TFL Vsebine / Revija SIR*IUS

Lessons from the past for the age of artificial intelligence

Organizacije javnega sektorja po vsem svetu že desetletja uporabljajo na algoritmih zasnovane informacijske rešitve za podporo odločanju – med drugim pri upravnih postopkih, odločanju o pravicah iz javnih sredstev, presoji kandidatov za pomilostitev, prediktivnem policijskem delu in v enem primeru celo pri določanju zaključnih ocen srednješolskih izpitov. Čeprav so bila pričakovanja ob uvedbi teh rešitev pogosto povezana z večjo učinkovitostjo, objektivnostjo in poenotenjem odločanja, se je v praksi pokazalo, da lahko taki sistemi povzročijo resne težave, kot so pristranskost, netransparentnost delovanja in druge neželene posledice za posameznike ter družbo. Prispevek osvetljuje nekatere bolj znane primere uspešnosti in neuspešnosti na algoritmih zasnovanih informacijskih rešitev ter iz njih izpeljuje ključna tveganja in izzive, s katerimi se bomo soočali ob uvedbi sodobnih rešitev, ki bodo temeljile na tehnologijah umetne inteligence.

Public sector organizations around the world have been using algorithmically based decision support systems for decades. These systems have been applied in administrative procedures, the adjudication of social benefits, parole assessments, predictive policing, and - in one particularly notable case - even in the assignment of final secondary school exam grades. Although these solutions were often introduced with the expectation of improving efficiency, objectivity, and consistency in decision-making, real-world experience has shown that such systems can also lead to serious problems, including bias, lack of transparency, and other unintended consequences for individuals and society at large. This article highlights some of the most well-known successes and failures of past algorithmically based decision support systems in the public sector and derives key risks and challenges relevant to the introduction of modern AI-based solutions.