How do machines learn? Evaluating the AIcon2abs method

• DOI: 10.48550/arxiv.2401.07386
• Publication Date: 2024-01-01
• Authors: Rubens Lacerda Qu...; Cabral Lima; Fábio Ferrentini ...; Priscila Machado ...
• Keywords: Realization (probability)

This paper evaluates AI from concrete to Abstract (Queiroz et al. 2021), a recently proposed method that enables awareness among the general public on machine learning. Such is possible due use of WiSARD, an easily understandable learning mechanism, thus requiring little effort and no technical background target users. WiSARD adherent digital computing; training consists writing RAM-type memories, classification reading these memories. The model easy visualization understanding tasks' internal realization through ludic activities. Furthermore, does not require Internet connection for classification, it can learn few or one example. also create "mental images" what has learned so far, evidencing key features pertaining given class. AIcon2abs method's effectiveness was assessed evaluation remote course with workload approximately 6 hours. It completed by thirty-four Brazilian subjects: 5 children between 8 11 years old; adolescents 12 17 24 adults 21 72 old. collected data analyzed two perspectives: (i) perspective pre-experiment (of mixed methods nature) (ii) phenomenological qualitative nature). well-rated almost 100% research subjects, revealed quite satisfactory results concerning intended outcomes. been approved CEP/HUCFF/FM/UFRJ Human Research Ethics Committee.