Inquiry-Based Research Essay

In this essay we had to come up with a question that wasn’t easily answered. Then using 4 academic sources and 3 popular media sources we had to explore the question in 8-10 pages. My topic was still artificial intelligence, however it was focused on diversity in the tech workplace. My question for this was…

Topic Reflection

My topic artificial intelligence, interests me because there are so many perspectives on artificial intelligence. So many people have different perspectives on artificial intelligence and what it will mean for the future. Some people think that it will be the end of humanity seen in cinematography like The 100, IRobot, and Ex-Machina. Some people think that it will be the singularity which propels humanity into a golden era where no one has to work and no one is impoverished. Some people think it’s impossible and that it will never happen, there are just so many possibilities. I think it’s one of the biggest factors in our future right now, which is why it passes the “So what” test.I think artificial intelligence is too broad of a question to derive a good research question on just the topic itself, but I need to delve deeper into the many aspects that artificial intelligence will affect. One aspect I think it will have a huge affect on is poverty. However, it’s good thing only if it’s implemented properly into our society. Which is where the research question comes in. How will artificial intelligence affect the future in terms of poverty? There are many compositions that currently exist in relation to artificial intelligence and poverty. Like peer reviewed articles on the ccny library that explain how AI can be used to benefit the poor’s health, and other ways that AI can affect poverty in a good way. Also an article which says how AI is being used to identify where poverty is at its worst and find ways to help the situation. I think I may have struggles parsing through the sources to find relevant information to explain the research question’s “answer”. Also I think I may have difficulty in finding an essay with the same premise as mine. If this question doesn’t work out I can maybe try a more broad question like,  How will artificial intelligence affect the future of humanity? or Will artificial intelligence ever exist?

My Research Proposal

My research question, How will artificial intelligence affect the future of poverty?, interests me because artificial intelligence is a huge topic right now, not only in the technology area, but just generally. There are many ways that artificial intelligence can be used to change our future. So I wanted to focus on one area that artificial intelligence could potentially change, poverty. It’s relevant to social action because poverty is a huge problem throughout the world and must be addressed before it gets worse with our growing population. In the U.S., according to a Census taken by the government, 11.8 % of the United States population in 2018 is classified as being in poverty. So it’s important to analyze what artificial intelligence will do to poverty in the future, whether it will decrease or increase this percentage. It might be assumed that artificial intelligence will be a detriment to this rate and actually increase it because as artificial intelligence becomes more competent, this means less jobs, which means the unemployment rate will increase and cause the poverty rate to increase as well. However this may not be true, if governments regulate a company’s use of artificial intelligence, it could decrease poverty rates. While there may be some losses of jobs due to artificial intelligence, there will almost certainly be gains in the job market as well. Technological innovation has been going on for a long time, and every time a big advancement in technology happens people think about the jobs being lost and not the jobs being gained. One example of artificial intelligence creating more jobs is for human translators. As artificial intelligence gets better at translating languages, companies are doing more and more business across language barriers and even though artificial intelligence is helping bridge the gap between languages, this doesn’t mean that artificial intelligence can translate everything perfectly, especially for more complex language. This means that companies will need more human translators for these more complex interactions. (https://singularityhub.com/2019/01/01/ai-will-create-millions-more-jobs-than-it-will-destroy-heres-how/)

I should have all my sources completed by 10/17.My intended audience is for those interested in how technology could shape our future, for better or worse and for people interested in potential solutions for global poverty. An appropriate publication that would fit my essay would be Forbes.com, because they have their own AI section on their platform, so anyone interested in AI would likely be there. A great model essay that I will be using to look at while doing my essay is, https://www.nbcnews.com/mach/tech/ai-game-changer-fight-against-hunger-poverty-here-s-why-ncna774696, which also talks about what I’m focusing on in my research question.

Bibliography

Report on Research in Progress

The conference gave me great insight on what I should change about my essay. It also gave me inspiration for more substance in my essay since it was a little short for the requirements. After the conference I went to change some of the informal word choices that I had made, this helped make the essay more formal, helping with the goal of the essay. I also took out some quotes in favor of paraphrasing the information. This provided a better flow to the essay and provided a little bit more of my insight into the quote or paraphrased text. After reading, I believe it was the Orwell piece with the list, and the suggestion to include a list into the piece I thought it was a great idea and implemented the list promptly. I believe that this helps with the organization, at least to me, it feels more organized than it was before. My conclusion was a bit redundant, so I took out  the redundant parts that just repeated parts stated in the body paragraph and made it more short and concise. I feel like I wrapped it up a bit better than i had before, without a formulaic structure. Lastly I went deeper into one of the questions that arose from my initial question.

The Essay

How do we avoid racial and sexist biases in artificial intelligence?

The end of the world may be coming in our near future, with talks of artificial intelligence taking over the world, the talks have also taken over the media. Elon musk and Stephen Hawking agree and have stated that artificial intelligence might be the end of the human race. Although these concerns are important, they aren’t issues we should be worrying about now. This is because that type of AI (Artificial Intelligence) hasn’t even been invented yet. The type of AI that could end the world would most likely be general AI; which is the type of AI that you see in movies like Ex-Machina and Chappie, these movies consist of robots being created that act like humans with little difference between them and an actual human. General artificial intelligence is when a computer has intelligence that’s indistinguishable from a human, where they can learn things without someone coding it into to them. The more relevant artificial intelligence would be narrow. Narrow artificial intelligence is developed to be more geared towards a specific task. Examples of this would include google maps (specific towards directions), auto-correct (specific towards spelling), and siri (specific for answering questions). Rather than focusing on general artificial intelligence ending the world, we should be focusing on narrow AI. There is evidence of some narrow artificial intelligence being biased towards specific races and between sexes. An example of this would be when Chukwuemeka Afigbo, a black facebook employee tried to use a hand-sensing soap dispenser, but it wasn’t detecting his hand. He then put a white napkin and the machine detected it (Harriet Pavey 2017). This is only one case of artificial intelligence being racially biased. How do we avoid racial and sexist biases in artificial intelligence like these?

In order to answer this question we need to first answer what artificial intelligence is on a more technical scale. Artificial intelligence is a type of program that uses a large amount of data, then it analyzes the data and detects patterns, it uses these patterns to predict future patterns, and it also takes in new data to see where it was wrong and adjust it’s algorithm to be more correct, this is a process called machine learning or also known as deep learning. Why is this issue of racist and sexual biases in AI important? It’s important because we are currently moving towards a world where AI dominates everything we do. From what pops up when you search for something on Google, to a plane’s autopilot, these are all examples of narrow AI, AI is a huge part of everyone’s lives today. The inherent racism and sexism surprisingly affects a lot of artificial intelligence systems. According to AI can be Sexist and Racist — It’s Time to Make it Fair by James Zou and Londa Schiebinger, in 2017 research done to identify skin cancer in pictures using deep learning (a form of AI), had used pictures of people who were dark-skinned for only 5% of the data. Also the algorithm wasn’t used on dark skinned people. This led to the algorithm not being able to function as well when looking at dark skinned people when compared to light skinned people. Another instance of racially biased AI is when a website called beauty.ai hosted the first beauty contest to be judged by artificial intelligence. Out of 6,000 admissions, including people of all races, 44 people were chosen as winners, and “Out of 44 winners, nearly all were white, a handful were Asian, and only one had dark skin”(Sam Levin 2016). Both these cases show that the developers of the AI didn’t have those types of people in mind. Lastly in results from an algorithm meant to detect toxic language on twitter, it was found that the algorithm identified tweets from black people as toxic more so than white people’s. This was because black people’s tweets tend to contain more vulgar language than white people, but the tweets themselves weren’t actually toxic (Katyanna Quach 2019). So how does this happen?

Well to answer that we need to know what types of biases there are, according to Bias Amplification in Artificial Intelligence Systems by Kirsten Lloyd, there are 5 types of ways these artificial intelligence biases happen. There’s dataset bias, association bias, automation bias, interaction bias and confirmation bias. 

  1.  A dataset bias is when the data set is either too small or isn’t very diverse. An example of this would be, if you are trying to take a survey of commute times to Manhattan in NYC, but you only consider people who live in Queens, the data wouldn’t be very diverse because there’s other places in NYC, like the Bronx.
  2.  An association bias would be when an AI amplifies a cultural bias in larger data sets. An example of this would be one mentioned earlier, with the beauty contest judged by artificial intelligence. The AI projected the idea that lighter skin tones are more beautiful into the results. It failed to consider that for different cultures there are different definitions of beauty. In many East Asian countries, specifically in China, Japan, and Korea, having a lighter skin tone is more beautiful (Jin Hyun 2019). Whereas in areas like Africa people tend to have darker skin tones. This exemplifies that beauty doesn’t equate to having lighter skin tone because there are many different cultures and skin tones.
  3. An automation bias is when “…the AI fails to take social or cultural factors into consideration.” (2, Kirsten Lloyd 2018). This differs from association bias because where association bias takes small cultural biases and implements them to a larger data set, Automation bias doesn’t take a culture’s sensitivities into account when computing it’s result. An example of this would be when Microsoft created a chatbot that would respond like a human on twitter. The bot used artificial intelligence by analyzing people’s responses on twitter to respond on it’s own. Things took a turn for the worst when the bot learned from other twitter users and said “‘Bush did 9/11 and Hitler would have done a better job than the monkey we have now. Donald Trump is the only hope we’ve got,’ explained Tay in one tweet. ‘I f***ing hate feminists and they should all die and burn in hell,’ she added…”(James Zou and Londa Schiebinger 2018) The bot learned from other twitter users and started to act like them, without taking into consideration any cultural or social sensitivities we may have.
  4.  An interaction bias occurs when an AI doesn’t take into account cultural and social harmful beliefs and then learns based on these beliefs.
  5. The last one would be confirmation bias, which is when the AI makes “improper generalizations or assumptions about a group or individual.”(2, Kirsten Lloyd 2018).

In most cases of artificial intelligence being biased one of these biases are in effect.

 These biases could end up being more harmful than just beauty competitions or bots on twitter. What happens if you’re trying to get a job at google and they use artificial intelligence to screen resumes? Hypothetically the artificial intelligence could look through the resumes and deny people from Detroit because it notices that previous applicants from Detroit didn’t have the qualities that Google is looking for. As it so happens Detroit’s estimated black population is 79.1% as of 2018 (census.gov), so the artificial intelligence would be unintentionally denying black applicants, even if they have the correct credentials for the position. Something similar to this is already happening, a study in 2015 found that Google shows ads for high paying jobs more often to men than to women (Amit Datta, Michael Carl Tschantz, Anupam Datta 2014). Which could be a reason to why these biases are being applied to artificial intelligent technologies. Most jobs related to developing artificially intelligent technologies are considered to be high-paying jobs (ranging from $100,000 to just under a million dollars a year)(Andy Patrizio 2018), which would then mean that women could be excluded out of ads advertising artificial intelligence engineer positions. A report released by Google said that their workforce was 70% males and 61% white, which is terrible because “By 2044, people of color are expected to comprise the majority of the U.S. population, and women already make up nearly half of our country’s workers.”(Caroline Craig 2015). This might be the reason most of the people who are developing artificial intelligence aren’t developing it with people of color or women in mind because mostly white males are being employed to develop these technologies. This reveals the heart of the issue of AI biases, under representation in tech.

How do we fix this issue? Without first recognizing there is a problem, nothing can be done. So the first step would be to spread awareness that the tech field has an under representation of women and minorities. One way we could do this is by making companies more transparent about their workforce demographic, “tech companies need to work hard on transparency. This ranges from publishing compensation levels broken down by race and gender, to publishing harassment and discrimination transparency reports, to recruiting more widely than elite US universities and creating more opportunities for under-represented groups by creating new pathways for contractors, temporary staff and vendors to become full time employees.”(James Stanier 2019). Mentioned earlier, Google has already released their employee demographic along with other companies such as intel (46.2% white demographic) (Intel.com) and apple (50% white demographic) (Apple.com). With these huge tech companies releasing their demographics, it really shows how much of a problem there is in tech at the moment in terms of race and gender. More ways in which awareness of the problem could be spread by using mainstream media to report on the gender and minority gaps in tech, and by schools informing students about these disparities.

 The next step would be to actually get more minorities and women involved in tech, which could be done in many ways. One would be to give more scholarships to women and minorities focusing on a tech related field in their studies. This would provide a financial incentive for students to look into the tech field and make it more affordable for them. Scholarships such as the New Relic Scholarship which provides money to any underrepresented groups in computer science, provided by HackBright academy and Women in Tech Scholarship which provides money to women in full stack programs provided by Claim Academy (Erica Freedman 2019), are already doing this. This could be further improved by making these scholarships more well known through advertisement, schools spreading knowledge of the scholarships existence, and many other ways. One huge way to get more women and minorities into tech would be to show that other women and minorities are also in the tech field. This would provide a sense of relatability for anyone deterred from going into tech because they don’t see anyone like them in the field. According to Julie Ann Crommett, Google’s program manager for computer science education in media, “If we want to expand the employee pipeline, we must tackle this because girls who don’t see others like them in the field tend not to go into it,”(qtd from Caroline Craig 2015). Google and Disney have realized this and have put resources into creating characters that represent women in tech in order to popularize women and minorities in tech (Caroline Craig 2015). We could also showcase that there are actually women and minorities in the tech field, by popularizing women in lead technical roles, this would provide women and minority role models. For instance two minority women that are leaders in the tech field are, Mahashweta Das, a woman of Indian origin, is a Senior Research Scientist at Visa Research and Shivani Rao, a senior applied researcher a LinkedIn(“Real Life Role Models: 34 Women to Watch in Computer Science.” 2019).

 Providing more computing classes in schools that are historically for females and minorities would help spread awareness and interest in the field as well. This is exactly what Google is doing, “Google software engineers are not only teaching introductory classes at schools like Howard University, Hampton University, Fisk University, Spelman College, and Morehouse College” (Caroline Craig 2015) This would help get more minorities and women involved in tech. You could also popularize the tech field by making it more available in high schools, since most people figure out what their interests are in high school. The next way would be to use government regulation to make sure companies are creating artificial intelligence for everyone. Governments could implement laws to either ensure more representation in the people designing the artificial intelligence or for more representation in the test cases when testing the AI. In the United States efforts are already being made to get more minorities and women into tech by the Equal Employment Opportunity Commision and the Labor Department’s Office of Federal Contract Compliance Programs. However they aren’t making much progress (Joe Davidson 2019). Although companies mentioned above (Google, Apple, Intel) and other big tech companies, such as Facebook and Microsoft, have said they wanted to increase diversity in tech, not much progress has been made since their statements in 2014. Apple’s black employee population hasn’t changed since 2014. While the black demographic in Apple hasn’t changed, the demographic increased for women at Facebook by 8% from 2014 to 2019 (Sara Harrison 2019).

So to answer the question, how do we avoid racial and sexual biases in artificial intelligence? There are many ways to fix this problem, however the most effective would be to increase the diversity of employees in the tech field. Could you imagine a world where you try to use an app to detect health defects and it doesn’t work because the developers didn’t have you in mind? Or where you went to apply for a loan but you don’t qualify because an AI voids your application because you were a minority? That’s not a world that anyone should want to live in. Which is why we should try our hardest to implement some of these solutions to artificial intelligence.

Sources

  • Lloyd, Kirsten. “Bias Amplification in Artificial Intelligence Systems.” ArXiv.org (2018): 20. Web.
  • “Artificial intelligence fails to beat real stupidity; Racist ‘chatbot’.” Times [London, England], 25 Mar. 2016, p. 1. Gale Academic Onefile, https://link.gale.com/apps/doc/A447375584/AONE?u=cuny_ccny&sid=AONE&xid=6098af8c. Accessed 25 Oct. 2019.
  • Shank, Daniel B, and Alyssa Desanti. “Attributions of Morality and Mind to Artificial Intelligence after Real-world Moral Violations.” Computers in Human Behavior 86 (2018): 401-11. Web.
  • Craig, Caroline. “Wanted: Greater diversity in tech — but how?” InfoWorld.com, 8 May 2015. Gale General OneFile, https://link.gale.com/apps/doc/A481608578/ITOF?u=cuny_ccny&sid=ITOF&xid=41576e50. Accessed 25 Oct. 2019.
  • Stanier, James. “We Must Fix AI’s Diversity Problem.” Medium, Medium, 4 May 2019, medium.com/@jstanier/we-must-fix-ais-diversity-problem-6ad5fddc2f8c. Accessed 25 Oct. 2019.
  • Quach, Katyanna. “Oh Dear… AI Models Used to Flag Hate Speech Online Are, Er, Racist against Black People.” The Register® – Biting the Hand That Feeds IT, The Register, 11 Oct. 2019, www.theregister.co.uk/2019/10/11/ai_black_people/. Accessed 25 Oct. 2019.
  • Pavey, Harriet. “This Soap Dispenser Has Been Accused of Racism.” Evening Standard, Evening Standard, 18 Aug. 2017, www.standard.co.uk/news/world/automatic-soap-dispenser-sparks-racism-outrage-after-footage-shows-it-doesnt-work-for-darkskinned-a3615096.html. Accessed 25 Oct. 2019.
  • Davidson, Joe. “Perspective | Mostly White Male Tech Sector Needs Government Help on Diversity.” The Washington Post, WP Company, 31 Mar. 2019, www.washingtonpost.com/news/powerpost/wp/2017/12/04/tech-sector-needs-uncle-sams-help-on-diversity/.
  • “Real Life Role Models: 34 Women to Watch in Computer Science.” AccreditedSchoolsOnline.org, AccreditedSchoolsOnline.org, 30 Oct. 2019, www.accreditedschoolsonline.org/resources/women-computer-tech-role-models/.
  • Freedman, Erica. “Women in Tech: A Comprehensive Scholarship Guide.” The Best Coding Bootcamps, Switchup, 1 July 2019, www.switchup.org/blog/women-in-tech-a-comprehensive-scholarship-guide.
  • Datta, Amit, et al. “Automated Experiments on Ad Privacy Settings: A Tale of Opacity, Choice, and Discrimination.” ArXiv.org, Cornell University, 17 Mar. 2015, arxiv.org/abs/1408.6491.
  • Patrizio, Andy. “Artificial Intelligence Salaries: Paychecks Heading Skyward.” Datamation, Datamation, 28 Aug. 2018, www.datamation.com/artificial-intelligence/ai-salaries.html.
  • “U.S. Census Bureau QuickFacts: Detroit City, Michigan; United States.” Census Bureau QuickFacts, www.census.gov/quickfacts/fact/table/detroitcitymichigan,US/PST045218.
  • Hyun, Jin, and Carl Samson. “Why Do East Asians Want Pale Skin? It Has Nothing to Do with Western Beauty Standards.” NextShark, 13 Mar. 2019, nextshark.com/east-asian-pale-skin-beauty-standard/.
  • Levin, Sam. “A Beauty Contest Was Judged by AI and the Robots Didn’t like Dark Skin.” The Guardian, Guardian News and Media, 8 Sept. 2016, www.theguardian.com/technology/2016/sep/08/artificial-intelligence-beauty-contest-doesnt-like-black-people.
  • Zou, James, and Londa Schiebinger. “AI Can Be Sexist and Racist – It’s Time to Make It Fair.” Nature News, Nature Publishing Group, 18 July 2018, www.nature.com/articles/d41586-018-05707-8.
  • Intel. “Intel Global Diversity and Inclusion.” Intel, Intel, 10 Oct. 2018, www.intel.com/content/www/us/en/diversity/diversity-at-intel.html.
  • Apple. “Inclusion & Diversity.” Apple, Apple, Dec. 2018, www.apple.com/diversity/.
  • Harrison, Sara. “Five Years of Tech Diversity Reports-and Little Progress.” Wired, Conde Nast, 1 Oct. 2019, www.wired.com/story/five-years-tech-diversity-reports-little-progress/.