Reply To: Facebook

I think the change that is happening is the type of code people are writing. It is less and less old style al-Khwarizmi ‘if this, then that’ algorithms, which as you rightly say can be fairly easily quality controlled.
We are seeing more and more machine learning. So sure, ‘politics’ is defined initially with a series of synonyms and related words which it is easy to control, but then the programme itself looks at a few million posts, and says ‘in that case, i think these posts are political – do you agree?’. It will start fine tuning the word list; adding and subtracting words and phrases, but also looking at patterns of usage that it sees which we didn’t define. It will change its view over time based on changes in usage. The purpose isn’t to get to a formal definition of ‘politics’, but to get to the point where a machine would make an assessment that couldn’t be distinguished from that of a human. To do that, it has to have advanced syntactical analysis. Bad meaning good etc. How it actually assesses a single post then becomes incredibly complex – there isn’t a simple piece of code that you can QA traditionally.
Does FB care about this? The only thing FB cares about is shareholder value. If it impacts advertising revenue, it cares about it HUGELY. The good news is that advertisers want brilliant algorithms. Disney needs to know on a town by town basis how adverts for its new ‘princess’ movie are being received, so real time it can adjust stock holdings in stores. That a 13 year old kid might tweet ‘it’s really bad’, Disney needs to know how to count that tweet accurately. And of course, a 13 year old tweeting that is highly likely to have a different meeting than a 63 year old tweeting the exact same words.
I don’t think we are doomed at all, but I do think control over data is changing. Leadership and high standards don’t change of course, but the ability of leadership to control standards is getting more and more complex. (And al-Khwarizmi is of course one of my heroes!)