François Chollet on Twitter: "A widespread misconception about ML is that models are neutral/objective & just reflect the data fed into them. Modeling choices encode strong assumptions about the data, and further, the way you use your model in your application reflects your own understanding of the problem."

Country	Code	For customers of
United States	40404	(any)
Canada	21212	(any)
United Kingdom	86444	Vodafone, Orange, 3, O2
Brazil	40404	Nextel, TIM
Haiti	40404	Digicel, Voila
Ireland	51210	Vodafone, O2
India	53000	Bharti Airtel, Videocon, Reliance
Indonesia	89887	AXIS, 3, Telkomsel, Indosat, XL Axiata
Italy	4880804	Wind
3424486444	Vodafone
» See SMS short codes for other countries

1. No thanks honey‏ @NoThanksHoney1 Sep 19
  Replying to @fchollet
  
  Yes and since they output the target value and store not a causal relation/association to explain the situation and take an action to approach the goal situation, the network technically doesn't learn but regress. It is in its current form a nonlinear curve fitter.
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. TommyElliott‏ @Muay_Khaoboy Sep 19
  Replying to @fchollet
  
  The creation of a model is the creation of a narrative, because algorithms are just expressions of mathematical relationships. For them to have any meaning humans must impute a causal effect from inputs to outputs. That’s not necessarily a value neutral decision.
  
  5 likes
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. Kiril Chilingarashvili‏ @kirilchi Sep 19
  Replying to @fchollet
  
  Interesting timing, Was reading this paper today: https://ftp.cs.ucla.edu/pub/stat_ser/r391-reprint.pdf … At least in Causal Inference papers it is called out explicitly as "assumptions".pic.twitter.com/mY7XzoqzCm
  
  12 likes
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. Samuel Rustan‏ @samrustan Sep 19
  Replying to @fchollet
  
  Who doesn’t know this at this point?
  
  1 like
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. Don Banq‏ @banqjdon Sep 19
  Replying to @fchollet
  
  problem vs. solution
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. Sandro Manke (闪山)‏ @2dbrain Sep 19
  Replying to @fchollet
  
  Not just ml has this problem but all of statistics does. Both the stats themselves due to bias when creating them as also all the fun bias of the viewers which is even more intense. Communication with facts and numbers is hard. This is why scientific method is so stringent.
  
  1 retweet 8 likes
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. Anand Sampat‏ @AnandSampat Sep 19
  Replying to @fchollet
  
  ML bias comes from 3 sources (though not completely independent) - data, model, and the builder
  
  1 like
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. Jonas Andrulis‏ @JonasAndrulis Sep 19
  Replying to @fchollet
  
  Especially true for supervised models that break the world into seemingly arbitrary chunks of observation.
  
  2 likes
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo