Tackling Traffic?

Traffic is one of the worst things you can encounter as a driver. It slows down your day, makes you late for meetings and provides aggravation that is never needed. With the improvement of mobile computing and applications, there have become many applications that will take traffic into account when mapping a route. WAZE and Google Maps, now both owned by Google, are two of the more prominent applications.
Both of these applications show where heavy traffic is located on your commute and will try and reroute you around it if it will greatly slow you down. WAZE puts an emphasis on collaboration on the roads and encourages its users to provide real time feedback of the current road conditions they are driving on. Both applications are used by millions each and every day. They have changed the ways Americans now travel and hopefully help save a lot of precious minutes for travelers.

What methods do these applications use?
Currently, WAZE and Google Maps applications both allow for users to predict traffic at a particular date or time. This allows for users to be able to plan when to leave and how much time they will need to a lot for their travels. The real questions from these features is: How does the application know what the traffic will be like in the future? Can it time travel? Is it just making a guess? The answer to all of these questions is the use Neural Nets.
Neural Nets are an artificial intelligence tool that allows for the algorithm to learn from previous data. This means that WAZE and Google Maps applications takes the data from previous days and times and then makes predictions on the time or date that the traveler needs. The algorithm will become more and more accurate with the analysis of many days and traffic conditions. The WAZE and Google Maps applications, as stated earlier, also allow for real time driver analysis. Google Maps will look at car speeds and WAZE allows passengers of the car to input traffic data into the application. Therefore, data is being fed to these applications every single second. With immense training, the algorithms have gotten to where they are today with the capabilities to almost accurately predict traffic flows.

Are there other methods that haven’t been as heavily explored?
While both Google Maps and WAZE are extremely successful with their neural net methods, there is another method, Q-Learning, that can be just as powerful. Q-Learning is a reinforcement learning algorithm. The Q-Learning agent learns rules from data that is fed through it and through each iteration will update the table to reach a set of optimal outcomes.
In Artificial Intelligence & Cognitive Science (CSCI 379) at Bucknell University, we spent some time trying to help find light traffic routes in the city of Chicago with a Q-Learning Algorithm. We first implemented an environment for the agent to learn in. This was a weighted graph that simulated the streets of Chicago, their direction and traffic patterns. The Q-Learning agent then worked to solve for the path of least traffic and therefore help predict routes that would be the quickest to navigate the city. The agent was able to predict these routes of least resistance which is similar to those of WAZE and Google Maps.
But, what are the ethical implications of all of these algorithms?
With any modern technology, there are societal implications from the implementation of these algorithms. One of the major issues of the applications is the ability to route or not route travelers through particular neighborhoods. The algorithm may make sure the travelers are avoiding “unsafe” neighborhoods or areas that do not want a lot of traffic. The question then becomes: “Who decides which areas to avoid and not avoid?” and biases come into the equation (literally). By avoiding or not avoiding these areas, there can be both economic and social repercussions. The biases of the algorithm can have effects that reach further than just extra time in a car.
Another implication is the use of marketing in these applications. They now allow for companies to market in the applications and will provide detours to particular companies on the route for fast food and different services. This then brings up the issue of which companies are allowed to market their on-route services, how far off route should you be brought by the application for a marketed service and how much time will be factored into the drive? This will then give prominence to companies who have the means to market on these applications and are located on or near less trafficked areas. The whole marketing world can be changed by the outcomes of these algorithms and is that something we are ready for?

So, what next?
Traffic is always going to be a part of American road travel. Both WAZE and Google Maps have changed the way we handle the traffic and have helped millions of people reach their intended destinations with less stress. With that said, we must be open to other options of Q-Learning for traffic prediction, in parallel with the current use of neural nets. We must also understand that these algorithms provide much more than extra time to grab a coffee. They provide social economical and operational layers to the American commute.
Bibliography
- Bham, Ghulam H., and Rahim F. Benekohal. A High Fidelity Traffic Simulation Model Based on Cellular Automata and Car-Following Concepts. 12 Vol. , 2004. Web.
- Choi, Samuel PM, and Dit-Yan Yeung. “Predictive Q-Routing: A Memory-Based Reinforcement Learning Approach to Adaptive Traffic Control”. Advances in Neural Information Processing Systems. Web.
- Gordon, Eric. “MAPPING DIGITAL NETWORKS from Cyberspace to Google.” Information, Communication & Society 10.6 (2007): 885–901. Web.
- Watkins, Christopher JCH, and Peter Dayan. “Q-Learning.” Machine Learning 8.3–4 (1992): 279–92. Web.
- Strom, Timothy. “Space, Cyberspace and Interface: The Trouble with Google Maps.” M/C Journal [Online], 14.3 (2011): n. pag. Web. 4 Nov. 2019