{"id":9313,"date":"2020-08-13T09:23:30","date_gmt":"2020-08-13T09:23:30","guid":{"rendered":"https:\/\/www.experfy.com\/blog\/?p=9313"},"modified":"2023-11-21T09:32:30","modified_gmt":"2023-11-21T09:32:30","slug":"ai-bias-and-what-you-can-do-about-it","status":"publish","type":"post","link":"https:\/\/www.experfy.com\/blog\/ai-ml\/ai-bias-and-what-you-can-do-about-it\/","title":{"rendered":"AI Bias And What You Can Do About It"},"content":{"rendered":"\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n

AI bias doesn\u2019t come from AI algorithms, it comes from people<\/a>. What does that mean and what can we do about it?<\/p>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t

Technology is not free of\u00a0humans<\/h2>\n<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

No technology is free of its creators. Despite our fondest sci-fi wishes, there\u2019s no such thing as\u00a0ML\/AI<\/a>\u00a0systems that are truly separate and autonomous\u2026 because they start with us.<\/p>\n\n\n\n

\n

All technology is an echo of the wishes of whoever built\u00a0it.<\/p>\n<\/blockquote>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

This isn\u2019t just about\u00a0ML\/AI<\/a>; any technology you\u2019re in danger of seeing as independent is simply the kind whose effect lingers longer after your button press.<\/p>\n\n\n\n

No matter how complex its echo, technology always comes from and is designed by people, which means it\u2019s no more objective than we are. This makes the definition of\u00a0algorithmic bias<\/a>\u00a0problematic.<\/p>\n\n\n\n

Algorithmic bias refers to situations where a computer system reflects the implicit\u00a0values\u00a0of the humans who created it, but doesn\u2019t all technology reflect its creators\u2019 implicit values?\u00a0And if you think humans can be completely unbiased, take a stroll through\u00a0this list<\/a>\u2026<\/p>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t

Data and math don\u2019t equal objectivity<\/h2>\n<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

If you\u2019re looking to\u00a0AI<\/a>\u00a0as your savior from human foibles, tread carefully.<\/p>\n\n\n\n

\n

Math can obscure the human element and give an illusion of objectivity.<\/p>\n<\/blockquote>\n\n\n\n

Sure, data and math can increase the amount of information you use in decision-making and\/or save you from heat-of-the-moment silliness, but how you use them is still up to you.<\/p>\n\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n\n

According to\u00a0Wikipedia<\/a>:<\/p>\n\n\n\n

\n

Objectivity<\/strong>\u00a0is a philosophical concept of being true independently from individual subjectivity caused by perception, emotions, or imagination.<\/p>\n<\/blockquote>\n\n\n\n

Tragicomically, a layer of math and data spread on top of the host of utterly subjective choices (What shall we apply AI to? Is it worth doing? In which circumstances? How shall we define success? How well does it need to work? etc.<\/a>) obscures the ever-present human element and gives an illusion of objectivity.<\/p>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n
\n

It\u2019s much flashier to say, \u201cThe AI learned to do this task all by itself\u2026\u201d than to tell the\u00a0truth.<\/p>\n<\/blockquote>\n\n\n\n

Look, I know sci-fi sells. It\u2019s much flashier to say,\u00a0\u201cThe AI learned to do this task all by itself\u2026\u201d<\/em>\u00a0than to tell the truth:\u00a0\u201cPeople used a tool with a cool name to help them write code. They fed in examples they considered appropriate, found some patterns in them, and turned those patterns into instructions. Then they checked whether they liked what those instructions do for them.\u201d<\/em><\/p>\n\n\n\n

The truth drips with human subjectivity\u200a\u2014\u200alook at all those little choices along the way which are left up to people running the project. Wrapping them in a glamorous coat of math doesn\u2019t make the core any less squishy.<\/p>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n
\n

When math is applied towards a purpose, that purpose is still shaped by the sensibilities of our\u00a0times.<\/p>\n<\/blockquote>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\"\"\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n

To make matters worse, the\u00a0whole point of AI<\/a>\u00a0is to let you explain your wishes to a computer using examples (data!<\/a>) instead of instructions. Which examples? Hey, that depends on what you\u2019re trying to teach your system to do. Datasets are like textbooks for your student to learn from. Guess what? Textbooks have authors.<\/p>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\"\"\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t“AI Bias And What You Can Do About It” \n
Did you answer \u201cbananas\u201d? Why didn\u2019t you mention the plastic bag roll? Or the color of the banana? This example comes from Google\u2019s AI Fairness training course<\/a> and demonstrates that although all three answers are technically correct, for some reason we have a bias to prefer one of them. Not all people would share that bias; what we perceive and how we respond is influenced by our norms. If you live on a planet where all bananas are blue, you might answer \u201cyellow bananas\u201d here. If you\u2019ve never seen a banana before, you might say \u201cshelves with yellow stuff on them.\u201d Also correct.<\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n

When I said that \u201cbias doesn\u2019t come from AI algorithms, it comes from people<\/em><\/a>\u201d\u00a0<\/em>some folks wrote to tell me that I\u2019m wrong because bias comes from data. Well, we can both be winners\u2026 because\u00a0people make the data<\/a>.<\/p>\n\n\n\n

\n

Textbooks reflect the biases of their authors. Like textbooks, datasets have authors. They\u2019re collected according to instructions made by\u00a0people.<\/p>\n<\/blockquote>\n\n\n\n

Other design choices like\u00a0launch criteria<\/a>, data\u00a0population<\/a>, and\u00a0more<\/a>\u00a0are also entirely up to\u00a0human decision-makers<\/a>, which is why it\u2019s so important that you take care to\u00a0choose your project leaders wisely<\/a>\u00a0and\u00a0train them well<\/a>.<\/p>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n
\n

It\u2019s safest to think of AI as a tool for writing\u00a0code.<\/p>\n<\/blockquote>\n\n\n\n

Please don\u2019t fall victim to the sci-fi hype. You already know that computer code is written by people, so it reflects their implicit values. Think of\u00a0AI<\/a>\u00a0as an excellent tool for writing code\u200a\u2014\u200abecause\u00a0that\u2019s what it is<\/a>\u200a\u2014\u200aso the same basics hold. Remember to\u00a0design<\/a>\u00a0and\u00a0test<\/a>\u00a0all technology carefully, especially when it scales.<\/p>\n\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t

This is no excuse to be a\u00a0jerk<\/h2><\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n

Machines are just tools. They\u2019re extensions of their creators, who, according to the gory definition in the last bullet point of\u00a0this list<\/a>\u00a0(\u201cwhen our past experiences distort our perception of and reaction to information\u201d<\/em>), are biased.<\/p>\n\n\n\n

\n

All of us had our perceptions shaped by our past experience and all of us are products of our personal histories. In that sense, every single human is\u00a0biased.<\/p>\n<\/blockquote>\n\n\n\n

Philosophical arguments invalidating the existence of truly unbiased and objective technology don\u2019t give anyone an excuse to be a jerk. If anything, the fact that you can\u2019t pass the ethical buck to a machine puts more responsibility on your shoulders, not less.<\/p>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\"\"\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n

Sure, our perceptions are shaped by our times. Societal ideas of virtue, justice, kindness, fairness, and honor aren\u2019t the same today as they were for peoples living a few thousand years ago and they may keep changing. That doesn\u2019t make the ideas unimportant, it only means we can\u2019t outsource them to a heap of wires. They\u2019re the responsibility of all of us, together. We should strive to do our best ethically and treat all people with the highest standards of respect and care.<\/p>\n\n\n\n

\n

The fact that you can\u2019t pass the ethical buck to a machine puts more responsibility on your shoulders, not\u00a0less.<\/p>\n<\/blockquote>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t

Fairness in\u00a0AI<\/h4><\/h4>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n

Another choice is which data to use for AI. You should expect better performance on examples that are similar to what your system learned from. If you choose not to use data from people like me, your system is more likely to make a mistake when I show up as your user. It\u2019s your duty to think about the pain you could cause when that happens. Yes, it takes effort and imagination (and extensive\u00a0analytics<\/a>).<\/p>\n\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\"\"\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t“AI Bias And What You Can Do About It” \n
Unfamiliar with analytics for comparing distributions<\/a>? Start your journey by reading up on histograms<\/a> like this one I made of the output of a model on two different simulated datasets.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

At a bare minimum, I hope you\u2019d have the common sense to check whether the\u00a0distribution<\/a>\u00a0of your user\u00a0population<\/a>\u00a0matches the\u00a0distribution<\/a>\u00a0in your data. For example, if 100% of your training examples come from residents of a single country, but your target users are global\u2026 expect a mess. Your system will treat that country\u2019s residents less sloppily than everyone else. Is that fair?<\/p>\n\n\n\n

If you\u2019re keen to take your fairness analytics to the next level, there are specialized tools out there to help you. One of my favorites is the\u00a0What-If Tool<\/a>.<\/p>\n\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\"\"\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Fair and\u00a0aware<\/h4>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n

There are\u00a0many choices<\/a>\u00a0that people are responsible for along the way to\u00a0AI<\/a>. Half of the solution is striving to design systems to be good and fair and just to everyone. The other half is annihilating your ignorance of the consequences of your\u00a0choices along the way<\/a>. Think. Carefully. (And use\u00a0analytics<\/a>\u00a0to help you think even more carefully.)<\/p>\n\n\n\n

\n

If you want to strive for fairness, work to annihilate your own ignorance of the consequences of your decisions.<\/p>\n<\/blockquote>\n\n\n\n

I\u2019ve written a lot of words here, when I could have just told you that most of the research on the topic of bias and fairness in\u00a0ML\/AI<\/a>\u00a0is about making sure that your system doesn\u2019t have a disproportionate effect on some group of users relative to others. The primary focus is on\u00a0distribution<\/a>\u00a0checks and similar\u00a0analytics<\/a>. The reason I wrote so much is that I want you to do even better. Automated distribution checks only go so far. No one knows a system better than its creators, so if you\u2019re building one, take the time to think.<\/p>\n\n\n\n

Think about how your actions will affect people and do your best to give those you\u2019ll affect a voice to guide you through your blind spots.<\/p>\n\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t
\n\t\t\t
<\/div>\t\t<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n

 <\/p>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<\/div>\n\t\t","protected":false},"excerpt":{"rendered":"

AI bias doesn\u2019t come from AI algorithms, it comes from people. What does that mean and what can we do about it? <\/p>\n","protected":false},"author":335,"featured_media":9314,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"content-type":"","footnotes":""},"categories":[183],"tags":[226,342,548],"ppma_author":[2050],"class_list":["post-9313","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-ml","tag-ai","tag-algorithm","tag-bias"],"authors":[{"term_id":2050,"user_id":335,"is_guest":0,"slug":"cassie-kozyrkov","display_name":"Cassie Kozyrkov","avatar_url":"https:\/\/www.experfy.com\/blog\/wp-content\/uploads\/2020\/04\/medium_df35f80d-2bff-4fe3-b741-a94d51320e00-150x150.jpg","user_url":"https:\/\/careers.google.com\/?src=Online\/LinkedIn\/linkedin_profilepage&utm_source","last_name":"Kozyrkov","first_name":"Cassie","job_title":"","description":"Cassie Kozyrkov is Chief Decision Scientist at Google, Inc. With a unique combination of deep technical expertise, and world-class public-speaking skills, she has provided guidance on more than 100 projects and designed Google's analytics program, personally training over 15000 Googlers in statistics, decision-making, and machine learning.\u00a0"}],"_links":{"self":[{"href":"https:\/\/www.experfy.com\/blog\/wp-json\/wp\/v2\/posts\/9313","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.experfy.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.experfy.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.experfy.com\/blog\/wp-json\/wp\/v2\/users\/335"}],"replies":[{"embeddable":true,"href":"https:\/\/www.experfy.com\/blog\/wp-json\/wp\/v2\/comments?post=9313"}],"version-history":[{"count":5,"href":"https:\/\/www.experfy.com\/blog\/wp-json\/wp\/v2\/posts\/9313\/revisions"}],"predecessor-version":[{"id":34203,"href":"https:\/\/www.experfy.com\/blog\/wp-json\/wp\/v2\/posts\/9313\/revisions\/34203"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.experfy.com\/blog\/wp-json\/wp\/v2\/media\/9314"}],"wp:attachment":[{"href":"https:\/\/www.experfy.com\/blog\/wp-json\/wp\/v2\/media?parent=9313"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.experfy.com\/blog\/wp-json\/wp\/v2\/categories?post=9313"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.experfy.com\/blog\/wp-json\/wp\/v2\/tags?post=9313"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/www.experfy.com\/blog\/wp-json\/wp\/v2\/ppma_author?post=9313"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}