In 1958, Frank Rosenblatt fed punch cards into a 5-ton computer the size of a room. By the end of the fiftieth trial, the computer had just about learnt to identify the difference between basic markings.
Just over half a century later, we now carry computers in our pockets that would have made Rosenblatt’s mind boggle. Artificial intelligence (AI) and machine learning (ML) regularly spot cancers; help run security, lighting and heating systems; and read and translate languages with increasing accuracy.
Unsurprisingly, this technology is also transforming the educational landscape.
Remote invigilation and online education has opened up opportunities to learners worldwide – allowing them to access knowledge and secure assessments, regardless of their location. We also know that advances in AI have the potential to revolutionise how exams are assessed. This evolution will have far-reaching consequences for high-stakes English language tests, such as IELTS.
Here we look at what the future may hold for English language testing as AI and machine learning continue to evolve.
The changing role of AI in student assessment
Computers, of course, have long been able to carry out rudimentary assessments. Multiple-choice testing can be efficiently marked by machines that don’t mind repetitive tasks and don’t make mistakes.
Over the last five years, though, AI has begun providing students with more complex feedback. This has included a Virtual Operative Assistant helping train medical students. AI systems have also supplied personalised learning plans alongside test results.
With AI increasingly able to make accurate predictions, English learning and assessment companies are also looking at ways to utilise its capabilities.
How AI is transforming English language assessment
In 2019, Bronagh Rolph, Assessment Group Manager at Cambridge Assessment English, predicted that AI would lead to greater automation of English testing. The most significant impact would be for the speaking element, which had traditionally required a human examiner.
That same year, online language teaching pioneers Duolingo were already updating their digital assessment to include long-form speaking and writing questions. The assessment enabled machines to randomly assign questions and reliably score answers, and it was built on extensive research.
Duolingo’s English language test, which 3,000 institutions currently accept, can be taken at any time, anywhere. Research has shown that although this test does not use human assessors, candidate results have a high correlation with exams that do.
Company co-founders Luis von Ahn and Severin Hacker developed the AI-powered test to democratise access to education. At a time when the global pandemic made travel and face-to-face examinations impossible, it’s unsurprising that 2020 saw demand for Duolingo’s test soar.
AI-powered technology is also behind Pearson’s Versant tests. Again, the digital assessment ensures a more streamlined, cost-effective process for employers assessing candidates’ language proficiency.
Cambridge Assessment English, one of the organisations behind IELTS, has also developed a new English test with a high-tech auto marking feature. Linguaskill can assess reading, listening, writing and speaking, and results are available in 48 hours.
All these developments are fuelled by the demand for convenient, affordable English testing.
The risks of AI in English language assessment
And yet, despite all the clear advantages of automated English language testing, IELTS retain their traditional model of in-person exams assessed by a human.
The argument for this approach has been that as IELTS is the gateway for many study and migration opportunities, the test must remain the gold standard in English assessment.
Undoubtedly, the use of AI comes with risks; the deep-learning algorithms used in many AI-supported assessments can be beset with bias.
These issues often stem from machines learning from a narrow data set. This has led to applications not recognising certain accents or skin colours and compounding existing prejudices.
These problems aren’t always straightforward to overcome, with biases introduced unwittingly at various points throughout development. Arguably, though, the unconscious biases held by human examiners are equally challenging to banish. As Andrew Selbst, assistant professor at UCLA School of Law, states, “‘Fixing’ discrimination in algorithmic systems is a process ongoing, just like discrimination in any other aspect of society.”
Steps are now being taken to mitigate these risks. For example, UCL and Pearson are currently working on a three-year project to ensure cutting-edge technologies provide an impartial and highly accurate English test.
UCL academic Dr Mary Richardson believes AI has the potential to lead to less biased assessment. As she told the UCL blog, “You want a system that gives you the marks you deserve, whether you learned English originating from America, Australia, Canada, India, Ireland, Malta or Hong Kong. All those different, slightly subtle local variations in vocabulary and sentence structure need to be built into the assessment model, so you get a fair mark. This is where artificial intelligence can really help us.”
The future of AI in English language assessment
As AI and machine learning become increasingly sophisticated, they will inevitably be a key feature in the next generation of English language tests. One future trend may be a hybrid model where AI systems and human examiners work together to produce a high-quality assessment.
There will continue to be a demand for affordable, student-focused English language testing in the post-pandemic world. However, the stage is set for the traditional monopoly to be challenged and for more innovative, relevant and modern alternatives to rise to the fore. Ultimately, this will benefit learners, and will help more people to access qualification and career opportunities which is surely the purpose of modern Education Technology.