Purchases, conversations, access to information, music and movies: more and more of our online life is mediated by complex algorithms that are designed to make the experience of the Web customized and more “personal”. These algorithms can process an amount of heterogeneous data that would take enormous resources for the human mind to cope with, and find valuable patterns in it. Their use is not limited to our online experience as similar algorithms have been also implemented, for example, to inform policymakers: suggesting where to deploy police forces around the urban context, assessing criminality risk scores of offenders, or allocating high school students to the most suited school. While the consequences of the decisions made by algorithms have a great impact on people’s lives, the way they are built and designed makes them de facto “black boxes”: a series of legal and technical barriers prevents from accessing and understanding how a certain input influences a given output. Overseeing their decision processes becomes then of the utmost importance. This paper argues that visualizations can become a powerful tool to monitor algorithms and make their complexity accessible and usable by visually showing the relation between the inputs and the outputs in a manner that mimics an observational study approach. The paper analyzes a case study developed as an experiment to test opportunities and criticalities in using visualization to represent the presence and the activity of algorithms. This represents a shift from the main purpose of visualizations and Data Visualization in general: since its strong suit is to support human decision making processes by transforming data into knowledge, the substitution of people by machines in this activity seems to make visualizations obsolete. A computer doesn’t need to “see” the data to make a decision – or at least not in the same way as people do – no matter how multidimensional and heterogeneous the data is. With the diffusion of algorithms, the need to inspect their accountability and performance will simply move visualizations at a later stage. From a decision making tool visualizations become a monitoring and awareness tool.

Visual Monitoring of Complex Algorithms

Mauri, Michele;Ciuccarelli, Paolo
2020

Abstract

Purchases, conversations, access to information, music and movies: more and more of our online life is mediated by complex algorithms that are designed to make the experience of the Web customized and more “personal”. These algorithms can process an amount of heterogeneous data that would take enormous resources for the human mind to cope with, and find valuable patterns in it. Their use is not limited to our online experience as similar algorithms have been also implemented, for example, to inform policymakers: suggesting where to deploy police forces around the urban context, assessing criminality risk scores of offenders, or allocating high school students to the most suited school. While the consequences of the decisions made by algorithms have a great impact on people’s lives, the way they are built and designed makes them de facto “black boxes”: a series of legal and technical barriers prevents from accessing and understanding how a certain input influences a given output. Overseeing their decision processes becomes then of the utmost importance. This paper argues that visualizations can become a powerful tool to monitor algorithms and make their complexity accessible and usable by visually showing the relation between the inputs and the outputs in a manner that mimics an observational study approach. The paper analyzes a case study developed as an experiment to test opportunities and criticalities in using visualization to represent the presence and the activity of algorithms. This represents a shift from the main purpose of visualizations and Data Visualization in general: since its strong suit is to support human decision making processes by transforming data into knowledge, the substitution of people by machines in this activity seems to make visualizations obsolete. A computer doesn’t need to “see” the data to make a decision – or at least not in the same way as people do – no matter how multidimensional and heterogeneous the data is. With the diffusion of algorithms, the need to inspect their accountability and performance will simply move visualizations at a later stage. From a decision making tool visualizations become a monitoring and awareness tool.
2CO: COmmunicating COmplexity Contributions from the 2017 Tenerife Conference
9788409101719
Data Visualization, Algorithmic Accountability, Visualizing complexity
File in questo prodotto:
File Dimensione Formato  
2CO Communicating Complexity_14.pdf

accesso aperto

: Publisher’s version
Dimensione 974.96 kB
Formato Adobe PDF
974.96 kB Adobe PDF Visualizza/Apri
2co-scientific-committee.jpg

Accesso riservato

Descrizione: Descrizione del comitato scientifico
: Altro materiale allegato
Dimensione 301.67 kB
Formato JPEG
301.67 kB JPEG   Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11311/1136559
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact