AI Could Make Car Accidents a Thing of the Past

Call it the Minority Report for cars. 

Scientists say they have developed a way to predict car crashes using artificial intelligence (AI), according to a new research paper. The deep learning model produces very high-resolution crash risk maps. It's part of a growing movement that uses AI to predict risks and help prevent accidents.

"AI technology inherently leverages historical data to deliver predictive insights," computer scientist Sameer Maskey, the CEO at FuseMachines, told Lifewire in an email interview. "Using artificial intelligence, it is possible to assess and study historical and behavioral data around everything from natural phenomena, such as wildfire, and manmade situations, such as car crashes and cyber attacks."

Precog AI?

In the movie Minority Report, actor Tom Cruise starred as a detective who used "precogs" to catch glimpses of the future and prevent crimes. Similarly, the AI technology that Massachusetts Institute of Technology (MIT) researchers have developed is intended to predict possible car accidents. 

The AI model is fed a combination of historical crash data, road maps, satellite imagery, and GPS. After crunching the numbers, the AI describes the expected number of crashes over some time in the future to identify high-risk areas and predict future impacts.

"By capturing the underlying risk distribution that determines the probability of future crashes at all places, and without any historical data, we can find safer routes, enable auto insurance companies to provide customized insurance plans based on driving trajectories of customers, help city planners design safer roads, and even predict future crashes," MIT Ph.D. student Songtao He, a lead author on a new paper about the research, said in a news release. 

In the autonomous driving vehicle industry, AI plays a significant role in planning and control, Maxwell Zhou, CEO of DeepRoute.ai, a company that develops autonomous driving solutions, told Lifewire in an email interview.

Sensors collect all the data of their surroundings and pass it to a computer to process utilizing deep learning models and machine learning algorithms. 

"We designed neural networks like the ones of human brains, so it receives training through massive road data which deepens its understanding of the environment and eventually draws out a complete perception system," Zhou said.

Look Into the Silicon Ball

Scientists are increasingly turning to AI to try to forecast events of all kinds. Some uses of AI include predicting cybersecurity threats and monitoring videos to predict natural phenomena such as wildfires, floods, and hurricanes.

"AI is useful in predicting unsafe events because of its ability to see further and infer at a faster rate than humans," Zhou said. 

AI is based on pattern recognition, Mike Betzer, the CEO of AI company Hypergiant, told Lifewire in an email interview. This means that machine learning models can compute large amounts of data and then make recommendations about a foreseeable outcome. 

"What the model is doing is creating a risk projection and helping people to understand the propensity for disaster," Betzer said. "We already see this with weather modeling, accidents modeling, and other dangerous events."

AI likely will be deeply integrated into autonomous vehicles of the future, Zhou predicted. In the future, cars and trucks, taxis, and buses will all have features such as a lane change system, sensor suite for collision avoidance, and a computing platform to process real-time information. 

"This means crash data will be collected and analyzed in real-time, post-crash response efficiency will be increased, and further safety issues can be reduced," he said. 

One promising area of current research that could help prevent traffic deaths is using AI to detect near-misses and dangerous behavior, AI road safety expert Sohaib Ahmad Khan told Lifewire in an email interview. 

"Each intersection can be given a safety rating based on their near-miss scores, and the city's resources can be directed to those which are more dangerous," he added. "This ability to quantitatively measure safety issues will have a lot of impact in the future."