This question is the topic of a recent article in Administrative Science Quarterly by Corentin Curchod, Gerardo Patriotta, Laurie Cohen, and Nicolas Neysen. They looked at eBay sellers, who
are a good example because they had algorithm bosses long before the Uber
drivers and Upwork programmers did. As you might imagine, they found that
algorithm bosses can be quite difficult.
The main problem with algorithm bosses is
their power. They make decisions based on rules and data, and no one can
question or appeal their decisions. For example, they can downgrade eBay sellers
or even exclude them from the platform if buyers’ feedback is poor. But what if
the data are manipulated? For eBay sellers, that is a serious problem because
buyers can rate the sellers, which means they can use the threat of low ratings
to negotiate better deals. There are ways to use the algorithm to detect
manipulative buyers, but these don’t work well because such buyers understand
the algorithm well enough to avoid detection, such as by changing identities.
That leaves sellers vulnerable because they can’t change identities, so they
are stuck with the buyers’ manipulation and the algorithm’s power.
From an eBay seller’s point of view, the
algorithm as a boss starts to look like a prison. The boss is always right,
even when the data it uses are wrong, and the data are often wrong because
buyers use the system to manipulate the seller. There is a way to appeal, but
the answers to appeals also sound like they come from an algorithm: eBay can’t monitor
buyers’ behaviors because it is a platform that helps buyers find sellers, not
more than that. It is not in the business of policing people.
So what to do when the boss is an
algorithm? The eBay sellers know the answer, and their approach will work for
others too. With human bosses, you manipulate emotions. With algorithm bosses,
you manipulate the data. eBay sellers with some experience are always on the
alert for buyers who look suspicious. Maybe their profile matches that of a known
manipulative buyer. Maybe the buyer is very recently registered, as would happen
if someone changes identities often. Maybe their behavior matches that of a
manipulative buyer – the tone or content of email messages before agreeing to
buy gives out warning signs. In each case, the response is the same: refuse to
sell to that buyer, avoiding the risk of buyer data manipulation. Naturally,
these refusals to sell are a form of seller data manipulation because they are
done to avoid low ratings.
Of course, eBay sellers protecting
themselves by excluding buyers is bad news for buyers who are innocent but
match the profile of manipulative buyers. That’s not news in management
research, because we know that workers who need to protect themselves against
overly powerful bosses end up harming someone or something else. Making the
boss non-human does not change the problem for the worker or their need to solve
it one way or the other. I find it interesting to see that one can take human
bosses out of organizations without also removing the problems that human
bosses cause.
Curchod, C., et al. 2019. "Working for an Algorithm: Power Asymmetries and Agency in Online Work Settings." Administrative Science Quarterly, forthcoming.
