Image source: McKinsey “The economic potential of generative AI: The next productivity frontier” , June 14, 2023

Training has been less susceptible to automation than many other job categories because it requires skills that, up to now, only humans possessed—creativity, audience analysis, dealing with ambiguity, nuanced communication, understanding human psychology, etc. McKinsey estimates that prior to the advent of generative AI, only 15% of the work time for educator and workforce training roles consisted of tasks that could be automated. This is the lowest of any occupational group.

However, with the advent of generative AI, McKinsey estimates that the potential for automation for training roles jumps to 54% of work time. This 39% jump in potential automation is the largest of any occupation group, although it’s worth noting that the 54% total time savings with AI is still slightly below average (42nd percentile) when compared to all job categories.

The possibility that roughly half of the work time for those in training roles could be automated in the future may feel both exciting and disconcerting. A worst-case reading is that training industry jobs could be cut in half. Before you panic over the worst-case scenario, consider these additional factors:

1. Potential automation is not the same as achieved automation. No company succeeds in automating every task that it is possible to automate. And introducing automation at scale can take years. Competing priorities may keep training lower down the priority list for large-scale AI initiatives.
2. Efficiency is not the only use case for AI in training organizations. It may not even be the most impactful. AI has the potential to improve training quality, consistency, learner engagement, and personalization. These improvements increase the business impact of training, which could result in additional investment. This type of innovation does not replace employees, it empowers employees.
3. Many of the tasks that AI can do well are rote—the kinds of tasks training developers would gladly give away. Pushing these tasks to AI leaves more time for creative tasks, making training work more interesting and rewarding. In fact, working with AI is itself an interesting and creative task. Surveys show that the majority of knowledge workers who have tried ChatGPT in the workplace come away with positive feelings toward the experience.
4. The coming changes are not a zero-sum game between employees and computers. As AI changes the nature of work, new roles and tasks will emerge. The World Economic Forum’s *Future of Jobs* report for 2023 shows that **25%** of companies surveyed predict that AI will lead to job losses, However, **50%** expect AI to lead to a net increase in jobs. Although automation may reduce the need for some training roles, new roles are likely to emerge.

The changing nature of training work

AI isn’t likely to make training professionals obsolete. However, it’s almost certain to change the nature of training work. Fortunately, the skills that make you good at training—creativity, problem-solving, precise use of language, analytical expertise, logic, research, etc.—are likely to make you good at using AI once you learn how to apply them. However, there is an important asterisk. Your previous skills alone won’t be enough to keep you competitive in your career moving forward.

As AI becomes increasingly common in training departments, those who have made the effort to develop related skill sets will have a leg up. If you want to be prepared for the evolution of the training industry, here are a few things you can begin to work on right away:

• Learn about AI – Study how AI algorithms work and explore use cases for AI in the training sphere. Forward-learning training organizations will always be looking to push the envelope. The more you know about AI, the better you will be at finding clever ways to use it.
• Get hands-on experience – It’s easy to get mediocre results from AI with simple prompts. However, to get the best out of AI you need to learn prompting and workflow best practices. Tutorials and sample prompts can help. But blindly following recipes won’t lead to proficiency because AI responds very differently across various user contexts and tasks. Only hands-on experience can help you figure out how to use AI effectively for your setting and use cases.
• Understand AI risks – Learn about responsible AI practices and how to address challenges around social bias, intellectual property issues, data leaks, and fact-checking AI outputs. This knowledge will be increasingly valuable as AI takes root and may become a job role of its own.
• Focus on human skills – Double down on soft skills like empathy, collaboration, creativity, and human-centered design. Current AI is most capable when dealing with routine or intellectually-oriented tasks. Employees who have both soft skills and AI proficiency will have a leg up.
• Improve data literacy – Whatever your training context, eventually you’ll bump against the need to supplement the built-in knowledge of AI models with your own data. Learn how to gather, clean, integrate, and analyze relevant data available in your organization.
• Embrace continuous Learning – AI isn’t something that you learn once and then move on. A commitment to continuous learning is critical if you want your AI skills to stay relevant. The techniques you use today will not be the techniques you need tomorrow because AI is evolving at a breathtaking pace.

When and how to make the jump to AI

If there are only two ways to react to exponential change—too early or too late—how do you decide when and how deeply to wade into the AI waters?

For your personal career

For your career, the answer is straightforward. If you haven’t already, it’s time to at least dip a toe in the water. There isn’t a “too early” because there’s no downside to learning. Start experimenting with what AI can do both in your personal and work life (assuming your organization permits it).

You’ve probably already had some exposure to AI via common training tools that have embedded AI features. These task-specific tools are an important part of the AI equation. But one of AI’s main superpowers is its flexibility. Task-specific tools don’t expose the full range of capabilities that make AI a disruptive technology. For that, you should focus on chatbots such as ChatGPT, Claude, or Google Gemini.

While there is no prerequisite knowledge needed to use generative AI, taking time to learn the basics of the underlying technology is a good investment. It will help you identify good AI use cases, employ effective prompting strategies, and discover innovative tools. So dive in. Follow AI experts on social media. Subscribe to AI newsletters. Read about AI advances and product launches. Try new tools. As your knowledge grows, so will your confidence and skill.

As helpful as learning from experts can be, don’t make the mistake of over-relying on external research. Hands-on learning is critical for any tool. But it’s even more important for AI because AI is not deterministic. Repeating the same prompt doesn’t yield the same response. And a prompt that works well in one context or for one subject matter may not work well for another. You’ll need to develop an intuitive feel for how to best use AI for your circumstances based on patterns you discover with actual use.

When doing hands-on exploration, don’t forget to experiment with workflows. For instance:

• Do you get better results from asking AI to write your first draft or from writing a rough draft and then asking AI to improve it?
• Is it better to engage in a lengthy conversation with an AI chatbot or write a single very detailed prompt that generates output all at once?
• Do you need different models for different tasks? Are there scenarios where it works best to use multiple models for the same task and then integrate the results?

AI’s capability has been described as a “jagged frontier.” How capable AI will be for a specific task is surprisingly hard to guess, even for experienced users. So my final advice is to try to get AI to do things you don’t think it can do. You might be surprised. If it doesn’t work the first time, experiment with different models and different prompting approaches. Whether or not your experiment is ultimately a success, you’ll learn critical lessons about the boundaries of what is and isn’t feasible.

For your training team

If you’re responsible for a training team, the answer is more complicated. There are risks when using any early technology. The risk/reward equation for using AI will depend on the nature of the training you provide, your available resources, your organizational culture, and your personal risk profile.

There are primarily two paths to using AI at the team level. Each path has advantages and disadvantages.

Individual use

This strategy involves facilitating the ability of individual team members to experiment with AI informally on their own. This represents the lion’s share of business use of AI at this stage of its development. Individual use doesn’t mean you can’t direct team members to work on particular challenges. It just means that you’re not ready to apply the full weight of organizational resources to your experiments, which generally rules out more advanced use cases.

As I noted previously, chatbots are the most flexible and powerful tool for individual AI experimentation. When researchers conduct studies on the effectiveness of AI for business, they often simply give workers access to chatbots with no training or supervision. The results have tended to be notably positive. However, you can substantially increase your odds by providing training and support.

Nearly all popular chatbots have both free and paid versions. Paid versions tend to have the most advanced models, more features, and higher usage limits. The problem with free accounts is that you can’t be sure whether failures are genuine AI limitations or simply the result of not using state-of-the-art tools. Given that the cost of paid AI accounts is low, I recommend providing paid accounts if you can swing it.

Be aware that organizations have different takes on individual employees using AI. Some organizations encourage it, either by explicitly promoting it or by knowingly allowing it to flourish. Others turn a blind eye with more of a “don’t ask, don’t tell” policy. This gray area can give you a window to demonstrate success, which will improve your chances of getting a formal project approved down the road. Still other organizations actively attempt to block individual AI use out of fear of its risks or because of regulatory issues. Make sure you stay in line with your organization’s approach.

With an individual user strategy, you are likely to have fewer short-term organizational barriers and enable rapid experimentation, which can be quite fruitful. But you still need to recognize and mitigate AI risks by providing ground rules. To get the most out of the ingenuity of your team members, give them an opportunity to collaborate and share their learnings.

Individual use of AI can be very productive. But it doesn’t scale. Over time, you’ll end up with uneven use of AI across your team, a wide range of quality, and possibly a hodgepodge of tools. Unwinding idiosyncratic use of AI when it’s time to move to a more systematic approach can require a significant effort. You may end up with unhappy team members when you eventually move away from the tools and workflows that they spent a lot of time perfecting. In addition, if your organization is hesitant about AI use, it may be harder (and riskier) to publicize your successes with an individual use approach.

Formal projects

Formal AI projects are bigger bets that follow traditional business processes. Since you’re working with a novel and relatively unproven technology, you still want to think of these projects as experiments and set stakeholder expectations accordingly. However, this strategy implies that you’ve done enough homework to anticipate an eventual “win” that will enhance your core training processes. Without this, it’s difficult to justify the larger investment of time and dollars that formal projects require.

Formal projects can start with discovering a particular tool that suits your needs or with thinking about your needs and then exploring whether AI can address them. If you take the approach of actively exploring use cases, you’ll need to identify use cases that have a high enough value proposition to justify the budget, effort, and risk of a formal project. What are your biggest training blockers, time stealers, and challenges? These are your AI project candidates.

Some use cases will stand out as impractical right away because of size, complexity, cost, risk, or other factors. The rest should be ranked based on their value vs the effort required. Since effort can be hard to predict when using a technology that has a jagged frontier of capabilities, you will typically want to do hands-on testing of your top use case to get a sense of how feasible each one is as you decide which project to choose.

One important factor in choosing an AI use case is what custom organizational data you will need to meet your objective. For example, if you want to create a chatbot to make it easier for product users to find information, you’ll need to gather source data from some combination of product documentation, knowledge bases, internal technical bulletins, and existing training material. Make sure you investigate what data is available, where it lives, what format it is in, how good it is, and how easy it is to get to. Obtaining, cleaning, and keeping data up to date is one of the most challenging and time-intensive tasks for an AI project.

You will need to decide early on the extent to which you want to go it alone versus getting outside assistance. For low-complexity projects, generative AI isn’t prohibitively difficult to learn. However, it’s easy to underestimate the amount of time and effort it takes to get up the learning curve for even seemingly simple use cases. And there is hidden work that is easy to overlook. For example, if you’re project involves using custom prompts, you will need tools and workflows for creating, testing, storing, versioning, and maintaining prompts. Rather than following a trial-and-error path, it often pays to bring in a consultant or experienced AI coach who can help you get results quickly, avoid common mistakes, and navigate organization-specific risks.

AI projects follow a similar path to other technology projects. You’ll need to define success metrics and measurement strategies, evaluate tools, create a project timeline, develop a change management plan, etc. One area where AI can be trickier is dealing with its known risks around security, intellectual property, inadvertent disclosure of data, systemic bias, etc., Expect to invest considerable time and energy shepherding your project through the approval process. It can help to have proposed risk mitigation strategies in hand. Also, an attitude that treats safety teams as partners in making sure you don’t blow something up rather than as blockers can go a long way toward greasing the wheels to the necessary approvals.

If all this sounds like a lot of work, that’s because it frequently is. Launching an AI initiative is an investment in the future. Because of the learning curve, your first attempts at AI may fail to give the results you hoped for or temporarily reduce your efficiency as you divert resources away from current activities. However, as you iterate, you’ll clear the way for potentially dramatic improvements in business outcomes in the future.

Why take the risk?

If you’ve been in the training industry for a while, you’ve undoubtedly seen instances of highly promoted new technologies that never took root. Perhaps, like me, you’ve developed a strong internal BS detector that goes off when you encounter a technology that is looking for problems to solve. AI is odd in this respect. It is unabashedly a technology looking for problems to solve. Not only that, we are clearly still in the over-hyped phase where AI does not live up to the promises of its more glowing promoters. So why would you risk using AI now?

While there are obvious risks, there are also factors that make being an AI early adopter a better bet compared to less disruptive technologies.

1. AI is unlikely to become an abandoned technology because of the billions of dollars being invested in it. Not every application of AI will ultimately be a winner. But the companies making these investments and building AI infrastructure are not going to walk away before they find a way to make it successful.
2. AI is arguably the fastest-evolving technology in history. What it can’t do well today, it may be able to do well in a year, or perhaps even in a few months. We don’t know which current AI limitations will be stubborn. But we do know that its trajectory will be upward, meaning that the value of your early investment will increase over time.
3. AI is such a flexible technology that in some ways it’s actually under-hyped. We’ve only explored a small fraction of what it’s currently capable of, much less what it will be capable of soon. One of AI’s main limitations is our failure of imagination in finding ways to use it. You control that part of the equation.

Despite its risks, there is a strong upside to being an early adopter of a truly disruptive technology. Research shows that embracing new technology is an attribute that frequently distinguishes top organizations from laggards. Even when early attempts at using AI aren’t entirely successful, early adopters learn lessons that get them to success sooner. And because of the head start, early-adopting organizations tend to maintain their advantage over slower organizations even after the technology reaches mainstream adoption.

In addition to propelling your organization forward, launching a successful AI initiative can boost your profile in the organization and help you create a reputation as an innovator. It can also grow the visibility and reputation of your team, creating new allies and opening the door to cross-team partnerships.

The secret to working with a fast-changing, revolutionary technology is to plan thoroughly, start small, and keep expectations reasonable. Don’t expect to hit a home run in your first at-bat (although it does happen). Instead, remind yourself that you’ll never hit a home run if you don’t step up to the plate.