Autonomous Cars: An Adaptable Feedback Mechanism For Customised Ethics

Abstract

Autonomous vehicles (AVs) are the next breakthrough in the automobile industry and will pave the way to the public market in the near future. Apart from traffic efficiency, increased mobility and a hope for sustainable future with reduced pollution and consumption of fuels, they also promise to bring down the number of accidents up to 90 percent owing to the fact that the majority of accidents in the entire world happens because of human errors. However, this technology would not only make travelling safer but would also raise several ethical concerns regarding decision making in crash scenarios such as what would the autonomous vehicle do in a kill or be-killed scenario and the responsibility of the stakeholders in such crashes. This research work proposes to provide solutions for some of the ethical issues raised by the introduction of autonomous vehicles. It proposes an idea to capture the ethics of the consumers by generating a priority list which would be encapsulated by answering questions that provides user to choose between multiple scenarios. This research work also solves the issue of assigning the responsibility of the crashes by generating the user ethics/priority list and proposes the idea of a default ethics/priority list which would be pre-programmed into the new AV and would be followed if the consumer doesn’t want his own ethics to be followed by the car. Based on ethics that is followed whether user-defined or default, the responsibility of crashes would be assigned. Now, when the user defined priority list is ready to use, then in an accident scenario the sensors of the AV will detect the objects encountered. Next, the detected object will be classified into a category using a machine learning classifier and the priority of the objects will be determined from the priority list and the corresponding action will be taken. We tested the accuracy with which the objects with certain features are classified into a particular category in the experimentation phase of our proposed work using classifiers namely naïve Bayes, decision tree C4.5 and C5.0 and random forest. We used Weka and Rstudio tools for the experimentation phase. We then have chosen the algorithm that best suited our requirements and implemented the generation of the user defined priority list and the results of the implementation have shown that the action taken is in accordance with the ethics of the user.