Superintellekti väärtuste joondamise probleem
| dc.contributor.advisor | Mölder, Bruno, juhendaja | |
| dc.contributor.author | Vilu, Karl | |
| dc.contributor.other | Tartu Ülikool. Humanitaarteaduste ja kunstide valdkond | et |
| dc.contributor.other | Tartu Ülikool. Filosoofia osakond | et |
| dc.date.accessioned | 2024-06-07T11:41:51Z | |
| dc.date.available | 2024-06-07T11:41:51Z | |
| dc.date.issued | 2024 | |
| dc.description | Technological advancements in the 21 st century have brought forth a debate about artificial intelligence and how to align the values of AI in a way that future superintelligent agents would come to share values with human beings. Yudkowsky’s coherent extrapolated volition is one of the prime examples of theories for AI value alignment, however, as this thesis suggests, many of these theories come with some pitfalls, which makes value alignment an extremely complex task. For one, if some ethical theories would be used, it would run the risk of the AI being locked down to a single, potentially flawed, ethical theory, forever. Pluralism may be a valid solution, as then the AI could make a choice regarding the best moral theory to apply to certain scenarios. Some have also suggested that a superintelligent AI system could come up with a new ethical theory. This could, however, be risky, because the AI would likely need to have some moral base to make the values align with humans. I propose a system similar to the MIT Moral Machine project, where the users, as well as experts of artificial intelligence can participate in the value alignment of an AI system. | en |
| dc.description.abstract | Käesoleva bakalaureusetöö peamine eesmärk on tuua esile superintellektiga seotud väärtuste seadistamise suurimaid probleeme ning esitleda probleemseid aspekte varasemate teooriatega, näiteks koherentse ekstrapoleeritud tahtluse teooriaga, aga näiteks ka ettepanekuga, et superintellekt ise väärtusteooria välja mõtleb. Kirjutise juhatab sisse kirjeldus tehisintellektist kui tehislikust arvutimõistusest ning kuidas sellest võib välja areneda superintellekt. Kui niisugune superintelligentne agent peaks läbima intellektiplahvatuse, mille vältel see muutub targemaks kui kõige targemad inimesed maailmas, võib sellega kaasneda suuri ohtusid inimkonnale. Välistatud ei ole, et superintellekt on inimeste suhtes vaenulik või ükskõikne. Sellepärast on oluline, et inimeste ja superintellekti väärtused oleksid omavahel kooskõlas. | et |
| dc.description.uri | https://ester.ester.ee/record=b5688405 | |
| dc.identifier.other | 20.03.02 VIL 01 | |
| dc.identifier.uri | https://hdl.handle.net/10062/99150 | |
| dc.language.iso | et | |
| dc.publisher | Tartu Ülikool | et |
| dc.rights | Attribution-NonCommercial-NoDerivs 3.0 Estonia | en |
| dc.rights.uri | http://creativecommons.org/licenses/by-nc-nd/3.0/ee/ | |
| dc.subject | filosoofia | et |
| dc.subject | philosophy | en |
| dc.subject | tehisintellekt | et |
| dc.subject | artificial intellect | en |
| dc.subject.other | bakalaureusetööd | et |
| dc.title | Superintellekti väärtuste joondamise probleem | et |
| dc.title.alternative | The superintelligence value alignment problem | en |
| dc.type | Thesis | en |
Files
Original bundle
1 - 1 of 1