Help Wanted: Turning AI Into Reality

STEVE HENDERSHOT

Artificial intelligence is officially mainstream—sparking big changes in everything from manufacturing and supply-chain management to risk monitoring and product development.

MANOLIS KOUBARAKIS

No stone will remain unturned, I think that’s what we should say, because we are really seeing a revolution. It all started about 10 years ago with breakthroughs in image processing, speech recognition, natural language processing using machine learning, basically deep learning, in particular. But now, everybody’s jumping in this exciting wagon, and I think we will see a lot of products being essentially produced using AI technologies.

STEVE HENDERSHOT

But do companies have the talent they need?

SNEHANSHU MITRA

We know that there are a number of innovative ways in which organizations today are trying to tackle the talent shortage, but it is there. It is something that needs to be addressed. And without the right kind of resources and skill sets at your disposal, no matter how sound your project management practices are, they’ll probably never see the light of the day.

NARRATOR

The world is changing fast. And every day, project professionals are turning ideas into reality—delivering value to their organizations and society as a whole. On Projectified^®, we’ll help you stay on top of the trends and see what’s ahead for The Project Economy—and your career.

STEVE HENDERSHOT

This is Projectified^®. I’m Steve Hendershot.

The Artificial Intelligence Age is here. Sure, the technology continues to evolve, but there’s no doubt it’s gaining traction across industries: 46 percent of project leaders say their use of AI has increased over the past year, according to PMI’s Pulse of the Profession^® report.

So what does it take to build teams that can make a difference in the AI space? Today we’ll explore that by talking with a couple of people working in the emerging field, beginning with Manolis Koubarakis, a professor at the University of Athens and the project coordinator of ExtremeEarth.

ExtremeEarth uses AI to study satellite images and translate that visual information into data that can help researchers understand—and even anticipate—environmental changes that will affect food security in Europe, as well as affect the polar ice cap. Let’s hear about how he’s finding the people who can turn that vision into reality.

MUSICAL TRANSITION

STEVE HENDERSHOT

Walk me through how this works—how you use AI to go from satellite images from the EU’s Copernicus Earth observation program to data extracted from those images to, for example, practical advice for farmers based on that data?

MANOLIS KOUBARAKIS

So essentially, the whole project can be understood as a pipeline. You start with satellite data, and then you want to develop machine learning algorithms, deep learning algorithms, that will be able to extract information from this data. In the case of food security, what we were using these deep learning algorithms to do was to identify crops, basically. Because then this data about crops together with other kinds of data—for example, models that can capture the evolution of plants over time and other models that describe the availability of water in a specific region and other data about that—would result into the irrigation recommendations.

One of the good contributions I think of ExtremeEarth has been not only to develop new machine learning algorithms for dealing with problems like crop type mapping, but also providing very large data sets for training such algorithms. And we make these data sets free and open to everybody so other researchers can also use them. And we do the same also for the polar regions use case that we have. We provide open and free satellite images that capture information about ice and its state in the polar regions. And this is again provided for free for other people also to use in order to train machine learning algorithms.

STEVE HENDERSHOT

What does a successful project team look like in this context—what are the different skills that need to be represented to ensure you’re converting this upstream satellite imagery into useful downstream insights?

MANOLIS KOUBARAKIS

I think what is important with respect to what you’re saying, Steve, is to get the right team for the right project. So in our case, for example, we have the people that work with satellite data and extract information from it. Then we have the people that actually work with this information. In our case, it’s people like the people in my group working with semantic technologies and big data technologies.

And then we also have the people that have the application. In our case, it’s companies like Vista in Germany, a company specializing in geo information for agricultural applications. We also have the Meteorological Institute of Norway and Polar View, one institution and one company that specialize in polar applications. So once you get these pillars, let’s say, for setting up the project, then if you make sure that the communication channels are set properly, then a whole lot can be done in a very easy kind of way. And you can see, as you say, things that come from upstream ending up in a nice downstream application.

STEVE HENDERSHOT

In a consortium like yours, with lots of different academic institutions and commercial interests involved, how do you structure the project teams so that they’re operating cohesively and moving the whole effort forward?

MANOLIS KOUBARAKIS

The good thing is to have the whole project team at the beginning being organized in a coherent and good way. What I try to do in the projects that I do is I try to have partners that complement each other because if you have partners working on the same thing, you might see people competing instead of collaborating. And this is never a good thing if you want the project to succeed.

Now, inside each team, I mean, I can talk about my team at the University of Athens. I always try to have some senior people working with junior people. And essentially, the junior people typically learn from the more senior, and then they can become senior people in a forthcoming project. So this is my recipe for success inside my own team.

STEVE HENDERSHOT

What are the greatest challenges to successfully executing a large-scale AI project like the one you’re working on?

MANOLIS KOUBARAKIS

There are some very specific things that one has to be aware with. For example, you have to find the right kinds of data for what you want to achieve. And once you have this data, then you also have to have the right technical people to develop algorithms that will work with this data. And then people that will also be able to develop the application that will be based on what you extract in this data. So you need to have very good data scientists, and you need to have very good software developers sitting together in teams and being able to deliver the product that you want.

And for me, a very good thing that we do often in projects like ExtremeEarth is to have these teams basically working together and doing hackathons every now and then, where they sit together and basically, they implement a data science pipeline from the beginning to end. And in that way, we are testing what each group is developing, whether it actually satisfies the end goal, and in that way we make progress.

STEVE HENDERSHOT

What’s your advice for teams and organizations looking to grow their AI maturity? What sort of structure and approach and expertise is needed to draw maximum value and understanding out of the technology’s potential?

MANOLIS KOUBARAKIS

What people need to understand is that we are still in a position where we are able to develop very good value through AI. But we are not still at the position where we can solve all the problems that anybody has. So in some sense, we need to be careful that we don’t overemphasize these technologies and expect everything from them. I mean, it’s just a new technology. And like new technologies, it might look very impressive at the beginning, but then when you see the details, you might see that it didn’t really satisfy the requirements that you have, and you have to go back and ask for a better version, work with your provider again to give feedback so you get a new version of the technology.

I think for AI technologies, like every other [technology], people need to be patient, and they shouldn’t expect everything to be delivered from the first start. It’s going to be a slow process, and you have to go through iterations to get it right.

MUSICAL TRANSITION

STEVE HENDERSHOT

Even as AI has hit an inflection point in its maturity and adoption, there’s still a role for education and advocacy as companies and teams work to scale up their AI capabilities. That’s what Snehanshu Mitra does as CEO of the Centre of Excellence for Data Science and AI at NASSCOM, a tech trade group in Bengaluru, India that worked alongside PMI India to produce the Playbook for Project Management in Data Science and Artificial Intelligence Projects. Projectified^®’s Hannah Schmidt asked Snehanshu about the ecosystem characteristics he believes will help AI reach its full potential.

MUSICAL TRANSITION

HANNAH SCHMIDT

AI is booming. So what are the must-have skills project leaders need to make the most of it?

SNEHANSHU MITRA

From project managers’ perspective, there’s this whole need to work in a cross-functional environment, to be able to collaborate with a wide range of teams. These are of the kind where you have probably got direct control over, and some that you may not exert direct control over. So how do you work with these diverse teams and exert this kind of soft impact on them? It will be extremely important as a skill to have.

The other thing that I can think of is this whole aspect of communication, and the more we become digital in our communications. Communication as a competency is becoming ever so important. You are now trying to communicate with a diverse set of stakeholders, to inform them of the challenges, to understand where the problem statement is coming from, to inform them of the milestones, to explaining the iterations and what is the increment that has happened from the last time around. All of these would include a lot of active communication.

The other thing that I can think of is just coming back to the whole point around data. Estimating the impact of data quality I feel is very, very central to the role that a project manager in the data science and AI space will have to be good at.

HANNAH SCHMIDT

One of the challenges project leaders are facing in delivering AI initiatives is talent shortage. Where are the biggest gaps, and how can companies solve that problem?

SNEHANSHU MITRA

AI project development is unique in that you require a multitude of skills to successfully accomplish a project. The type of talent most in demand is fairly uniform across enterprises that are established or they are developing or they are just getting started. Now, the talent that I’m talking about is around AI developers and engineers, AI researchers, data scientists. These are the kind of skill sets that are highly in demand and which also means that, at times, we are looking at a scarcity of the right kind of talent to don these roles.

I was going through this interesting survey that Deloitte conducted on AI talent gaps. Although most of the seasoned enterprises reported relatively less gap between the needs and abilities, about a fourth of the respondents said that they had a major or extreme sort of a gap when it comes to reporting talent shortages. Now, one theory is that the seasoned enterprises, having worked with AI technologies more extensively, now know what skills they actually need and what they think they need. And the other thought process here is that maybe once you’ve reached that far, you are pursuing more transformational projects, rather than just focusing on cost reduction. You are probably looking at creating new products, new services, and hence you’re looking at different kinds of skill sets, and that is creating a certain kind of shortage.

The point that you asked was very interesting. How do you solve for that? Well, the future of work will require employees to be agile and change their skills over time. We are seeing companies that are focusing on retraining, on upskilling current employees, and this is a trend that is likely to increase. The second thing that we are looking at is, can we make some smart compromises? In the sense that, can we try and induct people with good base skills? When you talk about the quantitative side of artificial intelligence, it has got a foundation in mathematics, statistics, in certain aspects, even physics, right? So some AI positions could be filled by those with mathematics or statistics or physics degrees who have a certain kind of aptitude toward getting into this particular field.

The third thing that I can think of, although this is relatively new, Hannah, but this is happening, which is, can AI skill gap be solved by AI? We are looking at Google developing AutoML, that is an AI program that can create other AI systems. Consider the implications if this initiative is actually successful. It could be very powerful for a lot of organizations, especially ones that are in the startup space or midsized space that cannot afford a lot of high-cost talent, right? This will also end up creating a more level playing field with larger enterprises.

HANNAH SCHMIDT

NASSCOM worked with PMI India on a playbook for project management and AI. And it mentions the dependence on high-performing talent, or “wizards” as they’re called in the playbook. How can project leaders get around this need?

SNEHANSHU MITRA

One of the big challenges, Hannah, is around inefficiencies, right? And as the study that we did together indicated that a significant portion of AI projects are not efficiently managed, which increases the cycle time. It becomes more resource dependent and costly. I feel that sound project management principles and practices will ensure that the entire workflow, right from the problem discovery to the conceptualization of projects to the development and implementation, is effectively managed, and the output per unit of effort in some ways will be maximized. So that is one way in which project management principles will be able to come to the rescue of managing resource shortages.

HANNAH SCHMIDT

So in your role, you’re working with organizations on AI projects and skilling and reskilling talent to execute them. What are three lessons learned for project teams working on AI initiatives?

SNEHANSHU MITRA

Enterprises today are concerned that only a fraction of their AI projects will have real business impact, measurable business impact. In some cases, investments that are made in AI projects are put under scrutiny. The projects do not take off or they’re abandoned when the implementation is not in line with the project objectives, etc. So I think a couple of things that I would propose as lessons learned in some ways, also going back in my own experience, first thing is, does the problem statement actually involve AI? I’m saying this with a sense of responsibility that AI projects are significantly resource intensive. They are costly to manage. They’re time-consuming. So can it be solved through other means? Question the economic value that a project is expected to create. That is the first thing that I can think of.

The second thing is, can we start small? Now, oftentimes AI problem statements are not very sharp, precise. These are open-ended statements. So once we have gone through sort of constructing it the right way and breaking it down, can we pick up very specifically that component of the problem statement that is really, really well-defined? And AI technology’s more complicated than we think. It will only help if we start small, pick up a really well-defined problem and then apply the AI project flow. I feel that way is creating more chances of us being successful.

Going back to the basics about data, I feel a big lesson for me personally, Hannah, is that data preparation is a far more important task than we have traditionally thought it to be. An average model with good quality data will any day beat a good model that is built on a data which is suspect. So go back and focus on the whole data strategy, data preparation, data quality.

MUSICAL TRANSITION

STEVE HENDERSHOT

AI has evolved from a mysterious futuristic technology into something that is powering new insights and efficiency gains—right now, today. The next move belongs to the organizations and project leaders assembling the talent that can harness AI’s vast potential.

An amazing tool is at your disposal … what will you build with it?

NARRATOR

Thanks for listening to Projectified^®. If you like what you heard, please subscribe to the show. And leave a rating or review—we’d love your feedback. To hear more episodes of Projectified^®, visit Apple Podcasts, Google Play Music, Stitcher, Spotify or SoundCloud. Or head to PMI.org/podcast.