THE idea of artificially intelligent personal assistants—digital secretaries, made of software, that take dictation, manage contacts, book lunches and arrange transport—has been around for many years. But truly successful examples have proved as elusive as flying cars. Perhaps the best known example is Siri, the personal assistant built into Apple’s iPhones and iPads, which can understand complex spoken sentences such as “Please reschedule tomorrow afternoon’s meeting from 4pm to 3pm”. As clever as this is, however, Siri still feels more like a technology demo than a really useful assistant.
The next generation of assistant software aims to go one step further by pursuing an approach known as “predictive intelligence”. It exploits the fact that smartphones now have access to fast internet links and location data, and can draw upon personal information, address books, e-mail and calendars. The aim of these new assistants is to anticipate what information users need, based on context and past behaviour, and to provide it before they have even asked for it. Such an assistant might, for example, spontaneously suggest that you leave early for a meeting, because it has spotted heavy traffic en route; present directions to your hotel when you arrive in a foreign country; offer to book a taxi or hotel based on analysis of an incoming e-mail or text message; or offer personalised suggestions for dinner in the evening.
Magnetic tape to the rescue
- Fatal distraction
- Made in China
- Sooty skies
- Particle physiology
- Clean that up
- Move over, Siri
- And the winners are…
- Thief in your pocket?
- Kill switches and safety catches
- Bacchus to the future
- Bitcoin under pressure
- The internet’s fifth man
“The phone has all of these data sources, and we are still the ones who pull out our phones and take the first step. Why does a human need to initiate this process?” asks Bill Ferrell, the founder of Osito, a predictive-intelligence application for the iPhone. Instead, software should reduce the friction involved in performing tasks on mobile devices and reduce the need for manual input. Today’s cumbersome interactions are, Mr Ferrell argues, an awkward transitionary stage in the evolution of mobile devices that predictive intelligence will see off. With the advent of wearable devices that support fewer input methods, like smart watches and Google Glass, there will be a need for less interaction and more automation.
Google, for its part, considers predictive intelligence to be the future of search. Its Google Now service is arguably the most advanced incarnation of predictive intelligence; installed in many millions of Android devices, it is certainly the most widespread. In recent years Google has improved its search engine through semantic analysis and, more recently, personalisation. Anticipating your need to search is a logical next step. It draws upon search history, stored personal data and past behaviour to provide “the right information at just the right time” in the form of cards that appear on a smartphone’s screen. Google Now’s algorithms can also take advantage of partnerships with third parties—including airlines and event organisers to access ticketing information. A new feature even allows Google Now-enabled devices to listen to the sound of your television, identify what you’re watching and provide programme information in real time.
There are many other contenders. The Android and Apple app stores are filled with personal assistants of various kinds. Some draw primarily on calendar data (Donna, Sunrise, Tempo) and others on e-mail or location (Osito, reQall). Apple, which looked somewhat behind the curve on predictive intelligence, bought Cue, a personalised-agenda app, in October. It included some predictive features in the latest update to its mobile-device operating system, iOS 7, such as estimated journey times to frequently visited locations.
As ever more personal information becomes centralised in smartphones and the accuracy of positioning technology increases, these assistants will be able to learn new tricks. But innovators must tread carefully. For one thing, predictive intelligence can be creepy. Some people may not like the idea that their phone is tracking what they are doing so closely. It’s also hard to get right: assistants will inevitably make mistaken predictions, leading to unwanted notifications. Many smartphone users are already struggling with “notification fatigue” as apps and online services compete for their attention.
“There’s a balance between coverage and precision,” says Baris Gultekin, one of the architects of Google Now. Predictive technologies rely on a technique called Bayesian inference to combine multiple signals and estimate the accuracy of a given prediction. Mr Gultekin says Google Now’s notifications have to pass a “high bar” of both likelihood and usefulness in order to be displayed. It is tempting to push information to users based on a clever deduction, says Mr Ferrell, but it’s better to be silent than wrong. Yet some users even complained that the first version of Osito did not pop up often enough. Who would have predicted that?
Powered by Facebook Comments