On Ethical Algorithms and Inclusion

Storytelling and technology continuously inform one another. As new technological advancements are made, new forms of storytelling are crafted. Even though narratives and its components have been the same since the beginning of time, the constraints that each technology brings creates a novel rule set which stories learn to conform to. These stories then permeate the socio-political and cultural land spheres explicitly, like how a feminist tale about surviving in a resource dry world bloodthirsty for gasoline and water is a way to motivate us to combat ecological collapse, and implicitly, like how software-sorted geographies marginalize specific populations via GIS based systems¹.

What does technology X mean for person Y?

The sharing economy may mean utilizing a resource collectively to benefit all to one group of people but may mean an uneven distribution of resources to those who have access to another. The same technological platform results in different meanings because of how trust of others is valued by people. And then these sentiments get accelerated and amplified by the underlying algorithms.

Algorithms are awesome. Because of algorithms I can quickly google and learn about apples, consume funny content on Facebook, and have peace of mind knowing an Amazon shipment will be at my doorstep in 3 minutes.

However, a filter bubble of only red apples is seen because of where I live, I feel an invasion of privacy as inaccurate targeted ads are shown to me based on what I click, and someone else in a different part of the same city doesn’t get the same day delivery as I do. Ethics gets sacrificed for efficiency sake because what is efficient saves costs and brings in the $$$. But can’t there be a balance between the two?

Here are 4 ways to combat the negative consequences of algorithms

  1. Fight Against Discrimination and Stereotyping
    Since algorithms are based on existing data, changing the data set to be more diverse will then result in more diverse outcomes. Fighting for civil rights is the most difficult yet important way here. Whether on a sociopolitical level or a personal one on one level, it takes a lot of effort to change how people have thought their entire lives. The impact when successful is life-changing though!
  2. Algorithm Transparency and Auditing
    On the flip side, changing the algorithm to be more fair to various populations is another method. Auditing algorithms to balance allocation levels the playing field. This can be done through adding variable weights and creating population-specific algorithms. Also, making algorithms transparent leads to accountability as companies can be more readily held responsible for their outcomes.
  3. Company Policy Changes
    Making a bold and actionable statement not only signifies steps in the right direction for a company, but also does for the tech field at large. Which is exactly what Airbnb did. In their report to fight against discrimination and build inclusion, Airbnb established a more strict nondiscrimination policy hosts must adhere to, a more diverse workplace as well as a dedicated team to fight bias and promote diversity, an acceleration of Instant Book which negates host bias of guests, an Open Doors initiative that guarantees lodging for guests who were discriminated against and more. Good job Airbnb!
  4. Design Opportunities
    The next step Airbnb should take is to manifest their policy changes into the design of their products. Which is what the next part’s about! 
     
    But before I get into my proposals, following these 4 ways isn’t just the ethical thing to do. They are lucrative for companies as they’ll be tapping into large existing markets.

Expanding Badges

Diverse representation tells members of those communities that they matter. The good thing is that the narratives surrounding these individuals is starting to take hold to financial, critical, and cultural success. There’s a reason why Frozen, a female sibling love story, is the 9th highest grossing movie of all time. Within the context of Airbnb, providing opportunities and encouraging inclusion to different markets tells those people that they are valued.

In 2015, Airbnb booked 80 million trips around the world. Since 70% of those trips occur outside of the US, there were roughly 24 million bookings in America. Nearly 700,000 of these bookings were taken by families. In other words, families in America encompassed almost 3% of total Airbnb trips in 2015. When paired with how there is little overlap between travelers booking at hotels and those with Airbnb, my guess is that families aren’t feeling valued by the platform. Furthermore, my gut says it’s because of worries about security and trusting strangers.

Badges!

With added badges, listings display if their housing and its amenities meets the needs of different populations. Guests are able to search by these different badges and see them on each listing. This reassures that different communities are valued in an up front and direct manner. Families are able to see family friendly houses with welcoming hosts, private entrances, and a pool for example. Just like Superhosts, to get a badge, hosts must meet several benchmarks as well as receive algorithmic validation through guest reviews, comments, and surveys.

Events

America has gotten increasingly polarized and this division is seen clearly in the rural and urban gap. Rural communities are feeling ostracized because of a lack of economic investment. The narrative matters here as well since the result ends up becoming the toxic political climate we currently are in.

Within the context of the sharing economy, Thebault-Spieker et al². have shown that there are significant systemic biases against low socio-economic and suburb areas. These areas are less effective because of distance decay, where interaction between places decreases as the distance between them increases, structured variation in population density, where higher density locations have higher access to resources, residential clustering, where people live close to others who are similar to them, and mental maps that people create of areas based on their comfort level.

Growing up in Binghamton NY, there are very few opportunities. Things are slowly improving with new initiatives for businesses as well as a new pharmacy school but ever since the decline of manufacturing industries, just like other small towns in America and in the Rust Belt, Binghamton has been surviving to say the least.

Airbnb says hosts in US urban areas have increased by 1,300 percent while those in US rural areas have increased by 1,800 percent. This is amazing yet the picture isn’t as clear as it seems. The data used to get these numbers is based on the 2010 US census definition of ‘rural’ and ‘urban’. Rural areas are those underneath a population of 2,500. Binghamton is by no means rural. Yet most people will colloquially refer to it as a town. Airbnb’s report categorizes urban clusters (2,500–50,000 people) alongside urbanized areas (50,000+ people). Thus not even considering small town America.

On the left, Urbanized Areas and Urban Clusters from the 2010 census (purple is urbanized area, green is urban cluster). On the right, Map of Rural Airbnb Listings and Hotels (blue-green is rural Airbnb listing, red is rural hotel).

When comparing these two maps of New York, the distinction becomes very clear. There are clusters of rural Airbnb listings at Ithaca, the Adirondacks, the Catskills, and the Hamptons. These areas are well known to New Yorkers for their beautiful gorges, mountains, and beaches. Also known as nature vacation areas. When you look at other areas of rural New York, also known as small town America, rural Airbnb listings become sparse. This made me 🤔.

Next I checked listings in various small towns in America and compared them to the number of hotels in each. Places like Binghamton, Horseheads, Oneonta, Geneseo etc. and smaller. Airbnb had small presences in each. But as soon as you get to one of those nature-y areas listed above, Airbnb’s presence gets large. I replicated both map comparisons and browsing in Pennsylvania, a state I am also well familiar with, and found the same discrepancy. A vast majority of rural listings in Airbnb go to nature based areas. Small town America has not yet embraced Airbnb as much as other non-urban areas.

However, when you drive down the Vestal Parkway in Binghamton, you’ll see hotels popping up left and right. A market does exist here. When thinking about why this is the case, the answer lies with the local university. Hotels keep springing up because of student move-in and events at Binghamton University. This is an incredibly wasteful allocation of resources as the hotels aren’t being used most of the time. This can be better utilized through smarter planning.

Events!

The Airbnb listings here shouldn’t focus on vacations but rather on events and happenings. To incentivize towns whose tourism market is event central rather than continuous, Airbnb can promote event based discounts. Additionally, specific rural-centered algorithms can be built as audits centered around areas that need that extra boost. For example, hosts and guests in stagnant Johnson City get better rates than more well of Vestal. Whether it’s the Spiedie Fest and Balloon Rally, the LUMA Projection Arts Festival, or student move-in, dynamically allocated guests will give capital to hosts who are buckling in the stalled local economy.

Hope this inspires algorithm designers to think about both the pros and cons when designing algorithms!

References

  1. Graham, S. (2005) ‘Software-sorted Geographies’, Progress in Human Geography. 29(5): 1–19
  2. Jacob Thebault-Spieker, Loren Terveen, and Brent Hecht. 2017. Toward a Geographic Understanding of the Sharing Economy: Systemic Biases in UberX and TaskRabbit. ACM Trans. Comput.-Hum. Interact. 24, 3, Article 21 (April 2017), 40 pages. DOI: https://doi.org/10.1145/3058499
    ________________________________________________________________
  3. https://fivethirtyeight.com/features/technology-is-biased-too-how-do-we-fix-it/
  4. Isaac Johnson, Connor McMahon, Johannes Schöning, and Brent Hecht. 2017. The Effect of Population and “Structural” Biases on Social Media-based Algorithms: A Case Study in Geolocation Inference Across the Urban-Rural Spectrum. In Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems (CHI ‘17). ACM, New York, NY, USA, 1167–1178. DOI: https://doi.org/10.1145/3025453.3026015
  5. Tawanna R. Dillahunt, Vaishnav Kameswaran, Linfeng Li, and Tanya Rosenblat. 2017. Uncovering the Values and Constraints of Real-time Ridesharing for Low-resource Populations. In Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems (CHI ‘17). ACM, New York, NY, USA, 2757–2769. DOI: https://doi.org/10.1145/3025453.3025470