Speech english 5

Speech english 5

1. examples

   • Two black people were catagorized as gorillas by google automatic image labeling
   • TayTweets, bot developed by microsoft to intereract with users on Twitter
     • started tweeting racist, sexist and Hitler-approving messages in the first 24 houres
   • Facial recegnition machines
     • 34% more erros with faces of darkskinned woman than white skinned males
       • the darker the shade, the more the machines guesses were like throwing a coin (50-50)
     • Joy Buolamwini discovered the problem
       • she did a study and tested it on facial recognitionprograms such as microsoft, IBM and Face++
         • she did it in 2018
   • google translate
     • when translating from genderneutral languages to genderlanguages, it favors the pronoun "he"
       • not just in google tranlsate, but in every search thingy
         • example google images
           • doctor vs. nurse
           • also indian brides gets labeled costumes, while white brides get labeled bride and dress etc.
   • google
     • suggestions based on your gender
       • if you're a woman you may not get ads for higer paid jobs
   • algorithmic bias
     • google search results are based upon what others search for and want to read

2. source of problem

   • Race is generalization; where you take ceratin characteristics and apply them to a big group of people
     • key point in machine learning is generalization
       • especially when you make classification algorithms
       • problem! if the data that you use to program the AI is biased, the AI will be biased
         • algorithms are programmed with sotck images from ImageNet, where there's an overrepresentation of people from the US
           • the data you program the AI with is often biased
             • also takes inforamtion from historical events, which are also biased
               • counsequense: the history we try to forget will never really disappear
         • famous case: mislabbelin of two African-American young guys
         • to avoid: do proper data collection and data organization
           • most compaines/programmers just uses "mainstream" data

3. källor

   • Is AI racist? (first article that inspired everyting): https://towardsdatascience.com/is-artificial-intelligence-racist-and-other-concerns-817fa60d75e9
   • Nature (talks about training data, the problem of trying the AI on google-the loop): https://www.nature.com/articles/d41586-018-05707-8
   • Joy Buolamwini (Gender Shades; facial recognition, different skintones): https://www.forbes.com/sites/parmyolson/2018/02/26/artificial-intelligence-ai-bias-google/#7a16bdda1a01
   • African-American guyes labeled wrong: https://www.forbes.com/sites/mzhang/2015/07/01/google-photos-tags-two-african-americans-as-gorillas-through-facial-recognition-software/#73c8f150713d
   • Tay-the tweetbot: https://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist
   • Washingtonpost article: https://www.washingtonpost.com/opinions/2018/12/17/why-your-ai-might-be-racist/?noredirect=on&utm_term=.b108a923d894