Policy Punchline is a podcast, created, produced, and edited by a group of very talented Princeton University students – led by Tiger Gao. The podcast promotes long-form dialogues on frontier ideas and urgent issues with scholars, policy makers, business executives, journalists, and entrepreneurs.
Policy Punchline with Martin Fleming
The following is the portion of the Policy Punchline podcast on AI and Automation.
“My punchline is we may be only four or five percent of the way down a path that’s going to take 20 or 30 years for us to fully take advantage of the capabilities. The notion that artificial intelligence is about to take over the world is nonsense. It really misses the challenge and the difficulty that organizations face in deploying these kinds of solutions.”
Martin Fleming was formerly the IBM’s Chief Economist and the head of IBM’s Chief Analytics Office. His research focuses on artificial intelligence, the future of work, and digital currencies.
Q: In your recent report, “Cognitive Enterprise,” you encourage businesses to rip up the playbook and make changes within their organization to accommodate the introduction of AI. How exactly can AI help businesses grow? Many criticize AI for overpromising its capabilities, and it’s mainly helped in areas such as marketing analytics but has done little in medical diagnosis, for example.
A: I’m sure that all of the listeners to the podcast today have used artificial intelligence. The simplest artificial intelligence application is when you’re typing a text message and your phone predicts the word that you’re trying to type. That is a small, simple artificial intelligence machine learning application.
Any application that is attempting to help you and predict the outcome of the action that you’re taking is, in principle, a machine learning or an artificial intelligence application. Another example is when Netflix makes a recommendation to you as to what video you might like to watch. The people at Netflix have done a great deal of artificial intelligence work to be able to provide that recommendation to you and hopefully improve your satisfaction with their service and enjoy whatever video you end up selecting. It’s all about helping to make decisions and producing better quality outcomes.
Q: You hear a lot of fantastic stories about companies like Google and Amazon reaping the benefits of AI, but are there any more conventional “brick and mortar” establishments or sectors that might not have embraced AI in the past, but that might benefit from its implementation in the future?
A: You’re right that a lot of the early applications have been in the technology industry. A close cousin of the technology industry is the financial services industry, where a lot of AI applications are beginning to emerge. These applications range from making personal finance recommendations to consumers to making recommendations to investors. Likewise, in insurance, many of the large property and casualty insurance firms are attempting to help both themselves and consumers by making better decisions.
A smaller area is the pharmaceutical industry, where a lot of pharmaceutical research is being done by data scientists using the applications of AI to look at the chemical compounds that come together to produce new drugs. Using the traditional physical efforts is a time-consuming process, and a number of the potential combinations can be eliminated through the use of artificial intelligence, helping the pharmaceutical firms arrive at a successful combination of compounds more rapidly. These are some of the areas where we’ve already seen some applications, and the work is continuing.
We’re on a long path here. We may be only four or five percent of the way down a path that’s going to take 20 or 30 years for us to fully take advantage of the capabilities. The notion that artificial intelligence is about to take over the world really misses the challenge and the difficulty that organizations face in deploying these kinds of solutions.
Q: If automation is fairly easily accessible for employers, and if it’s harder to bring employees back to work due to Covid-19, do you think employers may simply choose to automate the jobs and not bring workers back? It seems that this would be especially likely if the government subsequently chooses to raise the minimum wage or raise corporate taxes in the aftermath of Covid-19, as some labor economists have argued.
A: Automation doesn’t happen like that; it’s just not how it works. When businesses automate a business process, they first ask, “do we need the talent?” They need the data scientists; they need the developers; they need the folks with business acumen and strategy skills so that they can understand the business process.
Second, the business process has to be transformed. If the process is not working and a company wants to introduce some automation, it shouldn’t attempt to transform the process through automation. Rather, it should introduce the automation in the process of transformation. Third, companies have to change the behavior of individuals. It becomes a change management challenge, because nobody likes to have to change the way they do things.
All of that – the talent, the business process, the transformation, the change management – has to happen after the technology has been put in place. The shift is difficult. I would assert it’s not the most difficult piece of it, but it is still difficult nonetheless. Even a large organization like IBM thinks of itself as having dozens of business processes, all needing to be transformed. It doesn’t happen quickly. It takes time to do all of this change that’s occurring. You can’t just snap your fingers and wish for it to change. It’s real work.
Q: Where do you think data science and AI education should begin? Has IBM focused on changes in college education? What about on the high school level?
A: There are certainly high school students that are learning how to code in Python, which is a great first step. We’re seeing more and more of that. One of my roles at IBM is to lead the data science profession, and one element of the data science profession is a certification that we’ve created with an outside third-party group, so that those who get certified within IBM will also have that recognition externally.
We’re now in the early stages of working with two universities to begin to introduce that certification capability into their academic programs, so that students who are studying data science can earn a certification and learn about the recruitment and hiring process. We’re beginning to see more and more of that activity at the undergraduate level.
Q: When we interviewed Iwao Fusillo, the chief data officer at the NFL, he said that there will be more chief data officers who take on CEO roles in the future. Satya Nadella at Microsoft, Sundar Pichai at Alphabet, Shantanu Narayen at Adobe — and just recently from April onwards, Arvind Krishna at IBM… These are four Indian-born executives who were trained as engineers and rose through the ranks in technical positions. Their backgrounds are very much different from the stereotypical corporate America managers’ background – in sales and general management after receiving MBA degrees – such as Ginni Rometty, IBM’s previous CEO for the last eight years. Do you think we’re seeing a fundamental sea change, such as tech companies, or even companies in general, are better managed by people trained in more technically competitive backgrounds? In the age of “Big Data,” can you really manage a corporation well without knowing the tech yourself?
A: Absolutely. The combination of data science and business acumen is really what we’re looking for in data science. Now, there is no Renaissance person who has the whole package of skills, so you have to team people together. But, over time, we hope to develop those skills so that the folks who bring the data science skills can learn the business skills, and the folks who come with the business skills can learn the data science skills. We all have our strengths and limitations, and through experience, a company can help to even those things out.
Q: You spoke about how AI is beginning to take hold in emerging economies. How will these advancements affect workers in developing countries? What about lower-income workers in developed countries?
A: That is an interesting question because the technology could possibly have a differential impact across different regions of the world. In the United States, Western Europe, and Japan, the impact has been more on the mid-wage workers. We use the term job polarization, where the low wage and the high wage workers are where more employment is appearing, and it’s the middle wage workers that have lost employment share.
In the developing and emerging market world, where low cost labor has been important, many of these low-cost roles are perhaps more easily automated. One example is call center work, which we see quite a bit of in a country like India. As more and more natural language processing capability, both for voice and for text, becomes available, there is likely to be less demand for call center workers across many of these countries.
So, the impact could be quite different across different geographies. In the U.S. and Western Europe, it will likely be more of an issue around the distribution of income and wages, particularly for middle income workers, whereas in emerging market economies, it may be more of an issue of the share of employment for low wage workers.
Martin Fleming's Blog 