The Sirens are among the most famous creatures in Greek mythology. With their beautiful voices, they enchant sailors nearing their shore. Luring them astray to their island and let the ships shipwreck on the reef of the island. The story about the Sirens could be used as a metaphor about the media and news coverage of computerization in the workforce. The media use their beautiful voices to inform the public. They however - just like the Sirens - can lead people astray with the information they give. The stories they write are sometimes a bit of doom and gloom.
Going beyond opinions and estimations
Many discussions concerning robots replacing humans in the future as employees are often on an abstract level. Scientists estimate the probability of jobs being computerized and futurists envision the world of tomorrow. There's however no magic powder dust. There will need to be some serious amount of effort to create robots performing such difficult tasks. So let's go beyond opinions and estimations. What do these scary monsters actually look like?
When it comes down to artificial tutors helping or even replacing teachers, where do we even need to start? We're going to focus on one domain: chess. Chess has been at the forefront of Artificial Intelligence for many decades with other board games like Go. When I asked Tal Shaked, both a chess grandmaster and a machine learning expert at Google, which domains he believes will be on the forefront of machine learning and education, he believes playing guitar and chess are low hanging fruit. It's not that human society is inherently motivated in unraveling the secrets of the game of chess or playing guitar. Chess and guitar playing are however ideal domains for innovation in Artificial Intelligence. They can use this knowledge in chess and guitar playing to support endeavors which are of great value to human society. Think about programs to teach topics like math and reading.
If we would assume chess is indeed a frontrunner in education and Artificial Intelligence and could be of help to innovate in other fields, what is currently going on and what could be learned from this domain?
Back to the future: big data
For a robot teacher, it's important to have data about the performance of students in the domain they want to improve. They can look for patterns and offer a road to success based on learning profiles.
When the first computers where built, chess was one of the first domains to utilize the possibilities of computers and analytics. The German company Chessbase started to digitally collect millions of games from players since 1987. Former world champion Gary Kasparov and many other grandmasters directly saw the potential. Grandmasters had digital access to games of their opponent in one central database. Players used Chessbase to prepare against their opponents. They got help from computers to look for the strengths and weaknesses of both themselves and opponents. Computers made it possible to systematically and almost scientifically approach the game of chess. Later on, with the help of electronic chess boards, the entire logistics of logging, collecting, storing and distributing games all over the world in one central database was automated.
Ten years later, in 1997, IBM's Deep Blue defeated Kasparov. A hallmark for machine learning. Machines were not only able to defeat humans in a cognitively challenging task, but computers were also able to accurately rate the quality of every decision players ever made in their games. This tool offered unprecedented possibilities to improve analytics in chess. Chessplayers could look at their own historical development, compare their development to other players or efficiently scan for new ideas with computers.
Most sports are still at the start of even collecting such vast amounts of data of players on such a level. To compare chess with popular sports like tennis or soccer. They would need to film millions of soccer or tennis matches in the world (including youth games) over a few decades, store these matches in one central 'Soccerbase' or 'Tennisbase', analyze every decision each player made and have machine learning tools to automatically look for didactical patterns to successfully improve their game. Even if these sports wanted, technology, the current infrastructure of collecting and storing such data and the complexity of these sports makes this impossible.
Becoming a robot teacher
One of the challenges till this day in chess is the ability for computers to not only tell humans which brilliancies and mistakes they make, but also automatically explain WHY a certain move is good or bad. Artificial Intelligence lacks didactical skills. Exactly the skill a robot teacher will need to grasp. Didactical skills for AI are inevitable if they want to further enhance or even surpass the tasks teachers are doing. Robots will need to explain what their students are doing right or wrong and why this is the case.
It took a while to crack this nut, but in chess this is already possible. Not on the same level as a great chess teacher, but impressive steps are being made. ChessZebra created a program to automatically spot advanced didactical motifs and more in chess. To give a few examples:
The interesting thing about the possibility to spot these motifs, the program spotted all these motifs them self. The program did it all in a fraction of a second and is able to collect these key moments in billions of games. Double attacks, twofold attacks, profitable exchange, discovered attacks, removing the guard, x-ray, pins, preparatory moves and much more advanced didactical concepts in chess are recognized.
Ok, you might think. Good for them! Hooray! But why is this so valuable? Well, if an AI is able to automatically spot didactical motifs in chess on large scale, this offers scaling opportunities for personal analysis of players. At the same time, AI-based learning profiles based on own data can be created. Robots could recommend chess lessons based on the data. Or even better yet. Both coaches and robots can work together to create the best training programs.
To explain these accomplishments in chess in a more common language. Imagine Microsoft Word would have advanced teaching methods. We all know Word underlines misspelled words in a sentence: the spelling checker. A great function which helps us all to improve the quality of our letters.
But imagine if all letters you ever wrote in your life are collected in one central database. Each letter is automatically analyzed. But this time the analysis goes much further. Lexical, semantic, syntactic, logical errors, style errors and more. Every mistake you ever made is automatically categorized. Hundred thousands of learning moments in your writing will be analyzed. The program would automatically recognize nouns, verbs, adverbs, adjectives and much more. Every learning moment is categorized because the program would recognize simple or more advanced errors were made in your letters.
A historical analysis of your improvement in writing would be possible with this performance analysis tool. A personal training program would be possible based on the patterns found in the analysis of all your letters. Learning profiles can be created. You could compare the quality of your letters to classmates, friends or great writers. And this would only be the start of some of the possibilities.
Writing assistant programs like Grammarly and WhiteSmoke are making impressive steps and offer suggestions to their customers based on the errors they find. These improvements are however quite basic, if you compare it to the depth and complexity of great writing. The didactical value is at the same time limited. It's an assistant and not an analytics program to offer personalized training programs in writing. Customers will just correct their mistakes based on the assistant, but do customers understand their mistakes? Do their customers understand why a sentence needs restructuring beyond obvious spelling errors and how to improve their writing next time? In writing this isn't yet possible as far as I know. However in chess it is.
These feats by AI in chess might be scary for teachers and could justify the doom and gloom the 'Sirens' are advocating about in the media. Teachers might ask themselves: if computers are already able to perform such tasks in chess, which domains are next? What does the future even hold?
Don't let yourself however lead astray too fast. In the next article, we're going to look at concrete possibilities of having these machine learning capabilities. That even AI have limits and is not able to figuratively speaking transform lead into gold like the alchemists wanted to achieve in the past.