Duolingo uses “spaced repetition” to improve long-time learning. This means that rather than drill a word a number of times and then assume we have learned it, the program presents the word occasionally so that we don’t forget it. The ideal time to show the word again is just before we forget it, and there are a number of spaced repetition software (SRS) packages available for money or for free, which are most commonly used for vocabulary (although they can be used to learn a great number of things). They use various ways to measure how often to show you a particular phrase. The most obvious is that if you get it wrong, it has been too long. Ideally you should have to stop and think, so if you can answer without thinking, it was too early. Programs such as Supermemo, Mnemosyne and Anki ask you to rate how hard it was to remember, and adjust the schedule individually for each word/phrase/fact.
Duolingo boasts a similar ability to keep track of weak points and bring them up more frequently. How does that work? Not too well. The program tends to wait too long before the first repetition: Unless I manually select that particular topic later in the day or the next day, it can take days before it comes up, at which point I won’t remember it unless I happened to associate it with something memorable. Normally you need to repeat things within minutes first, then perhaps an hour (OK, it seems unreasonable for the program to wake you up after an hour) and then the next day, before you start on a long-term schedule. Duolingo is rather more optimistic.
Generally I don’t find any signs that Duolingo actually keeps track of my weak words or features. Perhaps it is just hiding this really well, but getting a word or phrase right every time does not seem to keep it from recurring. It seems more likely that Duolingo uses an average of the experience of many users: I find for instance that it tends to bring up longish less common words like kaplumbağa (turtle) rather than shorter, more common words like inek (cow), even though I actually learned the longer word almost instantly (because of a game I play) and actually fail the shorter word because it shows up so rarely. Despite this, the text has been crawling with turtles almost from the start. ^_^;
So there is definitely room for a competitor to disrupt Duolingo with a superior artificial intelligence. Hopefully that competitor is Duolingo’s own shadow research team, which is supposedly working to do just that. It would be a shame to not make use of the work that has already been put into Duolingo, even though it is still a work in progress and far from perfect. (Then again, neither am I…) Either way, the selling point of Duolingo is not really its technical excellence, but the fact that it is free so anyone can get started with it, and fun so anyone can continue with it.