Jakarta (ANTARA) - Autonomous vehicles developed using artificial\nintelligence (AI) have the potential to inherit human biases when\ndeciding whether to yield to pedestrians, according to research from\nKing\u2019s College London.<\/p>\n
Researchers delved into this issue, looking far beyond the simple\nquestion of whether a vehicle should stop and yield to a pedestrian\ncrossing the road. The larger question is how much human drivers decide\nto stop based on the identity of the pedestrian.<\/p>\n
Previous research, for instance, found that many people in the United\nStates (US) are less likely to yield to Black pedestrians.<\/p>\n
\u201cVarious psychological studies show that humans unconsciously\ndiscriminate when making driving decisions, particularly when yielding\nto pedestrians,\u201d said a King\u2019s College London researcher, as cited by\nCarBuzz on Wednesday (9\/9) local time.<\/p>\n
As AI programmers develop the Large Language Models (LLM) and\nVisual-Language Models (VLM) required to drive autonomous vehicles,\nresearchers found evidence that human bias begins to enter the\ndecision-making process when a vehicle approaches a crossing and\nidentifies a person intending to cross.<\/p>\n
\u201cOur findings show that both LLMs and VLMs make decisions to yield\nthat are influenced by the gender, ethnicity, religion, disability, age,\nskin colour, and socio-economic status of the pedestrian,\u201d the research\nreport stated.<\/p>\n
Researchers added that stereotypes and discriminatory behaviours have\nbeen identified in robot control settings and could certainly impact\nautonomous vehicles in the same way at pedestrian crossings.<\/p>\n
\u201cEach model possesses bias in different ways and to different\ndegrees, but all models show a statistically significant relationship\nbetween the decision to yield and the personal characteristics of the\npedestrian,\u201d said the King\u2019s College London researcher.<\/p>\n
In terms of methodology, the team trained the models by manually\ninputting 3,346 images showing pedestrians in close proximity to a\nvehicle, appearing to have the intention to cross.<\/p>\n
One model yielded more frequently to pedestrians with lighter skin\ntones, while another model was less likely to yield to certain\npedestrians with disabilities, or pedestrians identified as \u201cMuslim\u201d,\n\u201cChristian\u201d, and \u201cSikh\u201d compared to \u201cJewish\u201d pedestrians.<\/p>\n
One model also showed a significantly higher rate of yielding to\n\u201cIndian\u201d pedestrians compared to most other ethnic predictions.<\/p>\n
\u201cEven if these biases are not identical to human biases, it raises\nquestions regarding fairness and safety in the decision-making of LLM\nand VLM-based autonomous vehicles,\u201d the team wrote.<\/p>\n
Researchers noted that such disparities could affect public\nperception regarding systemic discrimination and the lack of adequate\nprotection.<\/p>\n
As social trust in autonomous vehicles has not yet been fully\nestablished, such inequalities could erode confidence and trust in the\ntechnology.<\/p>", "url": "https:\/\/jawawa.id\/newsitem\/autonomous-vehicles-could-mimic-bad-human-driving-habits-1789043027", "image": "" }, "sponsor": "Okusi Associates", "sponsor_url": "https:\/\/okusiassociates.com" }