Meet the founders of 'revolutionary' AI being rolled out in schools

Bot exam markers, robot tutors and AI-written reports are all now being used in schools as they look to ride the tech revolution to slash workload and boost outcomes for pupils. Jessica Hill and Jack Dyson speak to the founders behind some of the products being rolled out in schools…

An AI developer is trialling new software that could mark batches of pupils’ work after being sent images of their exercise books.

Mark Unwin, the chief executive of four-school MAT the Create Partnership Trust, toiled away during the summer to help create online platform askKira.

The CEO, one of the website’s three founders, said it has been used to plan lessons, digest leaders’ emails and even pen resignation letters since its launch in October.

But over the past two weeks, the team has been testing early versions of an update that could see it mark and provide feedback for up to 30 pieces of work at a time.

“Kira is like Jarvis from Iron Man, [but] for teachers,” Unwin said. “Teaching used to be a profession you could fit around your life, [but now] recruitment and retention is unbelievably poor. I think AI offers a route back to something more manageable if done correctly.”

As with ChatGPT, subscribers can send messages to the AI through a prompt box online – but beforehand they need to “wake it up” with the words: “Hi Kira.”

More than 2,000 people have signed up so far. Figures for December show about 14 per cent of users asked it “for help with maths”, while a similar proportion engaged it for lesson planning.

Unwin said it can also “read” CEO newsletters to pull out the most salient points and write social media updates or internal messages to staff.

Trials of the marking software – which will work for subjects including maths and English – started in January.

It is being tested by 100 school workers, including trust leaders, headteachers, classroom assistants and sencos.

Unwin described this as a “breakthrough” in his bid to slash workload.

“It will work with [up to] 30 photos so that by the end of the day you can make a pile of your books, take pictures of them, and drag and drop them into Kira. It will read each one and give them three ‘stars’ and a ‘wish’.”

It will be used for “any written” subject, like English, history and geography.

Unwin added that it would also judge the reading age of the child, provide individual feedback and suggest areas for teachers to address the following day.

“AskKira is programmed to only do formative marking at the moment for all students’ bookwork,” he said. “It will look at the work and how it could be improved – it’s not programmed to look at a piece of work and say, ‘this is GCSE grade 4’.” 

Unwin stressed the website will always remain “free in part”. But since February 1, those wanting unlimited usage of the programme must pay £4.99 a month.

The marking software is expected to launch towards the end of this month, costing £24.99 per month. 

‘We could mark the nation’s exams in hours’

Meanwhile, the makers of another AI tool claim it can deliver exam results to the country’s pupils in just a few hours.

Vision Marker is the brainchild of father–son duo Barry Lambert, a chief examiner who has written more than 50 GCSE papers, and Dr James Lambert, a machine-learning expert who previously ran an algorithmic hedge fund.

Dr Lambert has previously led science teams on projects including modelling gut micro bacteria, while his dad has designed assessments for GCSE and A-levels.

Vision Marker can mark handwritten exams, Dr Lambert claims, even considering where words have been scribbled out and provide feedback to learners to help them improve.

Manual marking takes weeks to complete, which Dr Lambert claims “delays results to students and hinders their outcomes”.

He claims his company could deliver exam results “in a couple of hours out to the entire nation, while also giving a detailed, auditable pathway of how the marks are awarded”.

Dr Lambert claims their tool has proven itself to be “more accurate than [human] markers”, who are “prone to error making in these highly repetitive tasks”.

Exam regulator Ofqual told Schools Week using AI as a sole exam marker was “not allowed”. However, Vision Marker is marketing the tool as a “co-pilot” for marking, rather than a complete solution.

Vision Marker is used in 25 schools on practice papers. The Lamberts are also working with exam boards, as well as some multi-academy trusts and higher education institutions, around using their services in the future. However, they won’t reveal which ones.

Ofqual’s pilot innovation service is helping awarding organisations to navigate the regulatory framework around AI. 

AQA’s chief executive Colin Hughes said in November that his organisation was interested in the potential for AI to mark exams, describing it as being “very easy” to have “machines marking human markers”. But “public confidence” was stopping them from doing so.

Exam marking is also a popular way for teachers to top up their wages each summer.

There has been controversy in recent years over the pay of some exam markers. Dr Lambert claims examiners are “paid below minimum wage for what they do” and that his tool “turns markers into moderators”.

“They evaluate the marks that we give and use the [learner] feedback to agree or disagree with our system. It’s the regulator’s dream rather than their nightmare, because we have a fully auditable process of how and why marks are awarded.”

Speaking more generally, edtech investor Richard Taylor, who is managing director of MediaTaylor, cautioned: “People in both edtech and education are prone to making stupid predictions.”

He pointed to the example of Pearson, saying in 2009 they had an automated exam marking system that replicated the accuracy of human markers.

“While hype may be seen as part of edtech’s DNA, buyers and users should always be very sceptical about claims that range from terminological inexactitude to pure snake oil. In my 20-plus years of looking at thousands of pieces of edtech, perhaps 5 per cent were either investable/likely to succeed, or worthwhile, educationally. It’s the 95 per cent you have to avoid.”

Speaking at the BETT edtech show last week, education secretary Gillian Keegan said: “We should have the same expectations for robust evidence in edtech as we do elsewhere in education.

“Ed tech business should be leading the way – being transparent with buyers and promoting products based on great evidence of what works.”

‘We’re not trying to take teachers out the loop’

Developers have also warned of AI’s limitations. SchoolOnline has recently launched an auto-marking function. Through this, English assignments between key stages 2 and 4 can be assessed against UK national curriculum and exam board criteria.

“Traditionally it took two hours to mark a key stage 2 essay, but we’re turning that into two minutes,” SchoolOnline CEO Lucien Bowater explained. “You are radically reducing the teacher workload.”

The software will offer feedback on a per-student and whole-class basis, and “recommend to the teacher a next step”. It can also create a subsequent lesson plan.

Despite this, Bowater stressed marked assignments aren’t sent straight back to pupils as “we believe the teacher has a really important role in reviewing and editing it”.

“The AI won’t always get it right. The vast majority of the time it’s right, but we’re not trying to take teachers out of the loop.”

Peter Gravell, of Real Fast Reports, which helps formulate school reports and UCAS references, also noted that early forms of AI were “inherently quite biased” in terms of gender and race.

Those using his paid-for service – which has 15,000 subscribers – plug in bullet points of information they want to include in each of their pupils’ reports. To eliminate the risk of these biases, Real Fast Reports changes the names of children to those in a pre-set list “of western standard names” and categorises all of them under the same gender.

“Another problem they [AI] have is they never know when they’re right or wrong… so they won’t flag when it be might wrong – it always gives the best answer it can give,” Gravell continued.

“No system’s perfect, so once the report’s written, teachers are encouraged to proofread it and make changes where necessary.”

Here come the robot tutors

But it’s not just marking and workload reduction being targeted. The learning platform Tassomai claims it’s new virtual tutor, Mai, can do almost anything that a human tutor does, but for a fraction of the cost. Its founder, Murray Morrison, believes that one day AI like his will do away with the need for tutors altogether – levelling the education playing field in the process.

Mai uses a trained version of OpenAI’s GPT3.5 to chat to learners, in the same way that an on-demand human tutor would guide them with their homework. Since January, it has been made available to all the 600 secondary schools in England that use Tassomai’s assessment platform, on which more than two-million pupil questions are answered every day.

It was while Morrison was propping up his teacher salary by tutoring for members of the British royal family and celebrities – including Elle Macpherson’s kids – that he realised how unfair it was that “people who can pay for it” get “amazing tutoring”, but “everyone else is missing out”.

Tassomai, which Morrison founded in 2013, helps plan and mark kids’ homework. With the addition of Mai, learners get help when they don’t understand something and instant feedback, without giving away the homework answers.

Morrison believes parental permission is not required for them to use Mai. The app does contain an opt-out function for parents, but Morrison is unaware of any parents having used it.

He points out that “our contract is with the school and the school owns the data”.

 However, one school with a Conservative Christian ethos insisted upon Mai not being able to teach some aspects of biology. Morrison describes such instances as causing a “real dilemma” for companies such as his.

“What if we were asked to write content for the Florida School Board and they don’t want us to say that evolution is true? That’s a whole other conversation.”

He also admits there have been “lots of hiccups” in teaching Mai how to tutor, as large language models are “not as clever as you want them to be… the art comes in training it to do the right thing. It’s certainly not just a plug in and play process.”

But he believes AI is rapidly becoming more sophisticated, saying it is “not beyond its reach” to replace tutors altogether within the next two years.

“When tutoring is as expensive and inconvenient as it is, there’s an opening there.”