In AI Learning, One Size Fits None

A: One of the striking trends I’m seeing in my classes is that students from a variety of different backgrounds are coming to study AI. I teach in the computer science department, but most of my Deep Learning students are not computer science majors. This is a new development. We have people from mechanical engineering, materials science, aeronautics, and medicine who want to learn how to apply AI in other areas in incredible ways. A few of my students with a background in energy took the course to build a predictive maintenance solution for a drill system.

I call these learners “AI+X” and they’re quickly becoming the top users of AI, whether at universities, enterprises, or in government. These AI+X individuals are studying the technology in connection with their domain expertise, and they’re coming with an intent to solve real-world challenges.

A: That’s right—this evolving cohort of AI learners is disrupting the way many of us used to think about education, whether at a university or on the job. A one-size-fits-all approach where people sit in a classroom for a set number of hours or complete a standard curriculum doesn’t serve AI learning today. The technology is changing so rapidly and there’s understandable hype about the shrinking “half-life” of digital skills. We’re not even at rock bottom of that half-life, so it’s becoming less practical or reasonable to spend time on irrelevant material.

For instance, a mechanical engineer knows linear algebra but may need to learn Python coding, and an electrical engineer understands the Fourier transform but may need to learn new neural network-based algorithms. The same is true for mission and business leaders who don’t have the luxury to take a 200-hour class but need to continually grow their AI skillset.

The key challenge today is about figuring out how we make smart decisions about what to learn and where to spend our time. That’s why we’ve seen a lot of recent innovation to bring AI learning together with AI agents that can verify a person’s skills, optimize for their gaps, and serve up fresh content that is targeted to individualized learning goals.

And the impact is tangible. For example, at Workera we’ve seen a fivefold increase in learning velocity thanks to AI agents used in skills verification. By applying AI agents themselves to the upskilling process, we’re able to optimize the process for the learner, serve the right content to the learner at the right time, and avoid material that the user already knows or that may not offer the fastest path to progress.

A: We’re likely past the generation where someone can expect to do the same job for 30 years and then retire. But I want to stress that subject matter experts aren’t going anywhere. In fact, there will be more demand for domain expertise as AI fuels new applications.

In terms of my career, I find it helpful to think about my development in the shape of a letter T.  The horizontal line at the top of the T represents the breadth of my durable skillset: These are skills that have long-term relevance—for example, the math theory and statistics I learned growing up in France. To this day, those durable skills still allow me to jump quickly into new subjects with confidence. Then, the vertical line of the T represents the many perishable skills that I acquired at points in time for the purpose of innovation.

While each of my individual skills might have different staying power, it’s the combination of perishable and durable skills that allows me to keep up with progress. So, for anyone worried about the longevity of your skills: Your future self has the advantage of a longer lasting and more durable skillset, even as the practice of technology evolves.

A: Organizations and their leadership teams are hyper-focused on measurement and understanding the progress of AI transformation. But with so many possible metrics, leaders have questions about how they should approach enterprise measurement. I recommend they track progress across two parallel categories: outcomes and skills.

By outcomes, I’m referring to the big changes an organization is seeking by investing in AI transformation. For example, these master outcomes might include increased productivity, innovation, and responsible use. When an organization is clear on these outcomes, they can start to track major indicators:

• How much new capacity have we unlocked through AI? (related outcome: productivity)
• How many projects have an AI-ready system in production? (related outcome: innovation)
• How many alerts or incidents have come up? (related outcome: responsible use)

In parallel to outcomes, enterprises should measure the progress of skills acquired across all participating employees. Are skills adequately developed to achieve the master outcomes over time? Does the organization have the right ontologies of skills that can be verified? Are employees able to get certified in critical areas (such as responsible AI), and are they getting feedback about their development?

Organizations can verify the skill levels of their employees in individual domains, in addition to tracking the most important metric for a skills-based organization: its overall learning velocity. Increasing overall learning velocity—how quickly employees progress when learning a new skill—can allow companies to quickly catch up to, and surpass, competitors. This kind of granular assessment of skills helps leaders see how mature their organization is today—and how they’re pacing toward enterprise proficiencies.

A: My first piece of advice is for the people at the top. In my work, I’ve seen leaders across the spectrum of engagement. There are executives who make themselves part of the training—measuring their own skills, closing their knowledge gaps, getting badged and certified—and those who don’t but pretend to know more than they do or don’t hold themselves up to the same expectations as the workforce. When leaders create high standards for their own growth, employees see that “we’re all on this journey together.” This leadership engagement goes a long way to fostering a positive culture around AI transformation.  

It’s also important to consider the culture that surrounds experimentation. Some organizations think about AI as a high-stakes effort and constantly push for priority-level projects. But that approach will eventually hit a wall, given the complexity and skills needed to push AI systems from prototype into production. My advice is to initially focus innovation on small but visible projects. Just think about the now-famous Google Brain project to detect cats in videos, which Andrew Ng led at the time. This AI project had zero risk to the company mission if the system didn’t work. But people love cats, and the project’s neural network approach became very visible. It ended up accelerating the production of neural networks within other Google programs with higher stakes. I often stress to organizations that those small efforts, ones that may seem insignificant or even frivolous, can inspire meaningful innovation down the road.

There are a growing number of organizations and foundations that are leading the way for AI literacy across communities, especially in the K–12 system. From on-the-ground training programs to National AI Literacy Day, they’re preparing young students with AI skills for university and the workforce—and offering programs to equip teachers and administrators with the tools they need.

But broad access to AI is a long-term challenge that won’t be tackled by any one player in the ecosystem and will need everyone at the table. We still have a journey ahead to make sure communities and populations are not being left behind. Just think about the aging workforce or the 20% of U.S. households that don’t have internet access—the digital divide will only get larger as AI starts to impact all aspects of society.

We talked earlier about the T-shaped model for thinking about durable and perishable skills. It’s important to emphasize that AI will eventually become a horizontal, durable skill—one of the historical power skills that will have an extremely long half-life. AI will soon show up in job descriptions you might have never expected, so we all have a role to play in eliminating barriers. With the Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence, the federal government has put a stake in the ground for what we need to get right—and how players need to come together for equitable access.

I’m excited to see the progress we will continue to make as an industry. Whether that’s collaborating with nongovernmental organizations, community partners, or the federal government, or offering no-cost training programs (which I’m proud that we offer at Workera), small efforts go a long way to put AI in the hands of more people. Ultimately, access is about more than strengthening tomorrow’s workforce—AI literacy will soon be a basic human need to navigate the world, from finances to healthcare to government benefits.