What is AI?
How smarter digital tools can improve learner outcomes
From how we find our way around a foreign city to how we book a taxi, we already live in a world made simpler by artificially intelligent processes.
Google Search and the PageRank algorithm
Google’s PageRank algorithm, which determines where a page should appear in a Google search, is based on how many other pages link to the page in question – with every web page on the Internet potentially contributing to the ranking of every other page. PageRank is an iterative algorithm - it repeats the calculations many times, getting ever closer to a true value, until the estimate doesn’t change significantly. This estimate is then used as the page’s PageRank.
Face detection on digital cameras and smartphones
There are many different algorithms for detecting faces in photographic images (this is a very busy area of current research). A typical face detection algorithm used in a smartphone scans the whole image section by section, at different scales, firstly looking for edges, then comparing detected edges with geometric models of the human face. All ‘potential faces’ in the image are then checked for overlaps, with two or three overlapping potential faces required to confirm an actual face, all of which gives the camera a target area for fixing focus.
Weather prediction ‘sliding window’ algorithm
Predicting the weather is well known to require huge amounts of computing power to analyze the massive amounts of data from atmospheric measuring devices around the world. It also uses a complex array of algorithms. One simple example algorithm uses weather variations in previous years. While the probability that the weather on any particular day will match that on the same day in the previous year is very small, the probability that it will match a day within the span of the adjacent fortnight in the previous year is actually quite high. So, a ‘sliding window’ algorithm contributes to weather predictions by looking at a two week window of data from the previous year.
Converting speech to text, either for transcription or for further semantic processing (in order to ‘understand’ what has been said), is another area that requires a range of approaches and algorithms. Many speech recognition systems use ‘Hidden Markov Models’, statistical models that break down the speech, the acoustic signal, into short pieces. The Viterbi algorithm then treats these pieces as an ‘observed’ sequence of events, while a string of text is assumed to be the ‘hidden cause’ of those events, allowing the algorithm to find the ‘most likely’ string of text that matches the speech.
The basics of how AI works
Artificial Intelligence can help to make the humanly impossible possible, but what are the main principles that make this a reality?
Big data: AI can amass huge amounts of data and analyse it very quickly - something which is not possible for a human to do, no matter how clever they are!
Machine learning: like people, AI (the machine) can learn through its own experience - so the more you use it, the better it gets.
Algorithms: algorithms are the backbone of AI because they are a set of instructions for the machine to follow.
There are many ways these are combined to create ‘intelligence’. One example is using bayesian networks, which collect data to make predictions (i.e. about what you might like to buy in an online shop, considering your past purchases and the season). The more it makes these predictions, the more accurate the predictions get, as it gathers more data and teaches itself to be more accurate.
In a classroom this could be used to predict student achievement.
A bayesian network could ask, “is the student confused or interested”, then ask “did the student answer the previous question correctly or not”, and give a predicted score based on this information. The teacher could then use this to help them decide the next most appropriate learning activity for that individual student.
Applying this to education
What are the potential applications in education (AIEd) and how will it benefit teachers and learners?
Open learner model
By opening up data gathered to learners and teachers, we can help them to see what they are learning and where they are struggling.
Systems that can personalise learning for students, by making recommendations to the student based on their past and current performance, for example if the student is struggling, the AI might recommend a new piece of supporting content to give them extra assistance.
Mapping the learner process
AI techniques to help us analyse learners’ behaviour, and teachers can use that to adjust their teaching styles and help the students how need help the most.
More advanced assessments
At the moment computers are good at telling you if a student has got the answer correct or not, but with machine learning and more complicated algorithms, AI can help us construct new meanings from a collection of existing data.
- AI can let the teacher know the probability that their individual students have actually understood and mastered the topic, rather than a test score.
- There’s the potential to assess more intangible ‘21st century skills’, at scale.
- Through game like simulations, students can be assessed without feeling like they’re being tested, for example, Pearson’s Alice in Arealand game.
“For me technology is my assistant, I am the expert, I am the one that can still guide the student in the right direction, but I am using technology as a tool to assist me so I can do my job even better.”
Mia Beyleveld, Lecturer at Midrand Graduate Institute in South Africa, What’s in store for the Future of Assessment? YouTube
For more on the benefits and potential benefits of AI in education, read Intelligence Unleashed: An argument for AI in education, a new report from Pearson’s Open Ideas team and the UCL Knowledge Lab.