Algorithms and Us


If you wonder how far machine learning has come, read these lines by Ben Evans:
“We used to call this ‘things that are easy for people but hard for computers’, but really, it was things that are easy for people to do but hard for people to describe to computers. The breakthrough of machine learning is that it gives us a way for the computer to work out the description.”

Today’s social networks are largely automated: algorithms decide what shows up on your feed, be it on Facebook or YouTube or Google Search. Almost always, no human has any role in deciding shows up on your feed. And therein lies the paradox:
“Yet they (the algorithms) are also totally dependent on human behavior, because what they’re really doing is observing, extracting and inferring things from what hundreds of millions or billions of people do.”

And that raises interesting (and uncomfortable) questions:
“What does this tell us about abuse of the platforms, and how much might machine learning change all of this?”
Who is to blame for what we dislike about social media? The people whose usage of it acts as input to the algorithms? Or the algorithm itself? Or both? And since the answer is almost certainly “both”, it opens the system to abuse from two ends:
“There are two ways to rob a bank - you can bypass the alarm and pick the lock on the safe, or you can con the manager. These are both ways that your processing systems are failing, but now one of the processing systems is us.”
And who analyzes the algorithm itself?
“Once you start using the computer to analyse the computer, you risk creating feedback cycles.”

So yes, algorithms are definitely part of the problem. But we the users are the other part. And while fixing both parts is necessary, fixing the algorithm needs a couple of companies to change. But fixing the users? Who’s going to do that?

Comments

Popular posts from this blog

Student of the Year

Why we Deceive Ourselves

Handling of the Satyam Scam