A Weather Risk Simulator… for ALL!
Abstract
For this project, we created an agent that simulates the human decision making process as to whether they should bring an umbrella outside depending on the chance of precipitation. For this project, we used the ACT-R cognitive architecture to create an agent that considers a numerical risk factor (chance of precipitation) and based on that decides whether the user should bring an umbrella outside.
Sounds great, right? To bring an umbrella or to not bring an umbrella — seems like a simple idea. However, no matter what technological artifact is created, it will inevitably have political qualities [9]. It is up to the creators to be able to recognize these qualities to make sure they are minimally harmful, and to be transparent about these qualities to its users [6, 8]. Lets dive a little bit deeper into how this agent’s decision making would differ if it was in certain demographic groups, so we can discuss how our simple weather risk simulator faces a potential race, gender, and class bias, if it is not implemented correctly. We must consider that a person’s desire to bring an umbrella outside depending on the chance of rain will vary depending on which demographic group the person belongs to. In general, people who are dependent on walking to places and people with non-straight hair are going to be more negatively affected by being in the rain with no umbrella. Therefore, they will be more likely to decide to bring an umbrella even if the risk (chance of precipitation) is low.
The Potential Biases
Getting caught in the rain with no umbrella is not a fun situation to be in. Although seemingly simple, this situation can carry more weight to certain demographic groups than others. A person’s race, gender, and income all affect how much being in this situation can affect them. These demographic groups, however, should not be considered individually. Evaluating biases of an artifact is only accurate and useful if we examine intersectional demographic groups [3].
Of course, the horror of being caught in the rain is the fact that you, your hair, and your clothes get sopping wet. However, this horror gets a whole lot more horrific for those with non-straight hair. More specifically, rain can be disastrous to Black hair, whether it is natural or not. If for some reason this is not enough to convince you that this is a racial bias issue within our agent, in 2019, New York expanded racial bias laws to include hairstyles and hair texture [7]. Even just a short shower of rain can cause Black hair to shrink, tangle, frizz, and can even wash away product that can cause stains on clothing [5]. Not only is it more easily damaged, but just one small rain shower can cost a lot of money. See Figure 1 for a visual of how much more Black consumers spend in the Hair and Beauty Industry (it’s 9x more!!).
Whether natural or straight, taking care of Black hair is both timely and costly. In 2018 alone, the Black hair care industry was valued at $2.51 billion, with the majority of that money being spent by women [10]. When Black women straighten their hair, it is either done with a straightener or a relaxer, both of which take up a lot of time (and money if it is done at a salon), and both of which will immediately make the hair curly again when it gets wet [2]. Any sort of damage to the hair or hairstyle, especially from an annoying little rain shower is a huge deal. Black people, specifically Black females will be much less willing to get caught in the rain due to how great of an inconvenience it can be to them. They therefore will be more likely to opt to take the umbrella with them if there is a chance of precipitation, no matter how low.
In another way, people who do not have the luxury of taking their own car from their garage to work every day also have to be more cautious about dressing properly for the weather. Figure 1 displays the percentage of adults who rely on public transportation on a daily basis. Many of which walk to their transportation or have to wait outside for the public transportation to arrive [1].
This figure shows that the majority of people who use public transportation on a daily basis are Black/Hispanic people as well as people with incomes that are <$30k. This, in turn, means that people who are Black, Hispanic, or have an income that is <$30k will be more likely to bring an umbrella despite a lower chance of precipitation. In order for our model to accurately simulate a human’s decision making process, we must take into account these groups of people who have to be more cautious when there is a chance of rain.
Our Solution
After researching the potential biases our artifact may have on certain groups, we decided to hard code a solution into our program. In order for our agent to learn from wrongly made decisions about bringing an umbrella, we used a point system as a reward value. When the agent brings an umbrella, but it doesn’t rain, the agent is deducted 10 points of their reward value because this is only a minor inconvenience. When the agent doesn’t bring an umbrella, but it does rain, the agent is deducted 100 points of their reward value in order to reflect how much of an inconvenience that situation would be for some demographic groups. We did this so that our agent would be swayed to want to bring an umbrella over not bringing one, just as an actual human who is negatively affected by rain would be.
Additionally, we the creators wanted to make sure we are following ACM’s Code of Ethics (specifically 1.1–1.4) in our agent [8]. So, we made our program come with a model card, which allows us to be totally transparent about how we created our artifact, its intentions, and its potential harms [6, 8]. This way, before even using our agent, a person is able to know about everything going on behind the scenes of our artifact, and they have the freedom to choose from there whether they will continue with relying on it.
The Results
After implementing the new point deduction for our agent’s reward value, we plotted the results. Figure 3 shows a dot plot graph of the model’s decision making, where “1” represents bringing an umbrella and “0” represents leaving it at home. Here, the blue indicates a correct decision, and the red indicates an incorrect decision.
As a result of our chosen point deduction in the agent’s reward value, we were happy to see that the agent tends to bring an umbrella over not bringing one when there is a chance of precipitation — just as we wanted it to!
Conclusion
In the end, we were content to see that our results came out the way we hoped they would. We did our honest best to try to minimize our artifact’s potential harm to Black people, Black females, Hispanic people, and people with incomes <$30k by researching how rain can affect them, and updating our program accordingly. This along with the model card’s transparency will allow users to decide for themselves how much they want to rely on our artifact. Although a human decision making simulation is not one size fits all, our weather risk simulator does its job as well as it can for the general population by taking into account the demographic groups that may be more negatively impacted. Thank you for reading!
References
[1] Anderson, M. (2016, April 7). Who relies on public transit in the U.S. Pew Research Center. https://www.pewresearch.org/fact-tank/2016/04/07/who-relies-on-public-transit-in-the-u-s/
[2] Antonia. (2013). Black Hair FAQs (10 Things Non-Blacks Want to Know). Un-Ruly. https://un-ruly.com/black-hair-faqs/#:%7E:text=What%20happens%20to%20black%20hair,goes%20into%20straightening%20our%20hair
[3] Gender Shades. (2018). http://proceedings.mlr.press/v81/buolamwini18a/buolamwini18a.pdf
[4] Harmon, S. (2020, June 9). Black Consumers Spend Nine Times More In Hair & Beauty: Report. Hype Hair. https://www.hypehair.com/86642/black-consumers-continue-to-spend-nine-times-more-in-beauty-report/
[5] Jackson, J., & Jackson, J. (2018, February 15). 5 Reasons Why Curly Women Still Run From The Rain. Curls Understood. https://curlsunderstood.com/rain-and-natural-hair/
[6] Margaret Mitchell. (2019). Model Cards for Model Reporting. https://arxiv.org/abs/1810.03993
[7] Press. (2019). New York expands racial bias law to include hairstyles, traits. WRGZ. https://www.wgrz.com/article/news/new-york-expands-racial-bias-law-to-include-hairstyles-traits/71-28aa2b15-7ada-4c3e-9c00-81866fda6f49
[8] The Code affirms an obligation of computing professionals to use their skills for the benefit of society. (2021). Acm.Org. https://www.acm.org/code-of-ethics
[9] Winner, L. (1980). Do artifacts have politics? MIT Press.
[10] Z. E. Holmes (2019, October 22). The Business of Black Beauty. Essence. https://www.essence.com/news/money-career/business-black-beauty/
