Follow Workday: LinkedIn, Facebook, and Twitter.
Credit: Google News