Making a Thing for People to Use

At Joel‘s patient urging, I finally got around to reading Will Oremus’s piece about Facebook’s news feed algorithm in Slate. I found it to be a really rewarding “long-read”. Here are my thoughts.


It’s always exciting to find faithful portraits of what I do for a living, since it can be difficult to describe to people. The art and science of it is basically:

  1. Put an initial stake in the ground in terms of the functionality we think is going to be useful.
  2. Gather data on the usage of that functionality.
  3. Interview real people about how they are using it and how they feel about it.
  4. Revisit the functionality with the data and human insights we’ve gleaned and improve it.
  5. Repeat.

user-behavior-venn

Numbers 2 & 3 above barely scratch the surface of a key tension in this work: What users do, what they want, what they think they want, and what they tell you they want are four different things. There’s some overlap, but not as much as one would hope or expect!


We need more solid, narrative descriptions of the limitations of machine learning (AI) at present. I find that non-technical people tend think machine learning can do both more and less than it currently can. The reality seems to be that in some ways our dystopian future is already upon us, but in other ways it’s further away than ever.


The article suggests a subtler point about making products that is generally overlooked: Luck and randomness are built into this process, but that doesn’t mean we shouldn’t approach it analytically.

Each person on a team like FB’s product team brings something different to the table, and each one of them has their own stuff going on that makes it harder or easier for them to hear their teammates’ views. The fact that feature development begins in such a chaotic place, but can end up with real, testable product recommendations that, once implemented, are frequently successful is amazing. It speaks to the fact that trying to do something methodically and scientifically, while never perfect, is better than the alternative.


If the Facebook employees featured in this article are a representative sample, Facebook is a very young, white, male place. That’s not entirely surprising, of course, but the fact that this cohort is setting the standard for technology of this kind globally is problematic.

Making a Thing for People to Use

Leave a Reply

Your email address will not be published.