Are algorithms going to choose our party leaders, too?

Let’s just hope those algorithms do a slightly better job of selecting our political leaders than our Spotify playlists (because otherwise… it’s going to be Nickelback, all the way…)

Interestingly enough, it was almost exactly a year ago – on July 20, 2020 – that MaltaToday published its now-infamous ‘Rate the Minister’ survey: you know, the one predicting how the Nationalist Party, under Adrian Delia, might lose the next election by ‘more than 72,000 votes’…

… thereby precipitated a full-blown crisis within the PN administration: which culminated, just two months later, in a change of party leadership, etc. etc.

Oh, OK: maybe there was a little more to it than just that one particular survey, alone. After all, it wasn’t exactly the first time that polls had painted a rather bleak picture of the Nationalist Party’s electoral prospects... and besides: the forces at work to overthrow Adrian Delia did not exactly begin in July 2020. (Indeed, you could almost argue that attempts had been made to ‘abort’ Delia’s leadership, long before it even had a chance to be ‘born’…)

But let’s not drag up ancient history again. The point is that, even if MaltaToday’s poll was not the only deciding factor … it did give the anti-Delia faction all the ammunition it needed, to justify what was (and still is) an absolutely unprecedented choice, for any Maltese political party to ever take.

I stand to be corrected, of course: but – at least, since Independence - I don’t think there’s ever been a single case, where the leader of a political party was ousted only midway through an electoral term (in this case, after just two years)… and even then, before that leader even had a chance to contest a single election.

But still: while it is undeniably tempting to play the ‘hindsight’ card today – in fact, it’s so tempting that I may even end up doing it unintentionally – the fact of the matter is that the circumstances faced by the PN were also quite ‘unprecedented’, at the time.

Again, I am limited only to what I can remember myself: which to be fair, excludes the whole Borg Olivier era… but as far as I can tell, the PN had never been quite so openly ‘at war with itself’, as it was by July last year.

And that works both ways. Not only was Adrian Delia’s authority internally undermined to a degree that none of his predecessors had ever quite experienced before… but his detractors also felt as though there had never been a more urgent cause, to take such drastic action in the first place.

As former PN president Mark Anthony Sammut – also an outspoken Delia critic - had put it to me in an interview, at the time: “we know that, without making this change, we will end up going into an election asking ourselves whether we will lose by 40, 50, 60 or 70,000 votes…”

But this, naturally, is also where it becomes increasingly difficult to avoid falling into that hindsight trap… even if, to be honest, you didn’t really need all that much, to predict at least a few of the possible consequences.

Like, for instance, that a Bernard Grech victory would come at the cost of most, if not all, of the Delia faction’s support… even just by virtue of having defeated (and, therefore, ‘humiliated’) their beloved leader…

Mathematically speaking: that can only ever ‘subtract from’ – as opposed to ‘add to’ - the PN’s already-depleted support base. And like all mathematical problems: the results are calculable from the outset (at least, if you know your arithmetic)…

All the same, however: the decision itself – i.e., to change party leader - remains understandable enough (even if just because ‘retaining Delia’, at the time, also meant simply resigning oneself to unavoidable electoral catastrophe.)

But… to justify such an incredibly dangerous political precedent – with all its foreseeable electoral consequences – only on the basis of a single newspaper survey?   

Far be it from me, of course, to discredit the accuracy our own surveys… or anyone else’s, for that matter… but again, you don’t exactly need to be Nostradamus, to figure out how something like that that might actually impact the future dice-roles of Fate.

For if a single, negative polling result – no matter how grievous – is suddenly enough to abruptly end a PN leader’s entire career: can we expect the same principle to apply to all future PN leaders equally?

Erm… hardly. Applied just to the present one: in theory, the PN’s Executive Council should really have re-convened at least three or four times, since Grech took over last September, to organise a new leadership contest.

But… it hasn’t happened, has it? And… well, let’s not even pretend to be surprised, either.

No, what really makes this precedent so dangerous, is precisely the importance that was given to surveys – not just ours last July; but all surveys, everywhere.

Again, I don’t wish to dispute the methodology of other people’s statistical research – still less, this newspaper’s foremost competitor– but just last weekend, The Times published a survey of its own: this time, predicting that the Nationalist Party stands to lose an election ‘by around 50,000 votes’.

By and large, that is pretty much in line with all other surveys published in recent months. As such, you could either argue that these latest figures mark an improvement of around 20,000 votes, since last year (and I imagine that’s the spin that today’s PN leadership will now put on it)…

…or else, you could point out that it still constitutes an even worse defeat than in either 2013 or 2017. Either way, it leaves the party itself stuck within roughly the same paradigm: i.e., “asking itself whether it will lose by 40, 50, 60 or 70,000 votes…”

But… how accurate is this latest prediction, anyway? (And I could ask the same about all the others, too… including our own survey last July).

And in this particular case: the answer also depends on how much trust you place in ‘artificial intelligence’. What makes this latest survey slightly different is, in fact, the methodology that was employed: which apparently uses digital technology – an algorithm, basically – to try and lower the traditional margin of error.

I won’t go into all the technical jargon, of course… but this is how it is described in the accompanying report:

“First, the pollsters carried out a survey of 600 people. But that first round of the survey gave Labour a much larger lead. […] However […] the survey scientists managed to bring the number of non-responses down [to just 2% from 24%] by building models of likely voting intentions based on respondents’ answers to several other questions. These answers were compared to the respondents’ declared vote in the 2017 election…”

In other words: stripped of all its algorithmic gloss, The Times’ original survey actually ‘gave Labour a much larger lead’ than the (reported) 50,000 margin of the final version. However, that same gap was later reduced…

… primarily, on the basis of how a computer algorithm interpreted all the ‘missing’ voter intentions; and even then… on how their responses to other questions compared with the situation as it stood in… um… 2017…

That is to say: on the eve of a very different election; held under very different political circumstances; at a time when both political parties were under very, VERY different leaderships…

… and above all, when all the factors influencing people’s voting intentions were totally (but TOTALLY) different from what they are today.

For instance: there was no ‘civil war’ raging in the corridors of the Stamperija back in 2017. Whatever his other flaws as PN leader: Simon Busuttil still presided over a largely unified party… that was still perfectly capable of, say, ‘filling the Fosos’ with a single mass-meeting (something that neither Adrian Delia, nor Bernard Grech, would even dream of attempting today…)

This also means that, whatever reasons people may have had for ‘voting Nationalist’ in 2017 – or, even more cogently, ‘NOT voting Nationalist’ – cannot even remotely be compared to their equivalents today.

Now: like I said earlier, I am hardly a world authority on ‘algorithms’, or ‘artificial intelligence’, or anything similar… but were any of those factors actually inputted into the computer’s learning mechanisms, before it decided to scale down the survey’s initial results?

And if not… how can we be so certain that the ‘new’ results are any more accurate than the ‘old’, anyway?

But in any case: in terms of how these considerations may actually impact the reliability of the figures themselves… well, it could go in two (at most, three) possible directions. Either the new methodology will indeed pay off; and henceforth, surveys will be slightly more accurate than they already are….

…or else, the predictions may end up slightly off the mark, in one direction or another. (It was the same with our pre-2017 election surveys, too, by the way: well within the margin of error… but still not 100% accurate. )

But that’s the full extent of what a survey – any survey, regardless of methodology – can be expected to achieve. (They are, after all, only designed to calculate probabilities… not to choose party leaders). If, on the other hand, we were to extend the same algorithmic approach to other, more sensitive decisions…

.. or, to put it as bluntly as possible: if this latest The Times survey were to have had the same effect today, as ours did last July… and actually precipitate another Pn leadership contest, to replace yet another, untested leader…

Ooh, I don’t know. Not only would it be the repetition of precisely the same, tragic mistake; but this time round, it would be the equivalent of dispensing with democratic structures altogether… and simply letting a computer programme take the decision for itself.

And…. well…. let’s just hope those algorithms do a slightly better job of selecting our political leaders, than our Spotify playlists. (Because otherwise… it’s going to be Nickelback, all the way…)