Özerk Tools Launches Human Managers’ Preferences

Although the development of autonomous vehicles is an exciting field in the technology world, a critical question for a new research industry: Can these systems take unconscious bias that human drivers have? The study published in arXiv reveals this risk during the use of general-purpose language models in autonomous vehicle decisions mechanisms.

The main findings of the research are:

• Psychology studies show that human drivers lead less and different proportions across pedestrians with certain demographic features Tags
• Language models can learn and copy these behavior bias
• The use of artificial intelligence in decision-making processes of autonomous vehicles means that these bias can be widespread at the system level
• Working makes these problems measurable by offering a new criteria (benchmark)
• Providing public trust can depend on the mechanisms of fair decision-making up to technical success

• Adalet issue* becomes critical for the adoption of autonomous vehicles. It is not enough to provide traffic security in technical terms; it is necessary to act in accordance with social justice at the same time. These findings highlight a vulnerability that develops autonomous vehicles and should pay attention to the selection of data set and model training stages.

As a result, when designing language models-based autonomous systems, a systematic approach is necessary to become digital versions of human prejudices.