We will teach machines to learn. But what will be the consequences of them taking an increasing role in teaching? Sam Smith argues that the growing use of machine learning in teaching and marking students’ work, risks undervaluing and losing the unquantifiable skills that drive diversity, creativity and innovation.
It was a 19th century folly that there was a hierarchy of progress - a canard that placed the aboriginal societies of Australia at the bottom, and the Strand in London at the peak. History may not repeat itself, but it certainly rhymes.
For years, automatic plagiarism checkers have had a role in higher education. As students submitted millions of essays to these platforms, and teachers used them to mark or comment on submissions, they provided the raw material for machine learning systems to offer their own advice. At first these were simple comments on grammar, spelling and word count. But they learned and evolved. Some now offer a full grading service, including content, style and use of genre for college-level essays.
Progress doesn't end. What happens after we pass the point where the machines teach as well as mark? How will algorithms and interactions evolve over time when they have started both teaching and grading papers? What measurements will they use to judge the worthiness of work? How will our education systems, and the values that they teach and demonstrate to our children co-evolve with these algorithms?
We only need to look at Universities today to see who does the teaching, the marking, and who makes the decisions (and equally, the variable quality of some teaching). Institutions and their structures often have perverse incentives; they must weigh the humanity of the post-docs responsible for the majority of teaching, against the priorities of the administration which cares primarily about what is measured and analysed, especially by authorities and by rankings read by prospective students. Is that all that matters? How will the algorithms know?
The Secretary of State for Education will at some point see a great saving in the teacher wage bill from having machines mark the exams, and then homework, and progressively more - if nothing else, the algorithms (probably) won’t be unionised (initially). Will the best algorithms and feedback be limited to the best schools or richest parents? The opening up of opportunities is powerful, but demonstrates the potential scale of the impact of these teaching machines on societies. They may have started with English essays, but they've expanded past that, and will continue to do so based on commercial interests.
The capacity of a human teacher to inspire with compassion and generosity is unlikely to be replicated by even the most advanced algorithms
While the first exam-marking algorithms from Shoreditch may not have the same effect on the playing fields of Eton as the first British explorers had on the Australian population, all decisions have consequences. What is the culture shift that these algorithms will create? People may prefer to be judged by people, but people’s preferences and raw humanity tend not to influence such high-level bureaucratic decisions.
When children learn they must please the algorithms to get good grades, will the knowledge and skills that generate new insight begin to disappear? When the pipeline for this is cut off in many secondary schools, the innovations of the future will not come from the many, but from the few who could afford human teaching, or were able to game the system and realise that not all that matters is measured. Or will schools with fewer resources gain more because of universally applicable algorithms? Will there be universally available algorithms? Or will development primarily assist certain types of student? Whatever happens first, decisions and consequences will evolve over time based on momentum, motivations and incentives.
The UK has disproportionate impact in the world in part because of its creativity (ok, and its nukes). The capacity of a human teacher to inspire with compassion and generosity is unlikely to be replicated by even the most advanced algorithms in the near future. It is that human touch - the human belief that makes life worth living.
Unfortunately, humanity and inspiration are difficult to quantify, and hard for machine learning systems to include in their assessments. These human qualities are not something that Government bureaucracies and hierarchies are renowned for promoting. Will humanity be a measure by which algorithms are chosen? Or will it be the best sales pitch to decision-makers? That would require a new humble acceptance of consequences of power and decisions.
To promote creativity, the whole basis of soft power, we must always choose wisely. Given the reasons and motivations of some who decide, that seems unlikely.
This blog is part of the Shadow of the Smart Machine series, looking at issues in the ethics and regulation of the growing use of machine learning technologies, particularly in government.