We’re all getting used to the thought that in a not-so-distant future, competition for jobs won’t just be other humans, it will also be an intelligent robot, self-driving car, or other artificial agent. But in our gut, we know this can’t be the full truth, that there’s a more nuanced story. We at least believe that elite human skills will remain valuable even as automation eats the world. The hard part is figuring out which ones will be the most valuable and where they will be the most prized.

• Every successful relationship is successful for the same exact reasons

As a parent, this can be a particularly vexing problem when thinking about how to advise your kids. Common wisdom–learn to code, cultivate empathy, study STEM–isn’t especially useful because it isn’t specific enough about what it takes to stay ahead of the robots for years to come. Many of the major advances in AI are happening in just these fields: Machine learning will ultimately eliminate a lot of coding work, perceptive and emotional AI is developing fast, machines are already good at math. So, instead of analyzing what jobs will be most threatened by AI, we turned our model upside down to look at what careers require human capabilities that the robots won’t be able to beat for a very long time.

We analyzed the 30 jobs that show the least potential for automation over time and found these jobs grouped into four categories:

We found one common factor in these clusters: unpredictability. Where the job requires people to deal with lots of unpredictable things and messiness–unpredictable people, unknown environments, highly complex and evolving situations, ambiguous data–people will stay ahead of robots. For instance, our people cluster doesn’t include just any job that handles people, it includes jobs that deal with people in unpredictable environments like school psychologists and supervisors of firefighters and repairers. And our bugs and bad things cluster doesn’t include just any health-care job, it includes jobs that handle complex relationships between ecological systems and human health like allergists, epidemiologists, and microbiologists.

We also found that these low-automation jobs relied on skilled management of real-world systems. For instance our spaces and structures cluster highlights that no matter how important the digital world becomes, we will always need people to design and manage the physical world like aerospace, civil, environmental, and marine engineers. And although software can automate plenty of math, humans are required to apply that math to real world systems in roles like economists, management analysts, and treasurers.

Our current digital environment can lull us into thinking that future “safe” career opportunities will be about using computers–being adept at manipulating code, mastering software, or creating a virtual world. But our research points to an important and valuable insight: the highest human value is not about being plugged in but being unplugged. And in order to excel at these tasks, we need to remove the digital filter and experience the people and physical world around us.

As AI pervades more of our physical world experience, AI determines how we interact and learn, offering us less experience in the physical world. That thereby reduces our skills in dealing with, say, quirky individuals or novel engineering challenges or rapidly evolving biological systems for which there are no data for an AI to use. And virtual experiences have their limit. At some point, things need to happen in the physical world, with in-person interaction. These are the skills that an AI won’t be able to beat us at.

Sign up for the Quartz Daily Brief, our free daily newsletter with the world’s most important and interesting news.

More stories from Quartz:

• A Nobel prize-winning physicist identified three simple steps to mastering any subject