How would you like to create your own algorithm?
Google has created an algorithm that can write Wikipedia articles. And what did it use to train the algorithm? You guessed: Wikipedia articles. In the space of a few years, encyclopedias have gone from monumental words requiring the input of hundreds or even thousands of experts on topics of all kinds to the point where they can now be written by algorithms capable of summarizing and documenting human knowledge with links.
Now we know a robot can be trusted to produce an article for Wikipedia, we might usefully ask how many repetitive or boring tasks we perform each day couldn’t simply be handed over to algorithms: one reasons is because some tasks are not as easy to automate as they might seem. For example: first thing every morning I check my Feedly feeds to see what’s happened in the world overnight. I then draw up a shortlist of the most interesting pieces and then do a second sweep to choose one or two of the best on different topics, posting them on my Pinterest board and my Flipboard magazine. Later I use these two tools to retrieve information, as do other people, who mainly use them as sort of a press summary curated by me. What are my criteria for deciding which are the best stories? The truth is I don’t know with any certainty, which makes me wonder whether I could educate an algorithm to do the same job by analyzing all the stories I have selected over time? In theory, yes, it must be possible.
But for the moment, I prefer to continue doing it myself: I find it relaxing, it offers me certain possibilities and creative freedoms and also bolsters my knowledge of certain topics. That said, adding the numerous links to my articles can be boring, because I make an effort to provide as many as possible to my sources (I am a teacher, after all, and so when I write, what I am also doing is preparing materials for my classes). Given that what I link generally comes from articles in my files or from other sources I use to document my writing, could an algorithm educated with the articles I have written for many years learn my linking criteria, when to use a link to Wikipedia to document a term, when to draw on an article from my archives, etc., and also to help me write? And once written, I share my articles on the social networks, a repetitive task the relevant tools don’t seem to do very well, so I end up doing it myself by copying and pasting. Could an algorithm learn my criteria for copying and pasting headlines and the link to my articles, and take over the task? Could an algorithm review the articles I have written over time, monitor the links to contents that have disappeared or are not accessible, and replace them with valid links, if they exist?
Again, I’m pretty sure the answer to all these questions is affirmative. After all, these are very specific tasks that very few people would do in the same way as I do, with the same combination of needs and motivations. In which case, when will we have algorithms that learn from relatively repetitive tasks — but that are not completely mechanical — and that we could use as personal assistants? How close are we to the point where somebody with no expertise in machine learning could define a task, provide the data for learning, and then monitor the progress of an algorithm carrying out some everyday task? I think we are pretty close to the point where after we have carried out a task a sufficient number of times to allow an algorithm to understand it and the criteria we apply, and that it would then repeat. How many daily tasks would you like to automate and how much would it improve your life and your work?
(En español, aquí)