top of page

Investors Are Impatient With AI Car Companies



INTRODUCTION

According to the June 13th Technology Quarterly edition of the Economist, the interest of investors in autonomous driving firms has taken a dip. After various promises that the world will reach complete autonomy by 2018 (made by Tesla founder Elon Musk), funders are getting impatient. Starsky Robotics, a self-driving car company in San Francisco, closed in March due to the recession but also because of investors’ inability to accept slow progress. It is fearful that other autonomous firms may follow. As a result, it is time that both investors and the public give up their rash fantasies of AI to understand how difficult it is to create self-driving cars.


PROBLEMS WITH TECHNOLOGY


Imagine trying to teach an ant the difference between “red” and “blue”. You may design an experiment placing two different fruits on two colored squares (raspberry on red; blueberry on blue). The ant may learn to identify red because it has the reward of a raspberry and the blue because it has the reward of a blueberry, but they do not know that “red” and “blue” are colors. Getting Artificial Intelligence to identify objects for what they are is the greatest roadblock for self-driving cars.

Computer Vision is the main technology that is used to create self-driving autonomy. Computer Vision takes inputs of images or video, passes the data through an algorithm, and the output is recognition, detection, and identification of objects.

For example: The input data is a picture of a stop sign. The output is the recognition, detection, and identification of the sign. A conditional in the computer coding then tells the car to perform the action of “stopping” when it sees a stop sign.

Training the AI computer vision of a car to identify objects is highly difficult. Humans must assemble vast amounts of correctly labeled data to teach AI the difference between a pedestrian and a traffic cone (neither of which the car should hit). Since there are countless instances and changes in a driving environment, the AI must learn from millions of scenarios or there will be consequences.



EXAMPLE

Even if the traffic signal is green, the AI needs to learn that they need to stop at an intersection if a car is coming towards them to avoid a head-on collision. A human driver would have the common sense to do this, but the Deep Learning technology only understands mapping inputs to outputs.


Dealing with weather is also difficult. There have been reports of AI vehicles getting confused on a sunny day when the light reflects off of surrounding white cars. The computer vision sensors were not able to detect or recognize that the reflective light was another vehicle, resulting in a car-crash. Heavy snow or rain also obscures the computer vision.

However, one of the greatest problems that prevents the autonomy of vehicles are the roads themselves. Cities may need to redesign their entire grids to make an easier landscape for autonomous vehicles to drive on. They will need to reorganize their infrastructure, street routes, backroads, and allies to accommodate the navigation abilities of self-driving cars.


CONCLUSION

From this basic discussion, it is seen how challenging it is to create self-driving autonomy. It is more extreme than creating an app that uses machine learning. Instead, autonomous driving companies are trying to create governing technology that will deliver humans unscathed from a dangerous, physical, environment.


Investors and the public need to understand the purpose behind this endeavor: According to the CDC, each year 1.35 million people on average are killed from road traffic accidents. Self-driving firms hope to reduce this number by implementing and harnessing AI technology. Surely then, the cause is worth it even if progress made by self-driving firms is slow.


If investors and the public dedicate more time to understanding the process and challenges that come from trying to improve road safety, they would not be as impatient with the current progress of self-driving firms. Therefore, it is recommended that the general population learn more about the tech of their dreams.


SOURCES

bottom of page