Perhaps second only to AR/VR applications in the education market (as we covered here for the EdTech Times), AI (or Artificial Intelligence) and related concepts of virtual tutors and student facing “chatbots” was on full display and discussion last week at SXSWedu. This should not be a surprise to any careful follower of the markets as venture investors poured over $4 billion into general AI technology start-ups last year1)A site called botfunded.com tracked over $200 million of venture investment in “chatbots” alone while CB Insights plotted a handy little timeline of such investments right here. And as early as the second week of January, this author was already snarkily suggesting would-be edupreneurs rebrand now to seize upon this hot trend.
Of course, edtech has been through the hype of artificial intelligence before with the 2011 – 2013 mania for adaptive / personalized learning – back when every publisher and school platform had to acquire an adaptive learning start-up.2)This was also when Knewton could still raise large venture capital rounds in Silicon Valley without having to schlep to Shoreditch, Shanghai, or Singapore. Actually, Knewton was perhaps a bit too early for the new “robo-tutor” buzz as all the way back in October 2015, their founder was fanstasizing about “Mind Reading Robo-Tutors in the Sky” to NPR.. Of course, hype need not get in the way of actual sustainable success as seen in the several bootstrapped adaptive learning start-ups that are actually scaling well outside of Silicon Valley (a forthcoming post to be published here) or all of the “Online Program Management” companies that are still scaling even after venture investors deemed the space too crowded (2U is up 2oo% since its IPO and other players like All Campus, Academic Partnerships, Everspring and HotChalk are growing profitably).
To that end, this post is intended to provide a simple survey of three current projects in education and the associated challenges facing AI in student tutoring and support.
At its most basic, AI-based chatbots can provide a similar role as is found in fairly pervasive customer service functions. A team in Norway, the EdTech Foundry, is leading a $1 million research project with a group of local universities to do just that: saving professor and classroom time by automating answers to students’ oft repeated questions, along with follow-up auto-messages and reminders. After applying rule-based “bot logic” to specific course curriculum, they found that many students switched from consulting the lesson plan or teacher to just asking the bot. Their pilot resulted in:
- Up to a 5x higher engagement on messages posted by a bot as compared to a human teacher
- Up to a 53% click through rate on educational content posted by a bot
The team anecdotally pegs their improved student engagement to two unanticipated benefits of bots: 1) when a professor posts, the LMS posts to the whole class, but with a bot, it comes across as a seemingly personalized direct message; 2) the bot is considered a lower stakes actor with no pressure to reply nor a fear of being assessed in the interaction. While EdTech Foundry has referred to these as “pedagogical chatbots” and are seeking to further apply machine learning, they are still quite limited as the data required to “train” the bots is not available (after all, which professor has literally saved all of their former students’ old questions?). Moreover, answering questions only takes a bot so far – most questions are never asked (as students fear appearing dumb).
Another start-up, Check It, has developed a chatbot to helps K-12 schools in NY talk with students in the format they like most – text messages. This app allows teachers to create their own content and transform the delivery of curriculum into conversations that can be followed up with questions. Teachers are able to program Check It’s chatbot to follow their scope and sequence, checking to see if students are ready to answer standards-based questions from state tests. Check It sends the questions and students text or select their answer choice, provide students with feedback and tips. In one example, at MS390, the chatbot has been programmed to send questions from the NY State Test in Math and ELA everyday. The school has installed the Check It leaderboard on a screen located at the entrance of the school. Students compete on a weekly basis to see which class gains more points by answering questions from the State Test.
Building off these more targeted “chatbots”, Pearson is seeking to build further through their partnership with IBM Watson 3)This appears to be somewhat of a shift in strategy back to enterprise partnerships for IBM, as after the initial splash of their announced $100 million Watson corporate venture fund in 2014 along with their former General Managers seed fund for even earlier staged AI start-ups, neither has appeared to be actively investing in start-ups aside from another recent partnership with Indiegogo and Arrow Electronics. At the Pearson Lounge at SXSWedu, their SVP of Personalized Learning and Analytics, Dr. Angie McAllister, presented the “virtual tutor” inside their Revel platform developed in partnership with IBM Watson to provide real instructional and learning impact value. Revel, powered by Watson augmented and artificial intelligence and Natural Language Processing features, can help students based on where they get stuck and what they need to get to their goals along with rich, data-driven decision-support for instructors to provide personalized learning to even more students.
Seen clearly as a supplement to educators, the typical fear of AI related job loss is less relevant for education as it is in industries like long-distance hauling, but education presents its own unique set of risks.
Both of the two above presentations generated audience criticism that AI prevents students from thinking holistically and learning to prepare. But as Dr. McAllister noted, you don’t care what a teacher knows until you know that your teacher cares about you. There is no amount of computer programming that can replicate empathy, but in homework and self-study away from a teacher, students too often rely on Google-search queries. Google can take a student in the wrong direction, but even if it does generate relevant content, this is not research and certainly does not mean a student learned the original objective. A virtual tutor provides unbiased, 24-7 available, eternally patient student support that is as much a learning experience as it is an assessment, further providing insight to the teacher and student around their level of mastery and where the next classroom lesson should go.
Its not hard to imagine how AI might also cross the education industry’s dreaded “third rail” of data privacy (what data is being collected on student learning, how is it being used, and who sees it?). Just last month an in-depth autopsy of the massive failure of inBloom was published, but as a closed-end system with an easily understood use case / benefit (tutoring!), AI chatbots should stir less concern. Still, Dr. McAllister (noting her husband is an ethicist) was keenly aware how the “do no harm principle” must be a base line in any such system and recommended that a governor be installed on devices to ensure data is not being used improperly.
Finally, educators also understand there is a distinction between data algorithms and “wisdom” (i.e., something an educator knows in their gut about a student that AI cannot). An audience member also described the risk of “curated perspectives” with students narrowly steered down a finite number of available learning pathways. Dr. McAllister agreed the artistry – and not just the science – of instruction is in curating an infinite number of such pathways and that the design principles of virtual tutors must be informed by this. She further espoused four pillars for AI in education:
- Transparency (to be clear about whats happening to all users and their data);
- Auditability (tracking data, even what has been purged);
- Incorruptibility; and
References [ + ]
|1.||↑||A site called botfunded.com tracked over $200 million of venture investment in “chatbots” alone while CB Insights plotted a handy little timeline of such investments right here|
|2.||↑||This was also when Knewton could still raise large venture capital rounds in Silicon Valley without having to schlep to Shoreditch, Shanghai, or Singapore. Actually, Knewton was perhaps a bit too early for the new “robo-tutor” buzz as all the way back in October 2015, their founder was fanstasizing about “Mind Reading Robo-Tutors in the Sky” to NPR.|
|3.||↑||This appears to be somewhat of a shift in strategy back to enterprise partnerships for IBM, as after the initial splash of their announced $100 million Watson corporate venture fund in 2014 along with their former General Managers seed fund for even earlier staged AI start-ups, neither has appeared to be actively investing in start-ups aside from another recent partnership with Indiegogo and Arrow Electronics|