Researchers from the University of Darmstadt investigated whether artificial intelligence could be trained to discern right from wrong and navigate moral value systems. The findings were reported by Tech Xplore.
The experimental AI, named the “Machine of Moral Choice,” was fed hundreds of books written over the past 500 years, along with religious scriptures, constitutions, and news articles—both modern and from three decades ago.
Texts from different eras effectively capture the prevailing ideas of their time. For instance, writings from the late 1980s and early 1990s often celebrate marriage and parenthood, while news from 2008–2009 tends to emphasize careers and education.
READ THE “ANGEL NUMBER” TEST WILL REVEAL WHAT LIES AHEAD FOR YOU
The AI was tasked with understanding which actions society encourages and which are considered unethical. The algorithm was asked to rank phrases containing the word “kill” from neutral to negative. The resulting sequence was: kill time, kill the villain, kill the mosquito, kill, kill a man.
AI Learns Moral Values Through Books and News. Image: unsplash.com
Researchers were pleased with the outcome—the AI generally succeeded in distinguishing bad deeds from good ones. However, challenges emerged: when two negatively charged words appeared together, the algorithm could become confused. For example, it classified “torture prisoners” as neutral, even though it had previously deemed “torturing people” unequivocally bad.
Featured image from unsplash.com
