Connect with us
humans control self-driving cars humans control self-driving cars


Humans Will Still Be in Control of Self-Driving Vehicles

 4 min read / 

Self-driving cars were born with the objective to eliminate the need to have a human behind the steering wheel. Driving is a ‘routine task’, consisting of a series of repetitive actions to move the vehicle from one point to another safely through a sea of other vehicles, pedestrians and bicyclists. If one drives through massive traffic jams over long distances, one has probably understood the stressful effort it demands, not to mention the amount of time wasted, time is money! With the increasing availability of data, artificial intelligence (AI) has been gaining ground over the recent years. Like other industrial machines, using AI, companies are now striving to automate the ‘mundane’ task of driving, so that we can utilise our time in better ways. Even with numerous companies investing billions of dollars and racing to build automated drive technology, no company has perfected the technology yet. The question though is: Can technology be really developed to 100%?

The Artificially Intelligent Machine

Whether it is machine learning, deep learning, neural networks or other cognitive methods, every industry is making use of AI to make profitable sense of data. There are a couple of simple examples.

Gmail successfully filters 99.9% of spam using machine learning algorithms.

Udacity’s self-driving car engineers are building projects consisting of deep learning algorithms that recognize traffic signs with 98% accuracy as part of their curriculum. Whether AI is being used for medical diagnostics or to play board games, the accuracy of the algorithms is not always 100% because in the end, software code is software code, and it always vulnerable to bugs in the system or in the code itself. Systems can only be as reliable as 99.999999%… but almost never a complete 100%. The automotive industry is highly regulated, and for good cause, to ensure human safety. That also means that unless vehicles are truly reliable, they will still pose a danger to life. So, how then do companies ensure that self-driving vehicles are at least 99.999999% reliable?

Human-in-the-Loop Architecture

Humans will play an important role in the safety operations of self-driving vehicles to ensure that the AI-based systems are more confident of the actions they take. Take a look at the input-output loop in the illustration below:

Source: Mahbubul Alam, BrightTalk

When the data is fed into the AI classifier (basically the algorithms), the system gauges the confidence level of the decision made by the AI agent. If it does not meet the threshold value of the confidence level required to perform the action, the task is then moved on to human annotation. In terms of self-driving vehicles, remote operators will be responsible for this and would make critical decisions for the vehicle. Once the threshold of the confidence level is reached, the most appropriate action is performed by the vehicle. Whenever human annotation is used, it is recorded in the system and the AI learns from that particular situation – also known as ‘supervised learning’, a memory that is stored and is used without human annotation when that particular situation arises again. When it comes to decisions regarding safety and security, especially for human lives, the confidence level threshold is set pretty high. This, in turn, ensures that the self-driving vehicle will operate at a higher reliability rate than just 99.999…%, if not a complete 100%, which in simple terms means fewer accidents.

Yes, self-driving vehicles are coming. The assumption is that these vehicles will be safer than the ones we drive ourselves. But how comfortable are we to actually ride one? Do we trust these vehicles to be reliable enough to give up all control? Self-driving vehicles will be more than just vehicles that drive themselves, they will be part of a bigger ecosystem that entails telecommunications and live human interactions with remote operators. Ultimately, a human will still be behind the ‘steering wheel’ of the vehicle, the only difference is that the ratio of human drivers to vehicles will change from 1:1 to 1:1000, meaning that one remote operator will be responsible for about 1000 self-driving vehicles in the future. Exciting times ahead!

Sign up to Mogul News.

Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Send this to a friend