Recommender Systems reinforce human bias?

Answer

There are two questions here, and I’ll try to answer them separately. The first question is, “In what ways do you think Recommender Systems reinforce human bias?”

While reserching on these questions, I read several learned articles on the net, but none of the articles seem to have directly addressed the questions at hand. So, I’ll attempt to answer, but be very narrowly focussed on the question, without digressing.

Well, let’s first recap what recommnder system (RS) does. Based on our reading (namely Mining of Massive Datasets, chapter 9"* and https://www.youtube.com/watch?v=2uxXPzm-7FY), and some additional reading, we learn that when a user wants to consume a merchandise, the RS recommendeds a set of items and thereby offers a more choices, in case the buyer wants to consume them. It works through my mind’s system of choices.

What is so biasing about recommendation? Is RS promulgating any ideology, any political doctrine? No. Essentially RS is just presenting a set of “similar” merchandise and offering it for sale–albeit with the ulterior motive of profiteering–as any store salesman at Best Buy or Radioshack (say) would try to do. But, since RS is electronically pulling up its merchandise from its database, it has the advantage of pulling up a plethora of them, but probably matching my needs. But, by no stretch of imagination, could I call it an attempt to bias my mind, or reinforce my biases. If at all, it’s a highly technologically driven sales that’s trying to target me, using my own biases. That’s not reinforcement.

At this part of the century, it’s not necessary to be too original to say that human beings are biased–findings of Anthropology, Culture Studies, Biology would vindicate that. And and therefore I am biased–about my food habbits, my choice of friends, choice of music, subjects of interest and so forth. Based on my previous purchases, Amazon keeps a database from which its RS can guess my biases and offer smart choices. But, the ultimate decision to consume mine.

While that should answer the first question, I would linger on a bit. Suppose, a system tells me that a poll shows that such and such candidate is popular in my kind of demographics. Is that an attempt to bias a voter? Not any more than what used to happen when RS was not invented. The question is whether the system is telling the truth, whether its database is robust enough to make correct predictions. That’s always doubtful–as we saw in the 2016 election. A shrewd and insightful person would always be skeptical. And if may dare say, he/she should.

The second question is, “do you think recommender systems reinforce or help to prevent unethical targeting or customer segmentation?”

There are two issues here. One is the RS that we learned about, and the other is the RS that a business entity can cook up, for profiteering. Let’s call the former as the “Ideal RS” and the latter as the “Distorted RS”. The “Ideal RS” would iteratively collect data, predict, update itself with more data and predict and move on, with its never ending quest, as it were. It might also gather demographic data about a community, their consumer behavior and target them for sales. What’s so unethical about it? What’s unethical in selling spices to lovers of spicy food, or selling masks at the time of COVID19? If it’s not in violation of law, then the issue of ethics remains within the zone of the subjective, in my opinion.

However, as for the latter i.e. “Distorted RS”, I am not so certain if it at all qualifies as a true RS. Ideally, an RS is purely driven by data. If a business entity codes algorithms, with the deliberate motive to provide false impression or false information, then I would think that is unethical. But, whether its actually happening is a matter of detaied investigation, about which I have no data. Would any business entity confess that they use algorithms to mislead consumers? It’s very hard to prove that.

Marker: 612-03_d