Accounting for Social Outliers – Teaching Software to Think More Like a Human

Recently a story got some press after Facebook launched it’s “Year in Review” app where a gentleman received a painfully accurate reminder that his year had been tragic: featuring a photo of his daughter who died.

His suggested fixes were:

– “don’t pre-fill a picture until you’re sure the user actually wants to see pictures from their year.”
– “instead of pushing the app at people, maybe ask them if they’d like to try a preview—just a simple yes or no”

While there should almost always be a opt out where a user can choose what to see in their feed after the fact, a human focused solution when it comes to software isn’t one that belongs so much in this technology era.
Artificial intelligence (AI), with the sub-branches of natural language processing (NLP) and sentiment analysis, the ever growing body of data, and the tools to access and process it well, are providing better and better tools to create more sensitive filters.

Using the right software techniques, machine learning etc., we can start to account for the human aspect of interaction; where things are omitted, phrased differently, or cast in an otherwise appropriate light, and eliminate the most egregious instances.

This is a perfect use case for machine learning/neural networks: we point a learning algorithm at tragic news and have it extract common features.  And with the advances in machine vision, things like detecting hospital environments, individuals with oxygen tubes, and people generally not feeling well are becoming more feasible.
Not perfect, but enough to piece together workable solutions that will become better and better.

Sentiment analysis style filters would be just the solution for this type of situation.
We don’t want to have to dial back pushing forward projects; most of time just putting something out there and gathering feedback is a very useful product option. Opt-in should not start being the rule; instead we should build, run lots of tests, try to think of all the edge cases, implement filters as needed, and launch.

I do want to say kudos to him for his follow up thoughts:

Accounting for Social Outliers – Teaching Software to Think More Like a Human

Leave a Reply

Your email address will not be published. Required fields are marked *

Confirm you are human * Time limit is exhausted. Please reload the CAPTCHA.