Real-Time Data-driven Technologies: Transparency and Fairness of Automated Decision-Making Processes Governed by Intricate Algorithms.

Author:Westbrook, Lorena
 
FREE EXCERPT
  1. Introduction

    Algorithmic decision-making processes may bring about more impartial decisions, consolidated in information that is typical of the group where the assessments apply, but they also generate unfairness, data irregularity and dearth of transparency. (Lepri et al., 2018). A substantial component of unfairness in algorithms derives from the statistics they are trained upon (numerous algorithmic systems are advanced by private firms and determined from proprietary code). (Shah, 2018) decision-making facilitated by algorithms advanced by machine learning is continuously more shaping individuals' lives, but complete ambiguity as regards the process constitutes the standard approach. (de Laat, 2018)

  2. Conceptual Framework and Literature Review

    The shift towards data-driven algorithms is an indication of an appeal for higher equity (Connolly-Barker, 2018; Meila, 2018; Lazaroiu, 2018; Nica, 2018; Park, 2018), evidence-based decision-making, and a superior grasp of individuals' separate and shared behaviors and needs. (Lepri et al., 2018) While algorithms are mechanisms that enable automated decision-making, or a concatenation of definite guidelines, bias constitutes an offshoot of such computations, affecting historically underprivileged groups. (Turner Lee, 2018) Being progressively programmed to enhance themselves without assistance, algorithms may become extremely ambiguous, surpassing the human capacity to comprehend. (Harambam et al., 2018)

  3. Methodology and Empirical Analysis

    Building our argument by drawing on data collected from Pew Research Center, we performed analyses and made estimates regarding % of Facebook users who say they think users have no/a little/a lot of control over the content that appears in their newsfeed and % of social media users who say it is acceptable for social media sites to use data about them and their online activities to recommend events in their area/recommend someone they might want to know/show them ads for products and services/show them messages from political campaigns (by age group). Structural equation modeling was used to analyze the collected data.

  4. Results and Discussion

    In algorithmic governance, unambiguousness and explicability are relevant ideals, but automated decision-making is being overestimated, maybe by virtue of an unreasonably high appraisal of the level of comprehensibility achievable from human decision-makers. (Zerilli et al., 2018) Algorithmic outcomes of machine learning, if all goes well, generate rigorous end results, but a clarification in coherent terms in respect to why a certain decision is suggested may not be provided. (de Laat, 2018)

    Spread-out neural networks encompassing various interrelated layers constitute algorithmic systems whose assessments...

To continue reading

FREE SIGN UP