Inbuilt biases and the problem of algorithms

As the founders of the Institute for Ethical Artificial Intelligence in Education, we are appalled by the manner in which an algorithm has been used to decide students’ A-level and GCSE grades (A-level results: No 10 hints algorithm set to be ditched in England, 17 August). The exam grading algorithm may not be sophisticated AI, but to be ethical it should still adhere to certain principles, including that it should impact positively on teaching and learning; be fair and explainable to the people on whom it impacts; and that it should use data that is not biased towards or against any particular group of people.

The algorithm used this year to decide the grades for students who have worked hard in extremely difficult circumstances adheres to none of these principles. It is, without question, unethical and harmful to education in general and a significant number of pupils in particular. It is a gross abuse of the algorithmic approach. Not all algorithms or all AI is bad, but its use this year for exam grading has rightly attracted criticism and vilification. The use of this algorithm should be abandoned and algorithms for use in future exam grading should be developed in line with ethical guidelines, as well as any additional principles that are necessary for algorithms used in education.
Prof Rose Luckin UCL Knowledge Lab, Sir Anthony Seldon Vice-chancellor, University of Buckingham, Priya Lakhani Founder and CEO, Century Tech

The exam fiasco illustrates the government’s misuse of algorithms. An algorithm can never predict the suitability of a student for university or further education. It seeks to provide the statistical probability of suitable students emerging from certain educational backgrounds. It tests the system, not the person.

Analogies may be helpful. An algorithm can never predict whether a particular teenager will be carrying drugs. It may be able to provide an opinion as to how many of those carrying drugs will be teenagers. It cannot predict whether an Oxford-educated ex-Etonian would make a suitable prime minister. It can provide a statistical opinion as to how many Tory prime ministers will come from Eton and Oxford.
Dr John Elsom
Kingston-upon-Thames, Greater London

Rather than argue the merits of modifying the algorithm, might it be time to offer an analogy with what Covid-19 has already taught us about the “necessity” of commuting and office working as against homeworking? Why, in other words, are we so preoccupied with exams? Are they any more essential to education than the office block is to the all the tasks traditionally undertaken there? Pulling the plant up to measure the roots is a peculiarly English vice, and one compounded by Conservative privateers who fetishise metrics as an essential adjunct to marketisation.

Obviously, universities need criteria to ensure that students have the aptitude, motivation and capacity to complete courses successfully. But perhaps it’s time to look at achieving a future balance between the role of teacher assessment, external examinations and aptitude testing for particular courses in higher education, just as we are working out which days to go into the office and which to spend working at home.
Paul Clarke
Horsham, West Sussex

Why not do away with working through 12 years of education and just have an algorithm from the start? It would save a lot of time. Now we know definitively that meritocracy is an illusion. What counts is where you live and what school you attend.
Calix Eden

Source: Read Full Article