I have a strange fascination with online education startups and their marketing copy. Here’s Udacity’s homepage from 2012:

Udacity is a totally new kind of learning experience. You learn by solving challenging problems and pursuing udacious projects with world-renowned university instructors (not by watching long, boring lectures). At Udacity, we put you, the student, at the center of the universe.

And their homepage a decade later:

Trust an IDC leader to overcome digital talent gaps.

Compared with 21 vendors, IDC recognized Udacity for creating job-ready digital talent and delivering measurable business outcomes.

Remarkable, isn’t it? You can almost see the sparkle fading from their eyes. Gone is the fun, the enthusiasm, the made-up adjective. Udacity and many of its peers have followed a similar trajectory. They start out pitching themselves as an education revolution. But over time, the companies evolve—usually into one of a few niches:

• Catering to consumers. “These fun, bite-sized lessons will fit into your busy schedule. Try it risk-free for 30 days.” Archetype: Masterclass
• Catering to employers. “Make your team more efficient. Customer surveys and reports prove that it works. Contact sales for more information.” Archetype: Pluralsight
• Catering to other educational institutions. “Teachers, assign our interactive textbooks and activities. Students, use our tools to boost your grades.” Archetype: Chegg

These companies provide valuable services. But revolutionizing education is notably absent from their list of accomplishments. In fact, some force seems to consistently push companies away from that. Why?

Maybe online education just isn’t that good. Something about being in-person—the human connection, the social pressure—is necessary for real learning. So software-oriented startups have to find a supplementary role that’s dictated by the other components of the educational experience.

Teaching through a flat glass rectangle is certainly a constraint. But I’m skeptical of how much it can explain. Depending on other factors, learning online can still be engaging, and learning offline can still be dull. What’s more, much of the former is supported by small teams and nonprofits—not a vibrant startup scene. App designers have gotten quite good at manipulating human motivations, and I expect online learning would only become more compelling with the level of effort that goes into optimizing, say, Instagram. When the goal is clear, motivational scaffolding can be constructed to make it reachable.

When the goal is clear. That qualifier, I think, is at the heart of the issue. Online education startups work when the goal is clear. An ideal education isn’t like that; it’s more about finding the goal. It’s about discovering skills and perspectives that will be valuable in unforeseeable ways 10 years from now. And here’s where the startups really fall down. Suppose a company could somehow provide that sort of ideal education for half the price of a college degree. Would it succeed?

In economics, there’s a classic story about the lemons problem. Buyers want used cars that are in good condition, and sellers are willing to supply them. But buyers are suspicious that the cars may have mechanical problems, and so they aren’t willing to pay what the cars are worth. If sellers don’t have a way to convince buyers that their cars are free of defects, none of those transactions can take place. Likewise, if education companies can’t convince students that their services will provide long-term value, they can’t sell the services for what they’re worth.

It’s very hard to tell if a specific educational path will have long-term value. For starters, the education industry has its own kind of lemons: low-quality coding schools are a salient example. But the failure modes can be more subtle than that. Students might walk away with a superficial understanding of the material. Or the skills they learn might not be the ones they need to get a job. Or they could find their calling in an unrelated field. Besides, how can you possibly know what value a company will provide over 10 years if it’s only existed for 2? You can’t. People are rightly cautious about paying money for the promise of a distant future benefit. And education, whether online or off, is largely about distant future benefits.

Because of this problem, society is pretty conservative when it comes to education. This is true for students, who often prefer to attend established and prestigious institutions, and for employers, who look for well-known credentials as a sign of talent. A centuries-old college may not be the ideal educational institution, but we know it’s good enough. That’s more than can be said about most of the alternatives. Conservatism persists here because it’s genuinely hard to improve upon.

Given all that, it’s clear that Udacity’s homepage from 2012 was not the start of a viable business. Creating a valuable online learning experience from scratch is hard enough. Convincing lots of people to invest their time and money into it is harder still: in the short term, it’s practically impossible. So if a company wants fast growth, it has to plug into some existing signal of value like fun or profits or grades; it has to find a niche like the ones I described above. When companies realize this, their bold mission statements disappear.

The next generation of education startups will be somewhat different. They’ll have better technologies to work with and new ideas to try. They might make a little more headway towards the important parts of education. I wish them the best of luck. But what’s holding back online education is not something these companies can improve directly. It’s our shared sense of what education should be, what success looks like, what’s “normal” and what’s “risky”. And these things necessarily change on the timescale of cultures, not startups.