Saturday, April 21, 2018

Lords report ‘AI in the UK: ready, willing and able?’ Let’s be honest - ready – no, willing – sort of, able not really…

Politicians love a good report. Problem is, we produce them like pills, in the hope that they will make things better, when all they do is act as a placebo. It seems as though things are happening but they ain’t. Whenever we are worried by something, in this case AI, we get a bunch of people, usually well past their sell by date to produce a ‘report’. To be fair this is a substantial piece of work, at 420 numbered sections and 74 recommendations, but it’s all over the place, lacks focus and at times is way off the mark.
Ethics heavy
First, I’m not sure about a document that tries to climb and descend a mountain at the same time. No sooner has something been stated as a way forward, than it’s drowned under a wave of repetitive moralising. Although they wisely stop short at blanket regulations, it full of pious statements about dangers, challenges and ethics. As Hume said, you can’t derive an ought from an is – and that’s exactly what they do, over and over again. It is hopelessly utopian in its assumption, even that AI can be defined, never mind regulated. Perhaps too much is attributed to its efficacy and promise. In the end it’s just software.
Crass identity politics
There’s the usual obsession with identity politics and the idea that bias in algorithms will be solved as follows,  The main ways to address these kinds of biases are to ensure that developers are drawn from diverse gender, ethnic and socio-economic backgrounds. Oh dear – not that tired old idea. All this shows is that the writers of the report have succumbed to the diversity lobby or suffer from a series of human biases, starting with confirmation bias – the confirmation that diversity will solve mathematical and ethical problems. Bias is a complex set of problems in both human affairs and AI – it needs sharp analysis, not Woolworthspick and mix team building. Theres one really puzzling sentence on this that sums their naivety up perfectly.The prejudices of the past must not be unwittingly built into automated systems, and such systems must be carefully designed from the beginning. Put aside the fact that this is largely what the House of Lords does for a living, it
is not even wrong. AI has 2300 years of mathematics behind it – from the first identified algorithm in Euclids Elements, through centuries of theory in logic, probability, statistics and other areas of mathematics. AI is built on the past.
LINK
Exploiting AI
The UK has an excellent track record of academic research in the field of artificial intelligence, but there is a long-standing issue with converting such research into commercially viable products. Damn right. They’re once again pained over the age-old problem the UK has on spending oodles of public money on world-class research, which doesn’t translate into commercial success. There is the usual error of equating AI SMEs with University start-ups. Actually, many have nothing to do with Universities. We need to support SMEs with business ideas. Yet where are the people like me, who put their own money and energy into starting an AI company and invest in others? Every AI academic in the land seems to have been consulted, along with many who wouldn’t know AI of they saw it in their soup. We know that our HE system is deeply anti-corporate. To assume that research equals success is a complete non sequitur. We need to encourage innovation AND commerce around AI – not just hose yet more money into Universities.
Usual suspects
Then there’s the usual tired old suspects. First, a Global Summit. Really? Nothing like a junket to advance our AI capability. Then a code of conduct. Yet another one? Politicians do love codes of conduct. Then there is the predictable call for a quango – creatively named the AI Council. Its all so unimaginative.
AI in education
But the worst section by far is the section on EDUCATION. There is a great deal of soul searching about AI in education but only in the sense of teachers and curricula about AI. The big win here is using AI to improve and accelerate teaching and learning. This is what happens when you only talk to teachers about AI. Its all about the curriculum and nothing about actual practice. This is a massive, wasted opportunity. Im selling an AI learning company to the US as I write this. Were already losing ground. Theres something called the Hall-Presenti review – whatever that is. Ive worked in AI in learning for years, run an AI company (WildFire), have invested in AI in learning companies, speak all over the world on the topic, write constantly on the topic – yet have no idea what this is. Thats the problem – Parliament is an echo-chamber. They dont really speak to the people who DO things.
Conclusion

To be fair theres some good stuff on healthcare and a few shells over the bow for defence and autonomous weapons, but it’s a bit tired, pious and lacks punch. It will, of course, fall stillborn from the press.

 Subscribe to RSS

Tuesday, April 10, 2018

The Fallacy of ‘Robot’ Teachers

I talk a lot on AI in learning but these days you can’t move for robot teacher articles and presentations, usually some diminutive piece of white plastic, sometimes, oddly, with a tablet stuck on its chest, that invariably responds with silence, something banal or falls over. This is seriously flawed thinking. I call it the ‘Robot Fallacy’, the idea that AI in learning is largely about physical robots. Fuelled by a century of cinema, where killer, and sometimes friendlier, robots dominate, due to the fact that it is a visual medium and needs ‘characters’ in drama, robots signify lazy thinking about AI. In practice, 99% of AI has nothing to do with robots. We are all enmeshed in AI, as AI is the new UI. Google, Facebook, Twitter, Netflix and most other online services are all mediated by AI, with not a robot in sight. Sure, Amazon uses them in its warehouses but this is a tiny portion of the process.
Robot teachers are, largely, as stupid an idea as robot drivers in self-driving cars, robot cleaners pushing a robot vacuum cleaner around the floor or a robot pilot sitting in the cockpit running autopilot on a plane. Auto pilot is a sophisticated piece of invisible software with secondary systems. The whole point of these self-driven systems is to ‘eliminate’ humans. Sure there’s a role for companion robots for people with severe learning difficulties or the very young, but on the whole the idea of a robot teacher is ridiculous. The point of this technology is to augment or disintermediate the physical teacher. Your automated banking is not a robot teller, it is online. Not a robot in sight.
The most ridiculous examples I know of AI in learning, are robot projects. They get tons of attention and grants. Doomed to succeed, they are usually a simple chatbot inside a big bit of plastic with barely moveable parts. Take Professor Hiroshi Ishiguro from Japan, whose robot self gives lectures, while he swans around conferences. To be fair his robot self looks more human than himself. This is bizarre and says more about the useless pedagogy of the lecture than any useful lessons in learning. My sense is that it’s a form of device fetish – education has disastrously focused on spending money on devices and not solutions to pedagogic problems. Tablets have been showered on schools in acts of folly. The robot thing is simply a another alluring device.
Robots in factories, that find, select and porter goods around factories make sense. Robots in manufacturing with their precision, speed and strength makes sense. Self-driving cars, make sense. Robot vehicles on Mars make sense. Robot teachers make no sense.
It is not just that AI has no significant cognition. AI is an ‘idiot savant’, incredibly good at specific, narrowly defined tasks but magnificently bad at generalist tasks – namely being a teacher. There is a huge amount of unwarranted hype around AI, not helped by the robotic presentation of robots as teachers, whereas in practice, AI can only be applied online to many specific parts of the learning journey. So far it is a story of augmentation not automation.
Find things out
That is not to say that AI has no role to play in learning. In fact, it will shape what we learn, why we learn ad how we learn. AI, in my opinion, will be the single most important technology to shape the learning landscape in the future. In many ways it already has. Google changed things for the better, a useful tool that heralded an irreversible pedagogic shift. Amazon revolutionized access to books, online and offline, as well as self-publishing. AI also shapes social media, as algorithms select personalized information on your timelines. It is a shame that the only form of AI you’re likely to see formally adopted by education is plagiarism checkers – but there you go – education can be a slow learner.
Online learning
That first wave of Google-led search and social media had had a profound influence on the learning landscape but the second wave is more significant, with AI-driven online content creation, curation, consolidation, adaption, personalization, retrieval and assessment. Tools now exist to do all of these using AI, and the efficacy is clear. Take one example, content creation. WildFire creates content in minutes not months, at a fraction of the cost of traditional online learning. Guided curation using AI is also possible. Spaced practice tools are now readily available and online assessment has benefited from online identification, face recognition, keyboard pattern checks and so on.
Learner interfaces
Another feature of this second generation AI is the shift in interfaces for learning. With NLP (Natural Language Processing) we also have text to speech (automated podcasts) and speech to text (speech recognition). This has opened up a switch from poor retention multiple-choice to open input, as well as spoken interaction. With WildFire, we have open input and interactive speech recognition for both navigation and interactive retrieval.
Chatbots
One recent advance has been in chatbots, which uses our natural propensity for dialogue to teach and learn. This return to a more Socratic approach to learning as been enabled by smart AI. These chatbots are used to find things, student support, deliver learning and mentor. Otto is a chatbot that sits above your content and find answers and learning opportunities for you as performance support, when you need it. We’ve developed an assessment chatbot that delivers questions on what you’ve learnt. Other chatbots provide help and support on courses. (10 uses of chatbots in learning)
Conclusion

So AI is present right across the learning journey. It can already deliver answers to questions, find things, create content, curate, allow natural language input and output, deliver personalized, adaptive learning as well as enable online assessment. Major learning services, such as Duolingo, are now delivering language learning to hundreds of millions of learners. With the introduction of software that learns (machine learning) we have software that get better the more you use it. Teachers have brains that are superb at general teaching but, bit by bit, aspects of teaching and learning practice will be automated. That has already happened. Every learner uses AI to search and find. Almost every learner uses AI-mediated social media. Teachers use it for CPD. AI at present augments teaching but a teacher is not replicable or scalable. If we want to solve the problem of increasing demand for learning we need to scale the process of teaching and learning. That has little or nothing to do with silly, teacher robots but everything to do with AI.

 Subscribe to RSS

Monday, April 02, 2018

AI-driven speech may revolutionise online learning

Over the last year or so we built the world’s first AI content creation service, WildFire, which creates online learning in minutes not months, at a much lower cost and with high retention. The reason for claiming it is high retention, is that we largely abandoned multiple-choice-questions for open-input, making the learner think, recall and actively input their thoughts. It is this ‘effortful’ learning that really matters in learning. This worked well and we have delivered online learning on factual knowledge, high-end academic content, processes, procedures and management content to a range of audiences in large organisations, from apprentices to high-end clinicians, in finance, healthcare, travel and manufacturing. Having seen how well open-input worked, we turned our attention to the use of AI to go several steps further and improve the interface. What if the learner could simply speak the answers? ... it was a revelation.
Speech
We learn to speak almost effortlessly, whereas, writing takes many years. So why not exploit what we do everyday in our lives - use speech input. It was thought that women spoke much more than men, a myth started in The Female Brain, by Louann Brizendine, who claimed that, whereas women spoke on average 20,000 words a day, it was only 6,000 for men. This proved to be nonsense. An actual study, at the University of Arizona by Mehl (2007), using an electronic recording device to sample everyday speech from 396 people found that we speak, on average, around 16,000 words a day, with no significant difference between men and women.
This is much greater than the average for writing. So it makes sense to use speech in learning. Consider also that if spelling is not part of the learning, you eliminate problems around misspelling, especially for those who are nervous on that score or who may have dyslexia. On top of this you do not have to make the physical effort to move a cursor around the screen into a field, then physically type.
Retention
However, it is interesting to compare different forms of input in terms of retention. So, you think of an answer, then:
   CHOOSE your answer from a list (multiple choice)
   TYPE your answer
   SPEAK your answer
CHOOSE (MCQs)
It is clear that simply clicking on an already provided answer from a list is the least effective of the three. The answer is there in front of your eyes, you have a 25% chance of getting it right without knowing anything, questions are often designed so that you can guess and the distractors are often remembered rather then the correct answers. It makes you wonder why the online learning industry is so wedded to MCQs.
TYPE
This has the advantage of making you recall the answer into your brain first (a powerful reinforcement event), then actively type in the answer, another reinforcement event, without having been given the answer or suffering from the drawbacks of simply choosing from a pre-written list. We have found this to be a much more powerful way of learning.
SPEAK
Things get interesting here, as you are communicating directly, without any of the artificiality of choosing from a list or typing. My initial impression (not based on any studies) is that this may be even better. Being hands free, your attention and cognitive focus is entirely on thinking and expressing your thoughts. None of your cognitive bandwidth is taken up by moving the cursor, typing and letter-by-letter spelling. You get a focus on meaning but there’s an additional advantage, as you get more of a flow and the learning is faster.
Podcasts
Using another form of AI, text to speech, we can also, automatically, create podcasts. This is built into the service. Simply tick a box and your online learning will create a podcast of the module or page by page speech. This is a useful supplement to the active learning.
Context
In addition, as we have audio only learning, including navigation, using the words, NEXT, BACK, GO and SCROLL, so we can place the learning experience within VR, which we have done, instantly and cheaply. We know that context helps retention, so speech input allows a further level of retention to be achieved. This is getting interesting in say, training fror healthcare professionals in a hospital or cabin crew inside an aircraft.
Conclusion
Simultaneously, using different forms of AI, we hope to have increased the efficacy of online learning by the:
1. Superfast creation of content
2. Higher retention open-input
3. Higher retention speech input
4. Automatically created podcasts
5. Full 3D VR delivery
All at lower costs and far greater speed than traditional and expensive methods. If you are interested we can show you all of this by Skype. Contact us here.
Bibliography

Mehl M (2007). Are Women Really More Talkative Than Men? Science ,Vol. 317, Issue 5834, pp. 82

 Subscribe to RSS

Monday, March 19, 2018

AI has and will change language learning forever

Just before the dawn of the internet I worked with the CEO of a major CD language learning company. His business model was fascinating, 
I don’t sell language learning, I sell the false promise…. My customers are ‘false starters’ mostly middle-class people who think they’ll learn a language in a few months before they go to Italy, Spain or France on their holidays… they never do.” He explained that the whole market was based on this model. The BBC packages at the time were the worst, he explained, “They’d send a film team to France for a month or so, come back and write a book around it…. it is literally impossible to learn a language from their materials”. I stayed out of that market. But times they are a changin’….
Internet
The technology moved on from CDs, along came the internet, and we saw the first big effect on learning languages – mainly English. The abundance of music, films, sport in all media on the web, allowed ready access to content, allowing contact, practice and immersion. Huge numbers have learnt languages without direct instruction. But learning a second language remains one of the most difficult things one can do in life and direct instruction still has a place. The problem with online instruction is that the technology was still too flat, text based and restricted to simple drill and practice. The content was too linear, often dull and struggled when it came to the spoken word, practice and immersion. Technology is now influencing not only what languages we learn but how and even why we learn languages. Some argue that the Anglo-saxon domination of the internet has accelerated the expansion of English as a global language. Machine translation raises the interesting possibility in that it may lead to less people learning new languages, if frictionless, real-time translation is available. But the most obvious and immediate impact will be on the practical teaching and learning of languages, where smart technology is already having a global impact.
AI
Of one thing we can be sure; AI brings a new paradigm to language learning. Natural Language Processing (NLP) has brought entity analysis, sentiment analysis, classification and machine translation. In addition we have text to speech and speech to text, now revolutionising interfaces. At the same time algorithmic techniques and machine learning brought adaptive, personalised and spaced learning. Even image recognition is being brought into identification and assessment. These technologies are being blended to produce sophisticated language learning and the possibility of learning a language without human instruction. One has to look across the whole learning journey to see how this is potentially possible.
Machine learning
To see how far AI has come in languages, Machine Translation is a good starting point . Google Translate can handle over 100 languages and us used over half a billion times a day. Launched in 2006 it used Statistical Machine Translation to match strings by probability against strings in another language, basically pattern matching. But in late 2016 it switched to Neural Machine Translation, making it much more successful and contextual. It is available as a browser extension and on Google Home and Google’s Pixel Buds. The ear ‘Buds’ can translate 40 languages in real time. To be fair, like Skype’s real time translation, it’s far from fluid and perfect but the direction of travel is clear – it will get better and better. 
Learning journey
So what about learning a language? Most successful language learning models take the learner on a learning journey from simple basics to practice then production. This progression normally starts with structural basics on the alphabet, vocabulary and grammar. Practice usually starts with limited and controlled practice and moves towards more open and free practice. Finally, there is generative production and use of the language. In addition to the actual learning there are also pedagogic issues such as motivation (a particular problem in language learning) and assessment. AI has a role to play across the whole of this learning journey.
Drill and practice
My first ever computer-based learning programme was teaching the Russian alphabet, which I built using the Commodore 64 graphic characters.  You saw a character and had to type in the corresponding English sound (as a letter or letters). I then programmed a behavioural drill and practice vocabulary programme. Randomisation was a feature, stratified with progress dependent on scores. This was typical of most early computer assisted language learning programmes.
Adaptive learning
Basic drill and practice is still a feature of most adaptive systems, such as Duolingo, with 200 million registered users, where structured topics are introduced, alongside basic grammar but adaptive algorithmic techniques track your progress and take into consideration, your forgetting curve, short-term success rate and effort. Adaptive systems can blend individual with aggregate data to optimise progress for the learner, depending on need. Every new learning event can be uniquely presented to that learner thus personalising the learning, an important form of optimisation in language learning, give the distribution of ability.
Spaced practice
Spaced practice, where the learners use retrieval techniques in a structured reinforcement pattern to push knowledge and skills from working to long term memory is a good starting point for the consolidation of acquired knowledge and skills. Anki is a free package that uses the algorithmic control of spaced practice to determine the learning path.
Chatbots
Controlled practice, to varying degrees, can also be delivered using chatbots. There are many species of chatbots from learning engagement, teaching, mentorbots, and practice bots. Chat has overtaken social media on mobiles and is clearly the preferred interface. We seem to have a natural affinity to chat interfaces and in some cases, with wellbeing bots, even the anonymity of the machine has been shown to be an advantage. They have been successfully used in educational and corporate training environments. They offer a dialogue interface, so are eminently suitable for language learning, with flexibility around the recognition of replies by the learner and, of course, speech. They have huge potential and when embodied in consumer, home devices can bring language learning into to the home.
Open practice
But active immersion is also now possible with home devices. You can switch your Amazon Echo to respond in German. Consumer technology, such as Alexa, Google Home and others will offer cheap, free and increasingly sophisticated language learning in your home. Ask it a question in English and it will reply in German. This is a bit like having a German person in your own home 24/7.
Immersion
The internet provides a wide and deep set of resources in most major languages. There’s an endless amount of content in your target language, in all media – text, audio and video - movies, box sets, music videos, Youtube, Wikipedia, whatever. Here other immersive technologies come into play, such as VR and AR. These are not AI technologies but AI techniques can be used within these environments to provide immersion, attention and context for language learning. In a current project (WildFire) we have successfully integrated speech input within VR, which not only allows you to navigate through the learning using just your voice but also input open response input and so on.
Assessment
Both Babbel and Doulingo offer paid English assessment testing. Face and digital recognition allow unique identification of candidates for assessment. Keyboard typing patterns can be recognised, along with adaptive assessment, which adapts to the candidate’s ability level, are all being used. Online assessment is now here, which increases accessibility and progress in language learning.
Conclusion
AI, with its rapid advances, specifically in technologies that aid language learning, may turn out to be the most significant technology in this field to date. The technology provides behind the scenes language processing that allows machine translation, speech recognition and many other services to be used across the learning journey to keep learners moving forward, optimising and personalising delivery. It has already accelerated the digitisation, disintermediation, decentralisation and democratisation of language learning.  Yet we must be careful in attributing too much efficacy to AI. Its translation ability is nowhere near as good as human translation, speech recognition still a bit ropey and with other services, such as chatbots you need to be a bit forgiving. Nevertheless, it is constantly improving and on current rates of progress, it seems likely that it will have a major impact in language learning.


 Subscribe to RSS

Thursday, March 15, 2018

Should you listen to music while studying? No... here's why

There are those who extoll the Mozart Effect, I know of one who extolled the virtues of playing Mozart to her children when they were very young and when they were learning. This, she claimed, had been proved scientifically to improve IQ and their ability to retain knowledge. Remarkably, she extended her claim to the foetus.
This baloney was sparked off by a paper in Nature by Rauscher, Shaw and Ky (1993), which showed a small improvement in spatial reasoning score (very specific), the effect lasted no longer than 15 minutes, then disappeared. The theory also disappeared, as several follow up studies could not replicate the effect. Rauscher herself, disclaimed the idea, saying that they had made no claim linking the playing of Mozart to intelligence. Chabris and Steele in a meta-studies paper in 1999 put the nail in the coffin by showing that such effects are merely the result of short-term and temporary ‘enjoyment arousal'.
But education can never resist a fad and there's always someone in education who can't let a bandwagon pass  in this case Don Campbell, who published The Mozart Effect (1997) and The Mozart Effect for Children. These books are, quite simply, bogus. His claims bear no resemblance to the actual research and, if you have this idea floating around in your brain, it’s largely down to him trade-marking the effect, then publishing these books, that were then taken up by lazy ill-informed journalists. This is how it ended up in the minds of so many parents and teachers. It was even funded and applied in some states in the US, notably Georgia and Florida.
Music in general
On the general proposition, that listening to music helps one learn, we have to be as equally careful. There is a large and complex literature on this subject, testing the effect of music on various cognitive phenomena and there is some evidence that it improves mood, even motivation, but one must be careful when it comes to actual learning.
In this interesting study, silence is used as a control, along with the two major components in popular music - music and lyrics. Perham and Currie (2014) created four groups:
Silence
Music without lyrics
Music with lyrics they liked
Music with lyrics they did not like
Results
The sample (30) was small, and I'd like to see this replicated with a larger group but the results were interesting:
Revising in silence was signifiantly better than revising while listening to music with yrics (liked or disliked)
'Silence' and 'music with no lyrics'
Revising to 'music without lyrics' was produced better scores than revising to 'music with lyrics'
Revising in 'silence' group could preict olearning outcomes better than other groups
Music in online learning
Moreno and Mayer (2000) tried e-learning with the following groups: 
Learning with music
Learning with sounds
Both
Neither

When retention and transfer were tested the groups with ‘music’ performed worse than those without music. This is a well known phenomenon where cogntitive overload inhibits learning.
Why?

It's to do with the overloading of working memory, especially with spoken words. One quick experiment you can do with your kids, or students, is to take a random page from a book on a subject they are unfamiliar with. Now tell them to read it in silence. Now choose another page and ask them to read it while repeating the word ‘boing-boing’ over and over. They will be unable to meaningfully learn from the text. The reason is the overloading of working memory, the phonological loop to be exact. Music takes up valuable bandwidth, therefore inhibits learning.
Conclusion

It may be devilishly difficult to convince your offspring that music is bad when they’re studying but when faced with a 60% differential it may be worth telling them about this study. There is lots of bad advice around study techniques that focus on superficial, low retention study methods and ignore attention, effort, retrieval and deliberate practice. No doubt some wag will tell us that music is good for those with an auditory learning style... that's also bullshit.

 Subscribe to RSS