Can an AI robot have preferences?
Historically, AI was built on rule engines. Wikipedia has a Category:Rule engines. Interestingly, “Rule engines” now resolves to Semantic reasoner on Wikipedia. “Regular expression”, or RegEx, engines are some of the most basic “rule engines” (Comparison of regular expression engines). Anyway, rules have defined “preferences” in AI. Rules, and therefore preferences, are defined by the “creator”, in this case the programmer acting as creator.
Machine learning basically substitutes for rule engines; though, machine learning may also be used together with rule engines. But, machine learning classically requires training by the creator, or programmer, and thereby also defining “preferences”. However, so-called deep learning, using the neural network model, seems to provide the promise forUnsupervised learning of “rules”; for example, the food is on the table, and the food is green. We wouldn’t necessarily call these “rules”, but once learned become a form of “preference”; for example, food probably goes on the table, and food is likely to be green.
See also my quick and dirty webpages: