Friday, August 05, 2016

10 ways BOTS can be used in learning

You know that bots are coming of age when Google hires comic writers from satirical site The Onion and scriptwriters from Pixar and they're being launched on major learning platforms such as Duolingo. They know that real conversations between humans and machines need to cope with light conversation idle talk and humour. The banter has to get better if we are to use voice or text activated bots regularly. Facebook, Amazon, Microsoft and Google are all in the chatbot game and dozens of startups are creating bots - MykAi (banking), GoButler (personal assistant), GoodService (Concierge). Conversational interfaces and conversational commerce have arrived and, as Chris Messina, Uber’s ‘experience’ guy says ‘chat is the new black’. Messaging services are among the most popular services online and young people have flocked to them, away from the more staid posting. Messaging is BIGGER than social media. Read that one again. Why? They’re natural and liberating, more akin to normal human behaviour than other interfaces. Gartner have predicted that by 2020, 85% of customer interaction will be through bots. That I doubt, but the figure will be substantial. Whatever the future, they are here to stay.
Socrates – the first learning bot
Socrates was probably the first bot or proto-bot. Having never written anything himself, Plato used his name in a series of dialogues, to tutor younger thinkers by asking questions that exposed their inconsistencies and lack of knowledge. The ‘Socratic’ method is exactly that employed by modern bots, though Plato is replaced by the invisible hand of AI (Artificial Intelligence).
What’s given bots wings are recent advances in AI-driven text and speech recognition along with machine learning and deep learning. This increases their speed (latency can be a problem), efficacy (they are better) and their ability to get better (they learn). In the same way that Socrates remains a model teacher, so Socratic bots may be useful in education and training.
If you’re on Twitter you are likely to have followed, or have follower bots and if that attractive model wants you to follow him/her, he/she’s probably a bot. That’s their primary skill, or trick, to weakly pass the Turing test by fooling you into thinking and/or behaving as if they were human. This has already happened, from ELIZA (one of the first AI conversational programs) to Little Ice (Chinese bot that has had hundreds of millions of conversations). Customer service bots are now common and Facebook.IBM and Microsoft have launched bot frameworks. These allow you to use bots that front AI systems to deliver business services. Interestingly, you needn't believe that a bot is human, as Duolingo have found that it is NOT being human that is the advantage in language learning. More of that later. Could this form of AI replace teachers? I've written about this here. This article focuses on the role of bots in teaching and learning.
1. Firendly face and voice
We are susceptible to chatbot dialogue. Nass and Reeves, in 35 studies of learning, published in The Media Equation: How People Treat Computers, Television and New Media Like Real People and Places, (summary here) explored this susceptibility in detail. They showed that we attribute human qualities to technology, especially interactive, computer tech that responds to our requests and actions. When Steve Jobs fought with Steve Wozniac to get the first Apple computer to open with the word ‘Hello…’, Wozniac couldn’t see the point – Jobs was right and that obsession with user experience became the driver for Apple’s success. That was the computer as bot.
We have now moved beyond this natural propensity towards seeing technology as having human qualities and agency, to actually creating AI technology, which is as good as, even surpassing human abilities. This started with tasks in specific domains, namely chequers, crosswords, scrabble, chess and now Go. This has accelerated exponentially to produce real-time trading and self-driving cars. It would be naïve to imagine that AI will be used in every form of human endeavour, OTHER than teaching. With bots, which are really just the front-end of certain forms of AI, it can surely be harnessed for learning.
2. Language
NLP (Natural Language Processing) has given us huge success in language recognition, whether text or speech. It is this form of AI that lies behind Siri (Apple), Cortana (Microsoft), VIV (bought by Samsung) and other voice systems. Most teaching is done via speech, and dialogue remains a key teaching skill. What powers, not just single query bots but longer dialogue, is AI. It all started with Markov, a Russian who used maths on Pushkin’s poem Eugene Onegin, to create Markov chains. This, with a battery of other AI techniques, has given us real speech and text recognition, essential for bot dialogue. 
Amazon’s Echo, Google Home, along with speech recognition on a range of computers, tablets and mobile devices, is making frictionless search, queries and transactions practical, as consumer technology. As the AI software raises recognition to 95% and above, consumer acceptance kicks in and with volume of use comes social acceptance. This new era of natural language interfaces, will bring opportunities for teaching and learning way beyond the current search or stilted e-learning.
3. Teaching assistant
When 300 AI students at Georgia Tech were fooled by a teaching assistant, that turned out to be a bot, we got a glimpse of their power. It was trained using data from previous electronic dialogue, to answer queries from students but only answered when it had a 97% certainty. So successful was this teaching bot that it was put forward by one student for a teaching award. Smart as these students were, they only realised it was a bot because it was too good – it replied almost immediately, something the real teaching assistants never did!
This idea, of a teaching assistant, that takes the admin and other trivial and repetitive tasks out of the teaching process, must surely be a laudable goal. Much of teaching, offline and online, is actually administration, which could surely be better handled by an intelligent bot. The fact that such bots can learn (through machine learning) means they can tailor themselves to specific subjects, courses and teachers. Assistive bots, therefore solve that age-old project of excessive paperwork and admin that so many teachers complain about.
4. Manage learning
Bots could help teachers and trainers manage delivery through the management of tasks through a LMS or VLE. Optimal scheduling, booking and delivery and other definable processes, could be requested and executed by a smart bot. We already have concierge and customer service bots that help people with queries and organisational tasks.
5. Curation
Content curation can be bot driven, as it crawls the web looking for relevant content based on filters deliverd as dialogue. Voice systems such as VIV already handle complex queries with Boolean logic. Such systems are rapidly being embedded in consumer technology like Amazon Echo and Googlr Home. Lesson plans and resources could be designed, found and shaped by bots. They could draw upon databases of existing lesson plans and crawl the web for suitable content.
6. Online subject teaching
Moving up a level to online teaching, bots have become learners in the sense of being capable of learning from real human data sets that capture human expertise. This can be subject knowledge, where content can delivered by an intelligent bot that has a broad and deep knowledge of a subject, way beyond that of even a trained teacher. With access to knowledge-bases, greater than anything a human memory could hold, at some point the queries and answers a learner would want within a subject domain could be handed by a bot. We saw evidence of this with Watson, when it beat the two World Champions in Jeopardy.
Subject bots have huge potential to teach, especially at more basic levels, where 101 courses are run repeatedly. Whenever there is a human task that is repetitive and replicable, it tends to be automated. I have seen this operate in HE courses from history to science, where significant attainment rises, along with correlated reductions in drop-out happen. This will happen in teaching at levels where it is possible to replicate the tasks.
7. Online teaching
Beyond this query level, where bots answer student queries, bots can be driven by algorithms that embed good, evidence-based, learning theory. Chunking, search, concept identification, recommendations, relevant feedback, spaced-practice – there are many principles in teaching and learning that can be ‘captured’ in software and replicated by technology. It is not unreasonable to suppose that the qualities of an expert teacher be captured and replicated, especially if the human delivery side can be replicated through bots.
Those elusive qualities in teaching - inspiration, motivation and emotional intelligence – may also now be possible through AI bots. Sentiment analysis is available through a number of APIs on the internet, allowing bots (AI) to real the emotional state of the learner, such as disinterest, boredom, puzzlement and so on. On motivation, online activity can be tracked and used to signal flagging students, then give them support to stay on track.
8. Language learning
Learning a language is tough. Millions try in school and fail. Millions try with online learning and fail. The one thing that is missing is often conversation, which is what language use is largely all about. Duolingo’s chatbot is a good example. They found that bots fill a gap. Most people learning a new language are too embarrassed to speak to a real person who know that language. A bot, which you know is not a real person, is a great middle-ground substitute. The interesting thing about bots is that they learn or at least widen their responses based on use. The more people who sue the bot, the better it gets. As Luis von Ahn, founder and CEO says “We’ve done the measurements: we know we’re as good as a classroom, a standard high-school classroom in the US. In a standard US classroom, kids are getting a minute of conversational practice a day. But we would like to be as good as a human tutor, and that’s where we want to go.” That means an all-purpose bot.
9. Performance support
We learn much of what we learn, not on formal courses, but in real-time, responding to real problems and overcoming those problems. In corporate learning and development, this form of needs-driven, just-in-time, online delivery is called performance support. Many attempts have been made at this but most fail because they don’t deliver on the promise of accurate and relevant help. With AI-driven support, the system learns from previous queries, successful solutions and adapts towards future efficiency. It learns by watching you learn, then delivers better learning. Bots that focus on customer care do exactly this, learning to answer queries and solve customer problems, learning how to do so dynamically. This sort of realtime performance support would be ideal in workplace learning.
10. Online assessment
There’s formative and summative assessment, although given the emphasis on the latter at the expense of the former, you see why education has become such a chore for learners. In formative assessment, the idea of bot-driven feedback makes sense, especially where teachers have to cope with large numbers of students (beyond a handful), where personalised, realtime feedback becomes impossible.
A good example of how far we’ve come on this front is in search, where every letter you type into Google, triggers a search query and, using AI, tries to provide an answer. In writing, spelling and grammar checkers do a great job in spotting your typos and mistakes. These increasingly use AI to do the checking. If your organisation uses a plagiarism tool, such as turnitin, then you’re using AI to spot things that have not been written by the supposed student. This is moving towards essay marking software that can be taught by training it with human data and allowing it to learn from every essay submitted.
Problems
Let's not get too carried away however, much of this is future implementation. We only have to look at Microsoft's Tay to see what can go wrong. If a bot can learn, it can be taught bad things. In this case the bot Tay was taught to become a sex-crazed Nazi. There's also a problem with sustained dialogue. That problem is a matter of degree and will get better. Oddball output is another common feature as the AI guesses badly.
Conclusion
What makes teaching bots not only possible but desirable, is AI. Technology gives them some beneficial staring points. A bot is free from cognitive biases along with racial, gender and socio-economic biases. They never get ill, don’t forget much of what they are taught, operate 24/7, and can deliver from anywhere to anywhere where there is an internet connection. Unlike our brains they don’t sleep for eight hours a day and, in a fatal objection to human frailty, neither get burnt out, retire or die.
Bots are already out there, operating without us even realising that they are either there, or not human. This is bad if they’re being used to defraud us or fool us in some malicious way. Bots, such as Microsoft’s Tay, have also been turned into sex, crazed Nazis, by deliberate false training on the web. On the other hand, if used for good, think bots-for-good, in education and training, they take some of the drudgery out of teaching, provide valuable assistance, support learners, even teach and assess. This technology enhanced teaching would be wonderful if it helped tackle the real problems we face – attainment gaps, high drop-out rates, education in poorer countries, expensive HE that produces huge debts for governments and students. Bots ahoy!

No comments: