What Einstein can tell us about AI in education

He’s the ulti­mate sci­ence teacher. History’s most bril­liant physi­cist brought back to life, promis­ing to unlock the mys­ter­ies of his sub­ject. He’s approach­able too, with twinkly eyes and a mis­chie­vous Teu­ton­ic lilt.

This is Dig­i­tal Ein­stein, a chat­bot that recre­ates the great man’s face and voice. The AI-pow­ered vir­tu­al Albert can set quizzes, answer ques­tions and even offer opin­ions – just ask him if he thinks the Earth is flat.

This is the prod­uct of an inter­na­tion­al col­lab­o­ra­tion involv­ing teams in Europe, the US and New Zealand. The voice ele­ment has been pro­vid­ed by Aflorith­mic, a Barcelona-based com­pa­ny whose COO, Matthias Lehmann, sees edu­ca­tion as a key use case.

“By using these chat­bots in schools, you could bring sub­jects to life for kids,” he says. “Wouldn’t it be great to learn from these amaz­ing fig­ures?” 

But beneath the glitzy visu­al ren­der­ing and the pat­tern-recog­ni­tion soft­ware that pro­duces Dig­i­tal Einstein’s tai­lored respons­es, there are ques­tions. Are these chat­bots appro­pri­ate for edu­ca­tion, with its empha­sis on fair­ness and mutu­al devel­op­ment? Just because we can bring these titans back from the past, is it right to entrust them with our children’s future?

Until now, such ques­tions would have been aca­d­e­m­ic. Chat­bots were always too unso­phis­ti­cat­ed to have any use­ful role in edu­ca­tion. Unless the con­ver­sa­tion stuck to a planned flow­chart, they would get con­fused. 

But oth­er forms of AI have been infil­trat­ing our schools. Intel­li­gent learn­ing pro­grammes, which give each user a per­son­al edu­ca­tion plan tai­lored to their exist­ing knowl­edge, are already help­ing thou­sands of stu­dents. Using games and quizzes, these pro­grammes offer tips on how to devel­op at a suit­able pace and rec­om­mend which mod­ules to study next.

Mark Ray­mont, a maths teacher at All Saints Church of Eng­land Acad­e­my in Ply­mouth, is an ear­ly adopter and fan of such assis­tance. Hav­ing imple­ment­ed Sparx, an AI-based sys­tem that pro­vides bespoke home­work plans, he reports that his class­es’ over­all home­work com­ple­tion rates have improved to more than 95%. 

“Instead of increas­ing home­work, it has actu­al­ly stream­lined it,” says Ray­mont, who adds that the tech­nol­o­gy has pro­vid­ed fur­ther ben­e­fits in the class­room. “We use Sparx to iden­ti­fy trends. Our sys­tem gives us more in-depth insights into how a class, or an indi­vid­ual stu­dent, is doing. It enables us to pro­vide more sup­port to dis­ad­van­taged learn­ers. Par­ents have been very pos­i­tive.”

Eureka moment

The recent sharp increase in chat­bot capa­bil­i­ty has result­ed from advances in nat­ur­al lan­guage pro­cess­ing (NLP) tech­nol­o­gy and con­tex­tu­al AI. These have giv­en the bots more emo­tion­al intel­li­gence and enables them to go off script and have back-and-forths with users. Instead of sim­ply map­ping out a student’s learn­ing path­way, they can be used as vir­tu­al teach­ing assis­tants.

Take-up has large­ly been lim­it­ed to a few uni­ver­si­ties, which have built their own Ein­stein-style sys­tems. But there are signs that the tech­nol­o­gy is trick­ling down.

The tech team at Bolton Col­lege, a provider of fur­ther edu­ca­tion and voca­tion­al train­ing, has cre­at­ed Ada, named after the 19th-cen­tu­ry com­put­ing pio­neer Ada Lovelace. The chat­bot is equipped with IBM Watson’s NLP plat­form and dozens of ques­tion-and-answer pairs, crowd­sourced from teach­ers. It can field sub­ject queries via Android, iOS and Alexa, and pro­vide near-instant respons­es.

The project is still in its infan­cy, but Aftab Hus­sain, the college’s learn­ing tech­nol­o­gy man­ag­er, reports that his team is already speak­ing to oth­er organ­i­sa­tions with a view to shar­ing the tech­nol­o­gy. He believes that vir­tu­al teach­ing assis­tants will become a com­mon fea­ture through­out the edu­ca­tion sys­tem. 

“Every stu­dent and teacher will have one in years to come,” Hus­sain enthus­es. “When stu­dents from oth­er schools have come here, they’ve said: ‘Why haven’t we got one?’ My ambi­tion is that every child, from the age of three upwards, has a per­son­al dig­i­tal com­pan­ion to sup­port their stud­ies.”

The limits of genius

For all the chat­bot cham­pi­ons’ opti­mism, con­straints remain. Pro­cess­ing pow­er is a big one – Dig­i­tal Ein­stein can still only speak to a hand­ful of peo­ple at a time, for instance, and answer about 300 ques­tions. 

The ulti­mate goal, for both Dig­i­tal Ein­stein and Ada, is to have mean­ing­ful respons­es to a vast num­ber of poten­tial ques­tions. But, even then, it remains to be seen whether chat­bots can branch out beyond fac­tu­al answers into sub­jects that require nuanced dis­cus­sion.

Wayne Holmes, a con­sul­tant on AI and edu­ca­tion at Unesco, is scep­ti­cal about that prospect. “So far, they can work only in spe­cif­ic domains, such as math­e­mat­ics, where there’s a right answer,” he says. “In sub­jects such as Eng­lish, they are less effec­tive. I don’t believe that any stu­dent will ever be able to have an in-depth con­ver­sa­tion with a chat­bot.”

Holmes issues a fur­ther caveat – a par­tic­u­lar­ly salient one, giv­en that recent research has found that up to 80% of chil­dren in the UK lack suf­fi­cient com­mu­ni­ca­tion skills when they start school. 

“These sys­tems sep­a­rate chil­dren from their peers and the teacher,” he says. “They claim they can do the job bet­ter than the teacher, but they have a lim­it­ed under­stand­ing of what edu­ca­tion is about. Edu­ca­tion is a social activ­i­ty that’s about per­son­al devel­op­ment. It’s not just knowl­edge trans­fer.” 

Per­haps the biggest issue with vir­tu­al learn­ing assis­tants is a socioe­co­nom­ic prob­lem. Even before Covid-19, the gap between rich and poor stu­dents in the UK was among the widest of any devel­oped nation. This has only widened dur­ing the lock­downs. Accord­ing to the Nation­al Foun­da­tion for Edu­ca­tion­al Research, stu­dents with dis­ad­van­taged back­grounds fell fur­ther behind in 2020, los­ing far more learn­ing time than their more afflu­ent coun­ter­parts. With this prob­lem in mind, many edu­ca­tors will feel that it would be unwise to intro­duce more teach­ing tech for use on expen­sive devices such as lap­tops, tablets and smart­phones.

As with many dilem­mas about dig­i­tal adop­tion, there is prob­a­bly a mid­dle ground to be found for vir­tu­al teach­ing assis­tants. By using these in the appro­pri­ate sit­u­a­tions, pro­mot­ing social equal­i­ty and pre­serv­ing teacher-pupil rela­tion­ships, edu­ca­tion providers can bring the best out of a huge­ly promis­ing tech­nol­o­gy.

But how exact­ly will they achieve that? For now, at least, it’s a ques­tion that even Ein­stein might strug­gle to answer.