Bill Gates recently claimed that humans won’t be needed for teaching within a decade. What makes him say that, and how true is it?

Like most people looking at AI, Gates is almost definitely referring to the substantive forecasts that are increasingly widespread on the exponential growth of AI power, complexity, and reach between now and 2030. The detailed June 2024 paper, “Situational Awareness: The Decade Ahead,” by the 23-year-old AI researcher wunderkind Leopold Aschenbrenner, is a central piece of writing to this effect. If you have not read it yet, I would definitely recommend it. It’s a 165-page essay that forecasts the rate of change (measured by order of magnitude) of AI computing power that we can expect by 2027 at the earliest, 2029 at the latest. Although the essay has its faults (a lot of jargon, some repetition – no one’s going to write a 165-page essay on anything without repetition) and slightly gauche, sweeping political statements, it is carried by formidable energy and some logical forecasting based on past patterns and historical examples, most especially the speed at which nuclear fission technology grew.

Chat GPT went through cycles of development from 2019 to 2020 to 2023, taking its thinking power from a preschooler to a high school student (where we pitch it at present). By 2027, the expectation is that it will be as intelligent as an engineer at a PhD level. Now, this opens up the rabbit hole of what we mean by intelligence, what about emotional intelligence, and so on, but let’s not go down that right now and stick to a rudimentary understanding of intelligence as technical problem-solving.

It is not so much the straight line of technical intelligence that will develop but, with the rise of agentic general artificial intelligence, the ability of AI to behave much more like an autonomous researcher than a simple chatbot. This will be the fundamental breakthrough: when AI can take decisions that are not responses to commands but part of a more elaborate project, and this will involve internal monologues, exchanging with other AI agents and essentially operating like a co-worker or project manager rather than a mere entity that executes commands one by one. This will happen through what Aschenbrenner calls “unhobbling,” in other words, coders and researchers will be able to iron out those few bottlenecks that are still preventing AI from being completely autonomous. He’s convinced that this will happen, mainly because it will be done by AI researchers and not human beings, and also because since the days of ChatGPT2, when hundreds of millions of dollars were being invested in AI development, we are heading to trillion-dollar investments in AI clusters and the very major impact such resource allocation will imply.

Somewhere between 2027 and 2030, we will be in the age of superintelligence, where companies will onboard agentic AI to develop reports, projects, business plans, and more; medical diagnoses will be done with not-yet-seen levels of accuracy, and students will work with AI tutors.

This leads us back to the original question, will AI really replace teachers?

My answer goes like this: today primary and middle school teachers are pedagogues whose main task is to socialise young children, enhance their language acquisition, life skills, and core literacies. The emotional, disciplinary, psychological, and social dimension of learning is scaffolded, necessarily (I would argue) by humans. It seems difficult to imagine how this will change. How could a non-carbon entity seriously do these things effectively? The very basis of primary education, social constructivism, is based on the idea of human contact and learning to live together. How satisfied and willing would parents be to have robots teaching their children how to play with other children, how to spell, and do basic arithmetic? Imagine how bizarre the appropriation of social cues would be if done by a machine. I’m not talking about occasional gimmicks with robot dogs or one or two automated programmes to supplement learning, I mean the entire teacher being removed and replaced by a robot or programme. I don’t believe that primary teachers, or middle school teachers, will see their work outsourced in the near future, and there’s not much in any serious forecast studies on AI to support that.

However, the pedagogy of high school is less resilient to the AI revolution. This is because teachers have been subordinated to a role whereby they are there primarily to help students do well on tests, prepare them for high-stakes examinations, and go through mark schemes and past papers with them. This is driven by the nature of the end of high school assessment, which is overwhelmingly dominated by a teach-to-the-test model. I’ve written about it at length in my open-access book, “Changing Assessment.” To be clear, high school teachers have so much more to offer, and one of the tragedies of this assessment design is that creativity, both in teachers and in students, is squandered because of the factory-line of test performance. I myself, as an end-of-high school Theory of Knowledge teacher, see the last months of the course dominated by the assessment model, for the simple reason that it is worth points, and those points will be needed by students to get the best possible IB, and that will have an effect on which university they are admitted to, and so on and so forth.

Currently, high school students use technology and AI where they can to help them learn, be it through online tutorials, such as Khan Academy, past examination question banks, or even very simple search engines like Google to find answers, strategies, and advice. Who can blame them? The point is that these aids will become increasingly powerful and at an exponential rate.

Therefore, I think that Gates is actually right about the end of high school, in any case, that the core function of examination preparation will be superseded by GenAI, and fairly sophisticated face-to-face examination tutoring will be superseded by superintelligent AI tutoring. This is why we should start integrating AI tutoring into high school, as part of the pedagogic voyage: why not complement the socio-emotive scaffolding of learning, so integral to human contact, with intentional and institutionalised student access to an artificial intelligence tutor to do further testing, corrections, practice runs, and simulations.

High school teaching will only be superseded by agentive superintelligence as long as it continues to be chained to the teach-to-the-test approach, as long as we insist on assessing students on declarative, lower-order factual recall through high-stakes testing. It is not by keeping technology out of schools and gatekeeping an archaic system that we will be empowering students or teachers for that matter; on the contrary, the whole question should be articulated around what students should be learning at the end of high school in order to be ready for a world where superintelligence will be part of society and the world of work.

We’ve actually known what students should be learning for a while, in any case, as long as we have identified that the future will be constituted by VUCA. The competences we need to engender and nourish, develop and value are lifelong learning (the ability to bounce back, reinvent yourself, show curiosity, and the ability to find elegant, aesthetic, and practical solutions within defined parameters), self-agency (critical thinking, standing on your own two feet, innovation and entrepreneurship, self-confidence, and risk-taking), interactive competencies (how to work in a team, with rapidly evolving resources, and with the environment), transdisciplinarity (knowing how to transfer skills across domains, to learn from one experience and see how the learning can be applied to another set of circumstances), and multi-literacies (having signature strengths and unique levels of mastery in given areas, allowing you to stand out from the crowd).

So, in the final analysis, as artificial intelligence grows, teaching should adapt, but for it to adapt, assessment must be reformed, and for assessment to be reformed, examination boards and universities have to accept new ways of recognising student gifts, and for that level of acceptance to take place, widespread educational reform at several levels is needed, more than ever.

Photo by Andrea De Santis on Unsplash