The latest in my series of FAQ posts follows on from the last one, in response to a question from an MA student at City University who posed the question “Do you think that an increase in algorithmic input is leading to a decline in human judgement?”. Here’s my response.
Does an increase in computation lead to a decline in human input?
Firstly, it’s important to emphasise that the vast majority of data journalism involves no algorithms or automation at all: it’s journalists making calculations, which historically they would have done manually.
You mention the possibility that “an increase in computation leads to a decline in human input”. An analogy would be to ask whether an increase in pencils leads to a decline in human input in art.
The computation is a vehicle for our input, not a replacement for it.
We don’t say to the computer ‘tell me a story’ and let it do all the work. We ask it to perform a series of operations: calculate a change, work out the percentage, order these numbers from lowest to highest, and so on.
When people talk about ‘technology’ it’s worth remembering that pencils are technologies, alphabets are technologies. Humans are characterised by their use of tools, and it’s just that after a certain time ‘technologies’ become invisible.
So I would counter with this: “an increase in computation allows for the magnification of human input“. It allows us to take an action and multiply it, or repeat it. But ultimately that action is ours: it is our input. It is computation that makes it possible to search through millions of documents or perform thousands of calculations — something that would have taken journalists months or years before.
Or another statement would be this: “an increase in computation allows for a reallocation of human input“. In other words, computation frees up time that would have been spent on repetitive activity, so that we can use that time to employ our ‘human input’ in other ways. For example, the invention of email means that we don’t have to spend time walking to the post box, so we can write more messages than before (take a look at your email inbox and tell me there’s less human input now than when people sent messages by post…)
Do you think that an increase in algorithms is leading to a decline in human judgement?
I certainly think there’s the potential for that, yes.
In essence an algorithm encodes someone’s judgement (rather than replacing it, initially).
So firstly that person has to think about the steps that they take in making that judgement, and things like how much any factor is weighted in that.
Once the algorithm has been created, the potential is that the person (or someone else using it) delegates their judgement to that algorithm — in other words, they delegate the judgement to their past self, or to their colleague.
There is a precedent for this. Historically journalists relied on interpretations of data by others: we delegated our own judgement to an ‘expert’. We reported the press release, or announcement, or quoted the researcher or statistician, rather than interrogating it ourselves.
So yes, having made so much progress in being able to exercise our own judgement about that data, we run the risk now of delegating that judgement again.
There are steps we can take to avoid that: we can review the algorithm regularly, for example, and indeed that’s quite a common practice: algorithms are often living documents that are constantly tweaked and updated. (I’m talking in very general terms here — again, algorithms are uncommon in journalism still).
You can also have quality checks to regularly check the output from algorithms.
Or — and this is the case in journalism — the results are checked manually by someone rather than being published directly by the algorithm itself.
By the way, one of the most interesting developments in data journalism has been ‘algorithmic accountability‘, in other words data journalists taking on a watchdog role to interrogate algorithms and subject them to scrutiny (see ProPublica’s Machine Bias series).
So one of the main roles of data journalism is actually holding algorithms to account. In a funny way, algorithms are making it possible to identify systemic problems in code, that previously existed in the culture of institutions and were harder to pin down.
So just to recap: algorithms encode human judgement and have the potential to facilitate a situation where judgement is not made contemporaneously.
When it comes to journalism, in reality the stories and data that we work with varies so much that there are not many opportunities for using algorithms in journalism. There are certain conditions that have to be met to justify the effort required: for example, it has to be similar data and similar stories, over and over again, which is why algorithms have mainly been used in areas such as robot journalism in sport, finance, weather, etc. and recommendation/targeting).
Or it has to be a journalistic problem that can only be solved effectively by an algorithm, rather than merely spreadsheet work* or calculations, and that relates to a big story.
When it does happen, data journalists often open up their algorithms for scrutiny from others (see BuzzFeed’s Tennis Racket investigation for example) and that should be expected.
*Strictly speaking a spreadsheet function like ‘SUM’ is an algorithm, but I am assuming that’s not what’s being referred to here.