Greengard, S. (2017). Gaming machine learning. Communications of the ACM, 60(12), 14-16.

From Digital Culture & Society

Article Review- Machine Learning & AI – Joshua Erdmann

Modern Transportation and more specifically self-driving autonomous vehicles have been an important topic in the past decade. More and more researchers are looking into the possibility of self-driving cars and trying to make a dream a reality. Looking at the article “Gaming Machine Learning” written by Samuel Greengard, he talks about a new path researchers are using to try and build a more functional and inexpensive way to test self-driving cars. According to this article there has been massive advances in both sensors and artificial intelligence (AI) that has helped research towards autonomous vehicles. One major technological piece that is missing in their research is a lack of algorithms that can accurately identify objects, movements, and road conditions. This has lead to researchers looking towards the use of videogame simulations and machine learning to build better algorithms leading to smarter vehicles. One of the key issues with this topic is that humans are able to analyze a traffic situation and adapt quickly, but autonomous vehicles aren’t able to detect road conditions and accurately adapt. Things like a stop sign covered by graffiti, a worn down lane marking or snow-covering road markings are just some potential situations that can throw an autonomous vehicle off track.(Greengard, 2017)What that means is the main issue for researchers today in this particular topic is the fact that they need to find a way where vehicles will be able to read dangerous road situations more accurately.

Due to this main issue, researchers have looked towards the use of computer games and computer simulations to help aid their research. These computer games include games like “TORCS”, which is an open source, VR (Virtual Reality) racing simulator as well as more popular games offered to the public such as Grand Theft Auto V. Grand Theft Auto V, is a game that has previously already been credited as revolutionizing the way researchers develop autonomous vehicles, robots, drones and other machine systems. (Greengard, 2017) Benefits of these computer games for the creation of autonomous vehicles is endless but one of the key things researchers have been using these games for is to replicate real life situations online. A quote from Artur Filipowicz, who is a recent graduate in operations research and financial engineering at Princeton University, stated that “ These games offer ex- tremely rich environments that allow you to drive through a broad range of road conditions that would be difficult to duplicate in the physical world,”. (Greengard, 2017) This quote is important because it illustrates the changes that are going on within the industry, which is moving away from real life simulations and more towards computer simulated data and AI.

According to the article, research into the idea of using video game simulations and AI has been around for at least a decade but recently has resurfaced and came into the forefront of innovation due to the increasing issue of transportation in a modern urban environment. One of the ways this technology works is by combining a vast number of images of things like stop signs, traffic signals and road markings and comparing/documenting on the actions and reactions of people while driving. These actions and reactions that are collected include things like steering motions, breaking and the rate of acceleration. (Greengard, 2017) When information on these aspects of driving is collected, they can be used to build a better picture as to what changes need to be altered to an algorithm to create a better preforming self driving car. These computer games allow researchers to study driving in difficult situations with any measurement they decide and with any road condition they need to look at. For an example a research would be able to upload any particular car specifications, weather conditions and what road the car is driving on then gather information as to how the car reacts when it is faced that that particular challenge.

Although the benefits of this form of research for autonomous vehicles is definitely important, for the idea to move forward there are definitely some challenges and weaknesses that these researchers face with this type of research. For an example the act of transforming pixels and RGB values into useful information so that the vehicles can understand what’s can be happening challenging. (Greengard, 2017) This is because although similar, virtual worlds and the physical world don’t always match up. What this means is that information collected for a particular algorithm that worked in the virtual would, could be ineffective when applied to a real life situation. What this means is that although this data is important and revolutionizing the way they go about developing these vehicles; researchers have to keep in mind that it is necessary that results are reviewed carefully and that information used by artificial intelligence is reviewed extensively before put into use by the public.

Probably the most significant contribution in the area of study this article offers would be in the solutions they offer to the autonomous vehicle industry. According to the article the software required to create a game like Grand Theft Auto V would require millions of dollars to develop, but since these games are offered commercially at a fraction of the price the game manufacturer is essentially sharing the cost of the research and development to create the software necessary to conduct this research. This is important because it sheds light as to how these computer game simulations can be put to use, eliminating costs, time and human resources involved with building and operating complex machines capable of driving without the help of a human being. This article is ultimately helpful because it provides a solution to the massive costs and human resources required to previously conduct this research and with the use of computer simulated AI research, car companies could finically bring driverless cars to the everyday average consumer.