Computers

MIT's "better Siri" helps you get there on time

MIT's "better Siri" helps you get there on time
MIT researchers are developing algorithms that factor risk into planning to help get users to where they want to be on time (Photo: Shutterstock)
MIT researchers are developing algorithms that factor risk into planning to help get users to where they want to be on time (Photo: Shutterstock)
View 1 Image
MIT researchers are developing algorithms that factor risk into planning to help get users to where they want to be on time (Photo: Shutterstock)
1/1
MIT researchers are developing algorithms that factor risk into planning to help get users to where they want to be on time (Photo: Shutterstock)

Researchers at MIT are building a sophisticated algorithm to help with time-sensitive planning, estimating your chances of success and even suggesting alternate approaches that are more likely to succeed. The software, described by its creators as "a better Siri," could help plan projects on all scales, from long drives to air travel to multi-billion dollar NASA missions.

Virtual assistants in our phones and tablets can use maps and navigation tools to estimate the time it’ll take you to get from A to B, but they can’t calculate accurate error margins on their predictions. That is to say, how easily a three-hour trip might end up taking four or five hours should you face delays, heavy traffic or any other inconvenience along the way.

The algorithms being worked on by Brian Williams and colleagues at MIT address this limitation by associating a confidence margin to each answer. For every user-specified scenario, whether a car trip or any complex time-sensitive project involving a large number of uncertain factors, their software will be able to estimate the probability that you’ll complete the task within the deadline you have specified.

Then, if the software thinks that your current plan is likely to fail, it will use a weighted graph (a mathematical construct common in computer science) to suggest the smallest possible change of plans that would make sure you meet your deadline with a very high likelihood.

For instance, you could specify you want to do the long drive to an important meeting by setting off at a certain time, picking up your friend two hours in, stopping together for lunch at your favorite restaurant, and still get there on time. The algorithm (which you could access on your phone) might answer back that you only have a fifty-fifty chance of succeeding as is, but that if you are willing to leave twenty minutes earlier or pick a different restaurant your chances will climb to 99 percent.

William's software can do this by combining a large number of complex factors, such as the probability distributions of travel time on a mile-by-mile basis for the entire length of the trip. This computation would be very taxing even for today’s most capable processors, but Williams and team have worked out a way to simplify things by managing the leeway.

"We added this idea of risk allocation," says team member and graduate student Cheng Fang. "We say, 'What’s your budget of risk for this entire mission? Let’s divide that up and use it as a resource.'"

Along with the constraints, in fact, planners can specify the amount of failure that they are willing to accept. In the example above, you could say that you don’t mind being up to five minutes late or eating a little faster if you absolutely have to. Building some slack into the system in this way makes things a lot easier to calculate.

Any scheduling problem  –  not just trip planning  –  could benefit from this system, and in fact both NASA and the Woods Hole Oceanographic Institution are already using a similar approach to plan their complicated multi-year missions and make sure their deadlines are met. Transportation companies could also use this software to schedule departures in such a way that trains and buses are highly likely to arrive on time consistently.

And of course, personal planning would be no longer "left to chance."

The research is being presented this month at the annual meeting of the Association for the Advancement of Artificial Intelligence (AAAI) and is described in two open-access papers [1, 2] available through the MIT website.

Source: MIT

2 comments
2 comments
Paul Anthony
I'd be happy to test it. I do multiple pick up and stops, then drop offs and I try to stay on schedule throughout the day. If you count getting to a destination early as a negative, then unless you have a crystal ball your not going to get 99 percent, not ever.
Bob Flint
Try just getting phones to find your location indoors is enough of a problem, without a GPS signal, good luck with that. Once your outside the transit time can be computed but only to the door, even accounting for traffic, but not a slow elevator or broken escalator, etc...