(Podcast) Cognitive and AI survey | Deloitte Insights

State of AI in the Enterprise, 2nd Edition Early adopters combine bullish enthusiasm with strategic investments

November 28, 2018

When implementing AI technologies, companies face social, ethical, and logistical challenges. Tom Davenport, of Deloitte’s analytics and cognitive practice, says a deeper understanding of AI and a pragmatic approach to its adoption can show a way forward.

 
Most industrial robots are not very smart right now, but they will get smarter in the future and we’ll have AI brains... more and more, the purpose of us humans is to provide training data to machines. (Laughs)

Learn More

​Read the report

View the infographic

Browse the AI & cognitive technologies collection

Subscribe on iTunes

Listen to more podcasts

Subscribe to receive related content from Deloitte Insights

TRANSCRIPT

TANYA OTT: I’m Tanya Ott and this is the Press Room, where we talk about the issues that are or should be important to your business. Today we’re talking technology.

In October, the Massachusetts Institute of Technology announced it’s going to spend $1 billion dollars—that's with a big capital B—to create a new college focused on Artificial Intelligence.1 That is a huge investment. MIT says it’s already raised two-thirds of the money and plans to start classes next fall. In announcing the school, MIT’s president said he wants to “educate the bilinguals of the future.”  What he meant by that was people who are schooled in the hard sciences, politics, history, and linguistics and are also skilled in applying modern computing technologies. It is a brave new world, although maybe more Shakespearean than Huxley-an.

Folks who’ve been paying attention know that Artificial Intelligence, or AI, isn’t just futuristic fiction. It’s in our daily lives and it raises all kinds of social and ethical questions. And today we’ve got a guide to walk us through those questions in a business context.

Tom Davenport is a professor of information technology and management at Babson College, research fellow at the MIT Initiative on the Digital Economy, and an independent senior advisor for Deloitte’s analytics and cognitive practice. He spent decades work in big data and analytics and more recently, Artificial Intelligence.

I asked him to start with some examples of AI that are in products we use every day.

TOM DAVENPORT: Anyone who uses a variety of Google-oriented products [knows it], from search to maps to certainly Google Home has various forms of AI built in. If you have an Amazon Echo and communicate with it—I don't want to say her name because there's one on my desk and she'll light up—so those are, I think, the most familiar, consumer-oriented offerings. But AI is soon to be incorporated in a whole variety of products. I was just interviewing this morning a company that does AI recommendations for banking that would tell you—you really need to move money from one account to another to avoid overdraft fees, or if you're trying to accumulate some savings, you need to spend less on this kind of thing. So, I think we'll see a lot of places where consumers will benefit from AI in the near future.

TANYA OTT: So that's the consumer space. What about industrial applications or maybe more back-end business applications that consumers don't directly interact with?

TOM DAVENPORT: Still in financial services, there's a lot of AI used for credit and fraud applications. If you're traveling around and there's an attempted fraudulent use of your credit card, for example, that probably [will] be detected through AI.

In industrial companies there's increasingly a lot of machine learning for predictive maintenance to fix things before they break and to identify anomalies in the performance of industrial devices. You have this idea of a digital twin for machines and whole factories that will analyze patterns in performance of machines and identify ways to improve them. And, of course, robots, I think, many people would define as AI. Most industrial robots are not very smart right now, but they will get smarter in the future and we'll have the sort of AI brains that we are increasingly seeing in other domains.

And then any sort of personalized recommendation from a marketing perspective is probably going to be driven by some form of AI. And even the idea of a credit score—that's probably the oldest, consumer-visible application of AI. It's been going for 25 or 30 years or so where machine learning is used to assign basically every adult consumer in the United States and many other countries a credit score.

TANYA OTT: So, you just mentioned machine learning, and Deloitte conducted what's called the Cognitive Awareness Survey, where you talked with companies about the current technologies that they're using, and machine learning was one of the most popular technologies amongst companies. What exactly is it?

TOM DAVENPORT: It's basically a technology that can learn from existing data, typically what's called “training data,” and then once a model has been produced that is very effective in that regard, it can be used to predict or classify data for which we don't know the outcome. You know, is this likely to be a fraud attempt? Or, is this customer likely to respond to this product offer? At its simplest, machine learning is really identical with predictive analytics, but there are many, many types of machine-learning applications—deep learning for image recognition and voice recognition is an increasingly popular technology. It's a much more complex set of algorithms and models than you would find with, say, credit scoring.

TANYA OTT: You mention visual recognition and audio recognition. As soon as we're done with this interview I'm going to take the audio from this interview and I'm going to throw it into an online application that does a transcription for me. And those transcriptions have gotten better and better over the last probably two to three years that I've been using it, and I think it's because everybody who uploads their audio, if it transcribes it with the wrong word we go in immediately and we correct that word, and I'm guessing the machine learns that. And that's how it becomes so much better at transcribing.

TOM DAVENPORT: Yeah... you're producing training data without necessarily realizing it. Google, for example, will now get you to label images. Supposedly [it’s] to verify that you're a human, but at the same time you're also providing training data for them in their image-recognition programs. So more and more the purpose of us humans is to provide training data to machines. [Laughs.]

TANYA OTT: What's your sense of why companies are choosing to deploy certain technologies? What are they hoping to see in return for the deployment of AI?

TOM DAVENPORT: That's a good question. Some AI technologies offer much greater ease of implementation and typically higher and faster returns on investment. Robotic process automation, which is not the smartest form of AI, but one that can perform structured digital tasks in an automated fashion ...  [for example] maybe replace lost ATM card for you or do some movement of information from one system to another in an automated way. It's not very smart, but it's also not very expensive; it's quite easy to implement. And so that would typically be a bit of low-hanging fruit that companies could implement and be pretty likely to see a quick return. Some of these more complex forms of AI technology require greater levels of expertise and would probably take a bit longer to pay off.

TANYA OTT: Speaking of paying off, I was reading a recent report that said AI could contribute more than 13 trillion dollars in global economic activity by the year 2030.2 That's only about 12 years from now, so we're talking about a real explosion in this sector. What kinds of new AI applications are in production right now? What are you seeing piloted, for instance?

TOM DAVENPORT: Fortunately or unfortunately, there's a lot being piloted. We aren't at the stage where a whole lot of companies have put in fully production applications yet, which is a bit unfortunate. It's relatively easy to pilot, but production applications typically require a lot of big changes in how companies do their work and may require some retraining of the part of employees and so on. It's quite common, for example, to see intelligent agents and chatbots being used within companies right now for employee purposes, but many companies are not quite ready to turn their customers over to them yet as a mainstream channel for customer interaction. We'll see those in the future as they get smarter and better trained and so on, but right now they're mostly being piloted for employees to check their vacation balances or get new passwords for their I.T. accounts.

TANYA OTT: You mentioned that robotic process automation is one of the easier things for companies to deploy. You refer to it as “low-hanging fruit” and you've got some great examples that you call “The Tale of Two Ambitions,” where you compare the low-hanging fruit with the “moonshot” by various companies. One of them is a major cancer treatment center. What was their low-hanging fruit and what was the moonshot? And then tell us what worked ... or what didn't.

TOM DAVENPORT: The low-hanging fruit were a set of applications related to some fairly prosaic ambitions like where should families stay and eat when they have a family member being treated at this institution, and who might have some trouble paying their bills. Predicting that in advance, you can start to work with them on a repayment options. Those all worked pretty well. The highly ambitious moonshot—and this organization actually referred to it as a moonshot—involved treating cancer. Basically, identifying certain forms of cancer and then recommending treatment for it. And that didn't work out so well. A lot of money was spent—I think $62 million is the agreed upon figure by the internal auditors—and a patient was never treated and no integration with the electronic medical records system, and a lot of organizational upheaval as a result of this sort of failure.

I generally feel AI is very task-oriented. It doesn't support entire jobs or processes, so it's probably better suited right now for a series of low-hanging-fruit projects. Each individual one may not be that incredibly impressive, but if you do a variety of things within a particular domain—say, customer relationships—you could actually make a fairly substantial improvement in it and maybe even dazzle a customer now and then; but in general technology is better suited to that kind of less ambitious application.

TANYA OTT: What I'm hearing you say is that the ambitious application, the moonshot, is perhaps really interesting but it's going to take a lot of money and it may not work, whereas if you do smaller, low-hanging fruit or a collection of them, you might get more bang for the buck, at least immediately.

TOM DAVENPORT: Exactly, and even at Amazon, which has arguably some of the strongest technology on this planet, Jeff Bezos said in his 2017 letter to shareholders that his primary focus with AI has been on these low-hanging-fruit applications that, I think he called it, “quietly but meaningfully improve core operations.”

TANYA OTT: One of the reasons that a lot of people are excited about AI is that many jobs can be automated or portions of jobs can be automated. But that can be a really scary idea for a lot of people because, on the one hand, all of us have routine tasks we'd be happy to give up to something else that's automated, but not if it means we're going to lose our jobs. So, there are a lot of ethical and social implications in the deployment of AI. How are you thinking about those things these days?

TOM DAVENPORT: Well, I've thought about it a lot. I wrote a book about it a couple of years ago called Only Humans Need Apply, which came out pretty strongly in favor of augmentation—humans and machines working closely with each other—rather than large-scale automation. And I still feel that way. Most of the surveys we've done at Deloitte have suggested that automation is a relatively low priority when you ask cognitive-aware managers what their objectives are with AI. But, in the latest 2018 survey, I think 63 percent said they were intent upon automating as many jobs as possible.

TANYA OTT: Wow!

TOM DAVENPORT: It was a relatively low priority, so that was a little scary. Most of the results thus far have suggested augmentation is a much more likely outcome than large-scale automation.

TANYA OTT: As we're looking at people that are what you might call knowledge workers and parts of those jobs being automated, you talk about a list of people that might have parts of their jobs automated: teachers and professors, lawyers, accountants, radiologists. I got to “reporter” on the list and it terrified me because the automated part of the job you identify is story writing. And as someone who's made her career, spent 30 years, writing stories, the idea of some automated system writing a story better than me. I just don't know. I don't know....

TOM DAVENPORT: Yes ... as you probably know the Associated Press does a lot of its story writing with an automated tool.

TANYA OTT: Yeah...

TOM DAVENPORT: If you're a reporter, the key is to focus on creative, feature-oriented writing, not on fairly structured writing that involves data. And the same could be said about financial advisers. Delving deeply into a customer's financial psyche is not going to be automated, but telling them how their investments performed last quarter probably is in the vast majority of cases.

TANYA OTT: I'm curious about the teacher side of it, because we think of teaching as such a hands-on and reflective practice. But I did talk to someone a couple of years ago that was running an experiment where they had video cameras in a classroom that track the movement of students, and would then translate that information into data related to which student seems to be confused, which student seems to be distracted, and notify the teacher that they might want to pay attention to that student at that moment in that classroom. I'm guessing that might have been some sort of AI technology that was taking a video image and interpreting body movement and subtle gestures to suggest which students might be struggling?

TOM DAVENPORT: That would be further advancement of technology in teaching than I've seen thus far. What I have seen is these adaptive learning technologies that if you're doing some sort of online content acquisition exercise, it can see who's proceeding at what pace and so on. And frankly, I think that is quite impressive and something that you know for a fairly large classroom human teachers would find difficult to do. There are systems that pay attention to whether you're bored or paying attention, but I haven't seen them installed in classrooms yet. Early days for that technology, I'd say.

TANYA OTT: Okay. One group of people who probably won't be losing their jobs are the folks who are actually designing and working in AI. It's interesting that recently MIT announced it was going to spend $1 billion dollars—that's with a big capital B—to create a new college focused on AI. That is a huge investment.

TOM DAVENPORT: It is, but I think it's fair to say that even hardcore data science types have some risk. There are these automated machine-learning technologies now that automate a fair amount of what we previously did in creating machine-learning models. None of us can afford to be totally complacent. I'm sure the graduates of that MIT program will do fine for a while, but even they have to embrace new technologies and figure out how to add value to them.

TANYA OTT: What challenges do companies face as they're looking to implement AI? What are your parting thoughts that you think leaders should be thinking about?

TOM DAVENPORT: In many cases it's the usual issues with systems. It's how do you integrate this into other systems and processes. There are a lot of data-related issues that companies say are critical. One, though, that's somewhat more easily addressed is just managerial awareness and understanding of these options. In addition to listening to this podcast, I think it's important for managers to understand what are the different AI technologies, how do they translate into different types of use cases, and how can they really benefit their business. That would be a bit of low-hanging fruit from a challenge standpoint that many companies could address pretty easily.

TANYA OTT: Thank you very much for your time. I really appreciate it.

TOM DAVENPORT: My pleasure. Glad it worked out.

TANYA OTT: Tom Davenport is a professor of information technology and management at Babson College, research fellow at the MIT Initiative on the Digital Economy, and a senior advisor for Deloitte’s analytics and cognitive practice.

If you want to learn more about applying AI in your business, there are a ton of resources in our archive, at deloitte.com/insights.

We’ve love to hear from you. Tweet us at @deloitteinsight.  I’m at @tanyaott1.

I’m Tanya Ott.  Have a great day.

This podcast is provided by Deloitte and is intended to provide general information only. This podcast is not intended to constitute advice or services of any kind. For additional information about Deloitte, go to Deloitte.com/about.