Google Scholar has over 2000 occurrences of "algorithmic bias" <https://scholar.google.com/scholar?q=%22algorithmic+bias%22> -- the idiom now apparently used to prevent machine learning from learning things that violate norms.
35 of them <https://scholar.google.com/scholar?q=%22algorithmic+bias%22++%22solomonoff%22+OR+%22minimum+description+length%22+OR+%22kolmogorov+complexity%22+OR+%22algorithmic+information%22++OR+%22algorithmic+probability%22> refer to "solomonoff" OR "minimum description length" OR "kolmogorov complexity" OR "algorithmic information" OR "algorithmic probability" On Sat, Feb 15, 2020 at 3:54 PM James Bowery <[email protected]> wrote: > On Sat, Feb 15, 2020 at 2:48 PM Robert Levy <[email protected]> wrote: > >> Are the mods just going to ignore James Bowery? >> > > Since algorithmically correcting "bias" is now seen as a central > responsibility of network effect content monopolies like Google, Youtube, > Twitter, Facebook, etc. rigorously measuring a dataset's "bias" is even > more urgent than is measuring "intelligence" or even "friendliness". > > Exactly _how_ urgent? > > Consider this: > > These content monopolies are intent on avoiding "a repeat of the 2016 > election", whatever that means. One thing is for certain: Claims that > they are attempting to provide an unbiased view of the world via their > machine learning algorithms in the run up to the 2020 election is viewed > with a great deal of suspicion by people wielding on the order of 400 guns > in the US alone. > > That's _exactly_ how urgent. > > Since we're stuck with some form of "prior" (speed prior, space prior, > etc.), and any prior will introduce bias in some sense, it seems the more > minimal that prior, the less bias it introduces to a minimum description > length of all available data. > > So why aren't all these content giants striving to create the largest > database of diverse, longitudinal social measures that their hardware and > human resources can support, and losslessly compressing it, so as to have > an unbiased platform upon which to measure "bias" in new data being added > to their content stores? > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T9ab9fba591214e64-M9562806743d05c6db7d09719 Delivery options: https://agi.topicbox.com/groups/agi/subscription
