Machines can learn, but can they learn to teach?
Machine Learning, simply put, is the process of creating a well-designed algorithm and feeding enough data to this algorithm, with the objective of finding patterns, improving the data and program itself, without the help of a human. The functions can hence be personalised for individual users. The machine 'learns' to behave in a certain way we want it to. Smart, right?
Machine Learning has been the buzzword for a while among mathematicians, software engineers, scientists, Artificial Intelligence (AI), enthusiasts and Steven Hawking.
Mr. Hawking, however, warns us in a 2014 paper along with his colleagues: “It’s tempting to dismiss the notion of highly intelligent machines as mere science fiction, but this would be a mistake, and potentially our worst mistake in history.”
Indeed, Microsoft's experimental 2016 AI Twitter chatbot Tay is an example of the risks machine learning can pose – the bot, posting from the handle @TayandYou, retweeted all sorts of misogynistic and racist comments within a span of 24 hours.
When Steven Hawking expresses his concerns regarding science and technology, we listen. Of course, the paper did not ignore the possible benefits of developing artificial intelligence: “They are huge; everything that civilisation has to offer is a product of human intelligence; we cannot predict what we might achieve when this intelligence is magnified by the tools that AI may provide, but the eradication of war, disease, and poverty would be high on anyone’s list. Success in creating AI would be the biggest event in human history.”
Today, companies such as Google invest so heavily in Artificial Intelligence research and implementation that computer scientists have now designed an Al to facilitate other AIs. Google's advanced Machine Learning algorithm – Google Vizier – was recently tested by asking to bake cookies. That's right, cookies. Their reasoning was that algorithms are no different from recipes. Then, Adobe released Photoshop Elements with an intriguing feature that can fix that otherwise perfect photograph in which a friend has her eyes closed. The feature is simply called 'Open Closed Eye'.
Machine Learning and Artificial Intelligence are often used interchangeably. But they are far from being the same. According to Google, AI is “the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages”.
Meanwhile, Machine Learning is a sensitive analysis of data by computer systems to extract meaningful information. A path (one among many) to Artificial Intelligence.
Once a machine can think like a human, there is a high possibility that personalisation of learning material and education will be changed and drastically improved. We all love it when the chef at a restaurant tweaks the recipe to our taste buds. Adapting content with technology is as satisfying as that. Add personalised, quality content to already accessible technology, and education reaches new heights, as it pleases every learner individually.
It has not been very long since computers have learned to read and comprehend the nuances of human language. This task is of great importance for computers as there is so much data that goes unclassified, and computers haven't yet learned to accurately find patterns. But a researcher from Stamford was recently able to design a Machine Learning algorithm that is only about 5% less in agreement with humans. “This is now approaching human levels of ability to understand the sentiment of natural simulation,” says Jeremy Howard, data strategist at the University of South Florida.
Can E-learning follow this path?
Websites like Coursera, Udacity, TedEd, Codecademy, Open Culture – the list is endless – offer online courses under a variety of thematic umbrellas: Art, Technology, Life Skills and more. These online learning platforms offer a range of topics from Quantum Physics to the physics of superhero movies to tips on getting dressed. We can access these courses anywhere, at any time.
E-learning has become a popular medium for making knowledge accessible to anyone with an Internet connection. Sometimes professionals give you lessons through scheduled videos; at others, text, homework, and evaluation are involved.
Any e-learning syllabus or course goes through a process called the ‘ADDIE’ model, which includes Analysis, Design, Development, Implementation, and Evaluation.
The Analysis stage defines the module of the course based on the objective, target audience and the timeline for completion.
The Design stage is where everything that is pedagogical – a logical flow of the topic, timelines, evaluations and assessment instruments – is taken care of.
In the Development stage, software such as Adobe Captivate, and Articulate Storyline are used to assemble the course content via programming, done by instructional designers and developers to make a story out of a lesson.
The Implementation phase is when the development of procedures for facilitators and learners happen. Training facilitators cover the course syllabus, learning objectives, methodology and testing procedures. The learning curve also includes getting comfortable with new tools (software or hardware) and, finally, student registration. This stage includes evaluation of the design.
The Evaluation step is the final one consisting of two aspects: formative and summative. Formative evaluation is present at every stage of the ADDIE process, whereas summative evaluation is conducted on the finished instructional programs or products.
This meticulous process and e-learning model is justifiably long to be accurate. But, consider allowing a machine to skim through this process with accuracy to deliver personalised online courses. Can Machine Learning be used to develop new content? Will this reduce costs, maintain or improve quality – and, the holy grail, save time?
What is Classification, and does it reduce costs?
'Classification' in Machine Learning lingo refers to finding patterns in data sets. E.g., grouping images as with faces and without faces, or image with animals in them. Most of us have experienced this on various social media platforms and in imaging software. Today, most social platforms recognise people and suggest naming tags based on our list of friends and followers.
The improvements made in classification processes have played a major role in identifying that Machine Learning as an accurate recognition of data has multiple applications: Image categorisation, translation, and caption generation. When classification is automated, its manual, time-consuming, yet indispensable alternative is rendered cost-effective.
An example from an article on elearningindustry.com: “While subjects like Physics and Optometry are completely different subjects, there are concepts within the material that would overlap – light for example. Sifting through all the content in each subject to identify related concepts would take many man hours and subject knowledge. Yet, with deep learning, machines have the potential to perform that task quickly and efficiently.”
Deep learning is another path to Artificial Intelligence and is a promising one to facilitate classification and instructional design.
While we see promising possibilities for – and determined research in – Machine Learning for what seems like the present and the near future, we also witness mind-boggling loss that some of the most intelligent human minds have faced. Google's loss after acquiring DeepMind is a shocking figure. Yet, the search goes on.
Therefore, the answer to the question is: we do not know, not yet. Machine Learning and Classification (where the former leads to the latter), when applied to AI in e-learning, can reduce content development cost and save time. But what does the algorithm cost?
Automation, by definition, makes processes faster than manual operation. Throw in Machine Learning and Artificial Intelligence into the e-learning mix, and you get a heady outcome of a sped-up ADDIE process of developing content – and, thus, save time.
While the human mind is pushing its limits every day, is it prepared to handle the potential risks and pitfalls of Machine Learning and Artificial Intelligence? We are clearly smart enough to achieve what was previously unimaginable, but is there a possibility of being outsmarted? Are we willing to learn from a machine that's supposedly has a 'mind' of its own, a machine that can 'think' for us? Utopia would be, being able to carefully reap all benefits and strategically avoid the risks.
Steven Hawking and his colleagues ask: “If the short-term impact of AI depends on who controls it, the long-term impact depends on whether it can be controlled at all?”.
Maleva Robert | Blogger
[Cover image credit: Kirillm | Getty Images]