Jump to content

Talk:Algorithmic transparency

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Good definition, needs more meat

[edit]

Many years, ago I worked for the Oxford English Dictionary, writing definitions of words. When I looked at this draft, my first thought was "this is a really well-written definition". My second thought was "but it's just a definition, it needs more body. It needs to explain why algorithmic transparency matters."

For example, if a bank has a policy of refusing to provide loans to black people, that's clearly unacceptable. If instead it uses a neural net or other such opaque decision process, which uses place of residence and other such data to decide on what loans to offer, and the result is that it rarely grants loans to black people, that should also be unacceptable. The bank should not be able to get away with saying "it's a highly trained algorithm, no-one here understands how it works."

I'm sure tThere are respected published sources that make the same point as the above paragraph.[1][2][3][4][5] (Ok, I've expressed an opinion in that paragraph, and it should not be expressed in the voice of Wikipedia. What matters is that many people hold that opinion.) Maproom (talk) 07:34, 4 September 2017 (UTC)[reply]

  1. ^ "Privacy expert argues "algorithmic transparency" is crucial for online freedoms at UNESCO knowledge café". unesco.org.
  2. ^ Tétu, Stéphanie. "TransAlgo: assessing the accountability and transparency of algorithmic systems". inria. INRIA.
  3. ^ Sadler, Denham. "Medcraft on ethics and algorithms". innovationaus.com. InnovationAus.
  4. ^ Petrasic; et al. "Algorithms and bias: What lenders need to know". whitecase.com. White & Case. {{cite web}}: Explicit use of et al. in: |last1= (help)
  5. ^ Lapowsky, Issie. "One State's Bail Reform Exposes the Promise and Pitfalls of Tech-Driven Justice". Wired.com. Wired.