The Evolution of AI Talent and Training (AI Zeitgeist 3 of 7)

Daniel Faggella is the founder and CEO at Emerj. Called upon by the United Nations, World Bank, INTERPOL, and many global enterprises, Daniel is a sought-after expert on the competitive strategy implications of AI for business and government leaders.

The Evolution of AI Talent and Training - AI Zeitgeist 3

This article is part 3 of a 7-part series called “AI Zeitgeist,” where we’ll be mapping out the details of AI adoption over the next 10 years and explore the critical changes in the AI ecosystem that business leaders need to understand.

In this installment of the Zeitgeist series, we’ll be talking about how the accessibility of AI talent and the requirements for training employees on the basics of AI functionality will evolve as we move from the “Emergence” phase, where AI talent is exceedingly rare, to the “Dispersion” phase, where the basic concepts of AI are as well understood by the average employee as the internet is today.

(Note: If you haven’t read the first article in the AI Zeitgeist series, you should familiarize yourself with it first. In that article, I explain what the 3 phases of “Emergence,” “Adoption,” and “Dispersion” actually mean).

Making artificial intelligence work in business is about more than data and algorithms; it’s about people and skills.

While there are many types of skill sets and company culture elements we could address under this broad umbrella, we’ve decided to discuss three critical “talent concerns” around AI in the enterprise to show how these dynamics will change over time as industries progress through the phases of AI in the enterprise:

  1. Engineers familiar with productizing AI: Developers and engineers who not only have experience “tinkering” with AI projects, but who have experience turning them into products and features that in-house users (people within the company) or customers (outside users) can actually use to reach their goals. We’ll talk about how rare this combination is, and how these skills will evolve.
  2. Organizational leaders with a firm grasp on AI: Board members, C-level execs, and prominent government officials will one day have a firm grasp on the general capabilities, advantages, and limitations of AI, but that day is not today. We’ll discuss the inherent challenges in upgrading the skills of top leadership, and how these skills are likely to develop through the phases of AI enterprise adoption.
  3. Management with a firm grasp on AI: VPs, managers, and “Heads of” will need to be able to speak the language of AI in order to develop and assess the right strategies and technology solutions to move their company’s goals forward.

We’ll begin by discussing the state of affairs for most industries at the time of this writing: the “Emergence” phase:

Phase 1: Emergence

Engineers familiar with productizing AI:

Students who leave school with an MA or PhD in machine learning from great schools (Stanford, MIT, Carnegie Mellon) know more about cutting-edge data science than 99.9999% of engineers alive in the world today.

However, experimenting with machine learning in academia is extremely different from putting an AI solution into productive use in an organization. Indeed, applying machine learning to build a tool or functionality for one’s own team (internal users) is very different from applying machine learning to build a tool or functionality for outside users or customers. Let’s examine each situation:

Academic use of machine learning:

  • Time urgency: Finish up whenever the project or initiative is due.
  • Financial considerations: You don’t have much to spend, but you certainly don’t have to balance a profit and loss statement. Finances aren’t taken into consideration.
  • End goal: Apply novel technology or apply technology in a novel way; potentially move the field forward.
  • Consequences of failure: Mess up all you want (that’s how you learn!), just make sure it’s fixed before your advisor sees it.

Machine learning products to be used by internal teams:

  • Time urgency: Pressing, important. Money and jobs are on the line.
  • Financial considerations: Pressing, important. These resources need to recoup themselves in terms of value for the business.
  • End goal: Improve revenue, improve efficiency, reduce costs, or drive an important business initiative.
  • Consequences of failure: Your teams might be mad at you. Your boss might be mad at you. You might lose tens or hundreds of thousands of dollars. Your company might develop a general pessimism about the ROI of AI in general, making it harder to get new projects off the ground and secure budget and buy-in. You might lose your job.
  • Example applications: Smart internal document search for unlabeled text and video files. An article suggestion engine for answering simple HR questions for employees.

Machine learning products to be used by customers and external users:

  • Time urgency: Pressing, important. Money and jobs are on the line. If customers have expectations around this functionality and when it’s delivered, that’s an added pressure and an added risk of failure.
  • Financial considerations: Pressing, important. These resources need to recoup themselves in terms of value for the business.
  • End goal: Improve revenue (might involve customer experience, or user experience), improve efficiency, reduce costs, or drive an important customer-facing initiative.
  • Consequences of failure: Your customers positively will be mad at you. You are extremely likely to lose your job.
  • Example applications: Product recommendation via email to existing eCommerce customers. Customer service chatbot.

See what I mean?

I’m not saying that every university is closed off from the business realities of implementing AI or that every internal project has more leeway for error and failure than every external (customer-facing) machine learning product. I’m generalizing, but I’m generalizing for good reason. Organizational leaders need to understand.

Academic AI and internal AI applications are not necessarily the best training ground for externally-focused machine learning products. Dr. Charles Martin of Calculation Consultants has applied AI at firms like BlackRock and eBay:

When you’re a PhD… you often work by yourself. You can take as long as you want to solve problems because data doesn’t go stale, and you’re never dependent on people who are totally ignorant about AI. That’s the opposite of AI in the business world.

Martin hopes to see more PhDs working actively on projects in industry while they’re getting their degree so that more of them are prepared for the fast pace and unique challenges of applying AI in a living, breathing enterprise. 

While data science talent is hard to come by and often settles in the Bay Area (thanks to the huge salaries and exciting projects to work on with big companies like Google, Facebook, or the hottest and latest startup), there is even worse news: the subset of data science and machine learning engineers who have real experience building machine learning products to support customers is astronomically rarer and even more concentrated in the Bay Area.

Whether we’re talking to startups in Boston, Montreal or Shanghai, AI startups that raise a certain amount of money (say $50MM or $100MM) will often set up a “second headquarters” in San Francisco for better access to VC money and genuinely experienced AI talent.

For this reason, companies can expect data science talent to be hard to come by. The modern tech giants will continue to offer the most interesting work, the highest salaries, and the most talented coworkers in the world – and smaller companies (or stodgy older companies) will have to compete on grounds other than salaries.

Smaller firms and older firms have mentioned using strategies such as:

  • Providing exciting equity options
  • Having an explicit, appealing purpose that Google, Facebook, and other tech giants can’t or don’t focus on
  • Providing a less stressful work life, and more time for hobbies and family
  • Providing an opportunity to use their skills in a city that isn’t so expensive, where they might buy a home and start a family (unlike SF and NYC)

These strategies seem to be more prominent among scrappy and interesting startups than they are among larger, established companies (such as banks or insurance firms). That said, I would argue that older firms need these tactics even more than startups.

It should also be noted that these strategies are less likely to garner the cream of the crop in terms of ambitious and hyper-intelligent data science talent.

God bless the people in Austin, Arizona, or Bucharest who are attracting smart AI engineers from bigger companies in order to offer them a better “lifestyle” in a less expensive and less stressful environment than NYC, London, or SF, but the most ambitious and hyper-capable machine learning geniuses are unlikely to want to move to Bangor, Maine or Wakefield, Rhode Island to make the most of themselves and leave a mark on this world. (I’m not actually religious, it’s a figure of speech, and here is being used as a somewhat sarcastic one. No offense).

Until you reach a critical mass of traction, growth, and (importantly) prominence, you will have an exceedingly hard time recruiting (never mind retaining) the top AI talent.

What if I Told You Deal With It

Am I saying that you’ll never attract talented people without being Microsoft or Google or Baidu? No, I’m not saying that. There will be occasional exceptions. Rarely will there be any exceptions for where truly top talent will go unless your company can provide one of the following:

  1. A clear path greater wealth than Facebook or Google could give them
  2. A clear path to greater prominence and importance in the world than Facebook or Google could give them

Some people assume that a nice work culture and an important vision are enough, but I think that these are generalized values intended for team members who work less than 45 or 50 hours per week. I’ve spoken with hundreds (if not thousands) of AI experts, many of them borderline geniuses who came out of exceedingly elite schools. If you think that most elite ML engineers are interested primarily in work-life balance, you are in for a harsh reality.

“Put a dent in the universe,” is what Steve Jobs famously said he was trying to do, and this core desire transfers well to a huge bulk of the smartest and most ambitious (i.e. often, the most capable) AI talent: the 1% of the field.

The engineers who get jobs in order to spend more time outdoors or in order to spend more time with their children are unlikely to be the 1% of talent whose crushing ambitions and massive intellect are most likely to grow something huge.

Those facts are cold. I am merely the messenger, and I ask that you not resent me for being straightforward with you here.

Training talent isn’t easy, either. At present, there is no clear path to developing AI talent through any of the following means:

  • Getting non-technical staff to learn to actively use data science tools in a business context
  • Getting existing non-ML engineers to become machine learning engineers
  • Getting IT personnel (particularly at enterprises) to understand the importance of data infrastructure
  • Getting non-technical people to grasp the basic capabilities and functions of AI so that they can work well with technical teams (that’s what we do with our free articles at Emerj, and in our AI Strategy work with enterprises, but the learning needs vary widely from company to company)

The skills are so new, and the applications (for the most part) are so nascent, that there’s hardly any people with experience bringing AI to life in business. In the coming five years, we can expect to see a lot more direct “best practices” for AI education enter the field, but we’ll talk about that in the “Adoption” phase (further down in this article).

It should be noted that sectors which are extremely late to reach the “Adoption” phase may enter the “Emergence” phase with a much more robust ecosystem of AI talent than companies today are able to reach. This might be the case in relatively stodgy and technologically underserved sectors (i.e. the lumber industry, carpet manufacturing, etc), or in geographic regions that are late to the game in terms of digital technologies, software, and the internet in general (much of the industry in India and Africa, for example).

These “late” sectors or regions will enter “Emergence” when talent is more plentiful on a global level and when “up-skilling” is astronomically more accessible than it is today. However, these sectors will still struggle to compete for talent or for retaining talent, as it is likely that the best data science skills will be somewhat absorbed into sectors that are digitally savvy and full of exciting startups. This will at least be the case during much of the “Emergence” and “Adoption” phase of the AI Zeitgeist.

Organizational leaders with a firm grasp on AI:

Leaders in some of the largest industries are beginning to tune into artificial intelligence, but most are at the early stages of understanding. They’re aware of it’s internet-like impact on industries, but they aren’t sure of what that means to them.

Most C-level leaders in enterprises seem smart enough to realize that (a) they don’t yet grasp artificial intelligence conceptually, (b) they have no firm idea of what applications have real ROI potential (versus those which are purely speculative), and (c) they’re organization may be ill-prepared to adopt artificial intelligence in the first place.

What does that mean?

It means that sales are that excitement and skepticism and vendors have to dance a delicate dance with execs. On the one hand, vendors need to mention “artificial intelligence” in order to seem “modern” and hip, but they also need to downplay the data science talent and the integration and iteration time required to see value from an AI solution. Some execs have been burned in this manner, and many higher-up leaders are wary, though many take the bait.

As a sidebar, I’d like to thank the “guinea pig” execs who purchased AI solutions without being truly ready to derive value from them. They might not have seen an ROI, but they gave vendors a chance to find what’s right and wrong with their product. If it brings you any solace, first-mover-but-not-yet-ready execs, you have helped the ecosystem as a whole. Pour some out for you guys:

It will be another two or three years until it’s commonplace for artificial intelligence to play a prominent role in the long-term strategy and top quarterly initiatives of most enterprises in the US, never mind in other countries.

Many companies are moving slowly towards AI adoption. It is those same executives who would release the funds needed to hire the data science experts in the first place, remember?

That being said, sectors like retail, finance, and pharma are bubbling with vendor solutions and venture money, and executives in some of those profitable and large sectors are getting up to speed on AI as best they can through events, newsletters, and articles.

Management with a firm grasp on AI:

From our experience here at Emerj, it feels clear that managers are often more interested in use-cases related to their specific function, as opposed to the business at large. In addition, functional experts today (when most industries are clearly in the “Emergence” phase) are often curious about artificial intelligence as a career-related skill rather than seeing artificial intelligence as a tool that their companies could realistically use.

In some regards, this is understandable because AI adoption is challenging, and functional managers don’t wield as much power into procurement or strategic decisions as do executives.

This career-oriented curiosity is what I suspect to be the main motive for managers to attend AI-related events. Many of them are reasonably convinced that AI will have a massive impact on their sector or functional role, and they’d like to have enough of a grounded understanding to ride that wave properly in the next decade of their career.

Phase 2: Adoption

Engineers familiar with productizing AI:

In the “Adoption” phase, experienced engineers will still be rare. Instead of being unicorns, however, they’ll be findable by many companies who (a) have the budget to afford then and (b) have a work environment where they can function, such as the right data science teammates and “connective tissue” of data-savvy subject-matter experts on their team.

In the coming decade, more of these experienced AI engineers won’t be as rare because:

  • AI-dense companies like Google and Facebook will have best-practices for AI education and productization that are widely used by other companies outside of Silicon Valley
  • Vastly more universities will offer reasonably good (or occasionally exceptional) grounding in data science for many of their degrees
  • Online learning will become vastly more popular, and data science skills in continued high demand online
  • Large AI-dense firms will churn talented employees who will continue to start their own companies or move into other sectors, disseminating much of the once-rare knowledge about AI use-cases to other sectors

Almost by definition, the “Adoption” phase won’t happen until a critical mass of these experienced technical experts (who know how to apply AI in a real, living, business environment) are disseminated into industry, making AI accessible to more firms.

In addition, AI tools will become easier to use, with more intuitive drag-and-drop interfaces and simpler methods for dealing with data and “tweaking” algorithms without the need for strong math skills from every team member. In article 4 of 7 of the AI Zeitgeist (coming up after this article), we’ll talk about how AI software tools will actually become easier to use.

Organizational leaders and managers with a firm grasp on AI:

I’ve joked frequently that in the “Emergence” phase, most of the money is made at artificial intelligence conferences and events, not with actual vendor companies.

In part, this is a necessary transition because, without the contextual insights into the fundamentals of AI, business leaders won’t know how to build AI strategies, assess AI vendors, or even speak with technical AI experts.

Right now, industries like banking, insurance, retail, and pharma are so saturated with AI events that it’s nearly impossible for businesspeople to avoid learning about the technologies and what they can do. Not only are there a lot of executives, managers, and VPs in these industries reading about AI online (as evidenced by our growing audiences from major sectors), they’re also investing in expensive events and training.

An industry in the “Adoption” phase will see a large percentage of VPs, managers, and executives with a solid understanding of AI use-cases, and the basic capabilities of the technology. Allowing them to:

  • More easily avoid AI hype and separate reasonable and unreasonable claims
  • Find opportunities for AI to solve their existing business problems
  • Understand the needs for data and data infrastructure
  • Be able to “talk shop” with data scientists (on a conceptual level)

All of this will extend further into the “Dispersion” phase:

Phase 3: Dispersion

Engineers familiar with productizing AI:

By the “Dispersion” phase, machine learning engineers and data scientists won’t be nearly as revered as they are today, for a number of reasons:

  • Most software that involves machine learning will not require technical coding or maths skills to operate
  • Gradually, more and more of the engineering population will have data science skills and knowledge from both school and work experience

Of course, there will remain a need for cutting-edge technical machine learning skills, but not for most of the day-to-day use of many enterprise technology solutions. The extreme end of machine learning and AI talent (the “wizard skills”) will still be needed for moving the field forward and for developing next-generation solutions, but these same skills won’t be necessary for even a mid-sized business to use AI daily in customer service, marketing, and other functions.

Who knows how long it will take a sector to reach the “Dispersion” phase. Even for extremely AI-advanced sectors (like the US eCommerce or Digital Media sectors), it may be well over a decade until we could safely say that “Dispersion” is reached. Once it is, however, data scientists won’t be the same kind of relatively rare specimens they are in the enterprise today.

Organizational leaders and managers with a firm grasp on AI:

By the “Dispersion” phase, managers and leaders will have a broad, taken-for-granted understanding of most of the critical concepts and capabilities of AI that they need in order to operate in their sector.

As I mentioned in our first AI Zeitgeist article (where I explained the three phases in depth), knowledge about how AI works will eventually be like knowledge of how the internet works today. Businesspeople don’t know how the internet works, technically, but they understand what it does, they’re familiar with basic terms (servers, wifi, etc), and that working knowledge serves them reasonably well in doing their work.

While AI will always have an expanding set of capabilities that will need to be followed and kept up with, a basic familiarity of the value of data, the requirements of data infrastructure, and the basics of “what AI can do” will eventually be taken for granted in most modern enterprises, and that’ll likely happen when we reach the “Dispersion” phase.

Concluding Thoughts

This has been week 3 of 7 from the “AI Zeitgeist” article series.

In the coming weeks, we’ll be exploring the following Zeitgeist topics, in order:

  1. The 3 Phases of AI in the Enterprise: Emergence, Adoption, and Dispersion (AI Zeitgeist 1)
  2. How “AI” Will be Discussed in the Future (AI Zeitgeist 2)
  3. The Evolution of AI Talent and Training (AI Zeitgeist 3) <— You are here
  4. The Increased Accessibility of AI in Business (AI Zeitgeist 4) 
  5. Buying and Adoption Readiness for AI (AI Zeitgeist 5)
  6. The Changing Landscape of AI Priorities of Business Leaders (AI Zeitgeist 6)
  7. The Competitive Dynamics of AI – Now and in the Future (AI Zeitgeist 7)

Next week we’ll explore how AI will become more accessible in the years ahead, including software developments and automated machine learning trends that will lessen the burden of technical skill requirements. Stay tuned next week.

Stay Ahead of the AI Curve

Discover the critical AI trends and applications that separate winners from losers in the future of business.

Sign up for the ‘AI Advantage’ newsletter:

Subscribe
subscribe-image
Stay Ahead of the Machine Learning Curve

At Emerj, we have the largest audience of AI-focused business readers online - join other industry leaders and receive our latest AI research, trends analysis, and interviews sent to your inbox weekly.

Thanks for subscribing to the Emerj "AI Advantage" newsletter, check your email inbox for confirmation.