Ethics in Artificial Intelligence: Bias, Fairness and Beyond by Unknown

Author:Unknown
Language: eng
Format: epub
ISBN: 9789819971848
Publisher: Springer Nature Singapore

4 Bridging Memory Ethics and AI-driven IR System Design

After identifying the common ethical challenges of human curation of genocide-related information, it is important to examine to what degree they are applicable for IR system-based curation and how IR system design can be bridged with memory ethics to address concerns associated with these challenges. For this purpose, this section will use the same framework of three groups of challenges related to respect for individual (consent, double vision, and privacy), beneficence (representation and distortion/manipulation), and fairness/justice (politicisation of curation and unequal treatment).

In the case of consent, the difficult part of bridging ethics and IR design relates to consent granting usually being part of the initial stage of data generation (e.g. the recording of a testimony or registering of an account to upload digital materials). One exception here relates to IR systems used in the context of web search, where indexing of digital-born materials is an ongoing process. However, in most cases, the IR systems process data for which consent has already been granted (e.g. in the case of collections stored in the heritage institution or materials generated through the online platform). While it can be possible to integrate consent checks or regular requests for consent re-granting, this specific aspect can arguably be more relevant for the overall model of institution/platform functionality and not necessarily for the IR design.

In terms of double vision, AI-driven IR systems are sometimes argued [30] to be capable of encouraging trust and empathy which can counter the potential dehumanisation of victims associated with this challenge. Such a problem can be particularly pressing with the passing of the living witnesses of the genocides which amplifies the risk of them being increasingly treated as objects and not the subjects of research. One particular example of the use of IR for countering this issue is the shift towards more human-to-human communication-like forms of IR, for instance, the use of conversational agents as a form of curation of genocide-related information.

The potential of IR systems for dealing with privacy-related challenges shares certain similarities with the case of content. In some cases, the matters of privacy are dealt with during the initial state of data generation (e.g. in the case of thorough anonymisation of genocide-related evidence). However, in other cases, IR systems can have a rather ambiguous impact on the privacy of individuals the information about whom they are curating. Advancements in several fields of AI (e.g. computer vision and natural language processing; [63]) enable novel possibilities for recognising the presence of individuals or mentions of specific entities in the data; however, the same advancements can be used for protecting individual privacy (e.g. by masking private information present in the documents). Under these circumstances, the inclusion of particular functionalities in the IR system design can either expose or protect private information. The choice of the functionalities can be informed by examining the selection of materials which the system is expected to work with and its intended uses.

Filtering out privacy-sensitive content can be another alternative to making modifications to the original data.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.