Pizza making robots

Everything AI touches is uncanny valley. Tutors, too.

What’s Happening in the Market

• It’s not a total desert out there in EdTech funding! Modal.io and Campus.edu have raised money this year, and both are focused on upskilling. Zen Educate has raised more than $37M. They match the supply of teachers to schools, which is an interesting premise; but it doesn’t help solve the top-of-funnel pipeline issue, which is that no one wants to be a teacher in an advanced economy anymore.

• A few weeks ago, Praktika raised $35M to provide 1-on-1 tutoring to students through AI avatars. Because definitely the biggest barrier to people learning languages is not having an uncanny valley, dead-eyed Sim to practice with. I suppose this is better than just having a normal chat encounter, but we shouldn’t rush to replace human interactions and learning with chatbots.

  

One Big Idea

In 2018, IBM Watson partnered with Pearson to launch an AI tutor attached to their psychology courses. It was a test that, by all accounts, failed. 

“We’ll have flying cars before we will have AI tutors. It is a deeply human process that AI is hopelessly incapable of meeting in a meaningful way. It’s like being a therapist or like being a nurse.” - Satya Nitta, Global Head of AI Solutions for Learning, IBM Research

There is a deep disconnect between more cautious voices (like Satya Nitta and Dan Meyer), and Sal Khan, whose positivity is effusive, and at times, contagious.

“We’re at the cusp of using A.I. for probably the biggest positive transformation that education has ever seen. And the way we’re going to do that is by giving every student on the planet an artificially intelligent but amazing personal tutor.” - Sal Khan

We have a tendency to use technology to replace things that humans do perfectly well, but that we’d like to pay less for as a society.

Such as art.

Or pizza making.

For some reason deep tech startups and deep pocketed VCs decided this is something to automate, because it can save pennies on the dollar and increase margin on pizzas (which we have to imagine already have really good margins). Zume raised over $440M before self-immolating (burnt crust!). Stellar Pizza Inc, founded by SpaceX engineers, has taken up the mantle, recently raising $16M to make robot-pizzas.

We get that dropping the costs of production can make things ubiquitous, and that’s what we want in education, and probably–most of the time–in life. Opportunity should be ubiquitous! Pizzas should be ubiquitous! TUTORS SHOULD BE UBIQUITOUS!!

But AI isn’t making anything ubiquitous yet, and I doubt it ever will. AI is not increasing access. As Lawrence Holt, in Education Next, pointed out, only 4.7% students in Khanmigo’s efficacy study benefitted because 5% of anyone will probably succeed at anything (Dan Meyer has an insightful response to Sal Khan in Math World that’s inspiring some of our thinking). I think there’s a plausible explanation that these 5% are early adopters, and as AI accelerates, more students will cross the chasm like in most adoption curves. That’s certainly what people at Khan Academy think. But many students who would benefit from a tutor can’t access a digital one (digital divides, asymmetric information, etc.).

But I think there is a better argument to be made that, instead of focusing on reducing the cost of a tutor to zero, we find a way to provide high-dosage, high quality tutoring to everyone.

There is reasonable evidence that high-dosage working works. There is reasonable evidence that AI chatbots are not great for learning, especially for socio-emotional learning.

We have high quality tutors out there. I was a reasonably good tutor in our college literacy corps. Preply is filled with good tutors: Teachers in Mexico or Taiwan who subsidize their income and use instructional tools. But Praktika wants to substitute humans with an AI-generated tutor. You lose out on almost all of the benefit that a linguistic interaction can have, like chatting about local politics or rambling about your favorite films, or deeply understanding how to conjugate the the future tense. Worst still, we will be distracted from the solution.

The solution is to build a sustainable pipeline of tutors, find the money (ie willpower) to scale it, and then iterate and make it effective. Ben Kornell has called for a National Tutoring Corps. That’s really hard and complex, but it’s needed. Kids need to learn now. Kids need positive human interactions and to benefit from proximal development. Tutoring also led to me joining JumpStart, to joining Teach for America, to teaching. It can help, even if just incrementally, our teacher pipeline.

Where do we want to go

Is this the end-goal of AI? To replace humans with lower quality, low marginal cost substitutes? We talk about the revolutionary nature of AI, and it’s sent Nvidia stock to the moon. But many of the futures it is promising are uninspiring. AI generating tons of awful graphic t-shirts and advertisements (Okay those two things are actually ubiquitous).

Customer support? Here’s a chatbot. Talented copy writer? Chatbot. Photographer? Here’s an AI-generated drawing with awful hands. Making a pizza!? Tutoring? Here’s a chatbot.

Kids just went through a two year pandemic where learning was interrupted, they couldn’t socialize with their peers, and they missed out on proximal development by staring at screens across town from all of their classmates. Why do we assume that they would want or need more of that? Do we think so little of tutoring? Of learning? Of kids?

What person actually wants to learn Spanish from Praktika’s tutors, when Preply offers a very human, very capable experience that is already arbitraged across borders? AI will change our lives. No doubt. But it won’t change tutoring and education in the ways that we need.

These technologists do not seem to appreciate the wonder that human teachers provide when they launch a lesson with a great hook, or probe student thinking. Having a human voice on the other end of the phone or service interaction when you are at your wit’s end can change the course of your day. Would Mystic Pizza have been a classic if a robot made all the pies?

Judging by this AI-generated movie poster, the answer is YES.

Khan Academy and Khanmigo are doing some really good things, but we need to help children navigate and interact with the world around them. Not with a chatbot.