The end of the year is a useful time to catch up on reading — with lots of time to even reflect on what you read. This December I was particularly struck by two data-points. The first is this chart shared by sociologist Philip Cohen that shows three decades of correlation between violent crime and single-mother families from 1960–1990 totally change from 1991 onwards.
The second is this paper from economists Amalia Miller and Carmit Segal that shows integrating more women into frontline police service led to greater reporting of gender-based violence (including domestic violence) and also “significant declines in rates of intimate partner homicide and non-fatal domestic abuse”.
We all know that correlation is not causation. That sometimes it can be sheer coincidence, that at the least causation can be messy with lots of other factors involved (and sometimes it may even be unclear what direction causality is running in).
Yet that doesn’t stop so many people from relying so much on classification engines that are really nothing other than correlation-finders. Machine-learning, artificial intelligence, human intuition — all of these are methods that look at existing data sets for patterns and fit those on to current real-time data and even make forward-looking projections. For example, we have social-network based credit-scoring and lending, reoffending rate estimators, music or book recommendation engines — the list goes on and on, driven by the ingenuity of people (and computation power) in interrogating data in more and more sophisticated ways.
And a lot of attention is finally being applied to the “hidden” biases that many of these engines are revealing. Some argue that these biases are wilful, while others note that perhaps this is just a case of data holding up a mirror we don’t like — that the biases that are being revealed so starkly are the ones that are indeed quantifiably there (but which until recently were only “alleged” by the minority groups affected).
You could also be a purist and say that to the extent that the predictive models show too much “bias” or too much “variance” (using machine-learning jargon) that’s just a sign of poor algorithmic design or weak “cross-validation” testing, or a need to “retrain” the algorithms. And others build on this by saying that algorithmic transparency would guard against this — although I wish them well when trying to justify an intuitive explanation for the number of hidden layers and activation nodes and weightings in a complex neural network that is making startling classifications that unsettle some parts of society.
But the bell that those two data-points kept ringing for me was that all this attention on faster, bigger, data-led classification was actually missing something even bigger. And that is human agency. The past is not the future, and things can change!
Those two data-points right at the top are compelling illustrations that human behaviour can change in profound ways. And importantly, that that change can happen at a person-to-person level, seemingly unnoticed in aggregate, and build like that into a tidal shift where long entrenched statistical patterns are dramatically altered.
To me, this type of tidal (as opposed to seismic) change seems to drive two important lessons. The first is that it happens best by people connecting with other people, by the norms and self-beliefs that people have for themselves and their communities being shaped and inspired (either for or against!) by those that they see around them. This can of course work for better or worse, but it shows that plasticity is there all around us, and within us.
The second is that these changes do not happen as inspirational gifts from visionary leaders on high. That this sort of emergent bootstrapped change — from one person to the next, at a peer to peer level — can and does happen organically. And that this organic shift in norms is what leads to tidal changes, not the inspirational diktats or seismic orations of charismatic ‘leaders’. (To be clear, I’m not saying that such leaders cannot influence things — sadly all too often they do, and in not very pleasant ways. My point is that change that relies on charisma does not seem to have the same longevity, or resilience, as change that has happened at a bootstrap level: Ireland’s referendum victory for gay marriage seems but one example of that?).
So far, so obvious — what’s the point I’m trying to make?
Its this. We need to get back to focusing on the human, and we need to support and respect their agency. Sometimes I’m struck by the tendency of self-proclaimed social change actors (“doers”, “funders” or “leaders”) to focus on the technology, on the scale, on whichever putative solution they are promoting.
But I don’t think you can appify or automate that precious human-to-human spark and impulse — that magical interaction that can prompt a change in a person’s self-belief.
Yet it still seems like too much energy, resource and attention is diverted on the technology, on the platform, on the classification of the people and the anointed entrepreneur-leaders. All of these seem to be basically different technologies of control.
Whereas what needs investing in, and nurturing, is the army of unsung change agents — the teachers, the nurses, the peer role models among many others. In a world where capital is so skewed in the hands of the few, is it not bad conscience to advocate for, or support, control-based models as opposed to those that nourish and leverage this human spark and relinquish agency back to where it has been overly depleted from?
When the British conquered India, they deliberately manipulated and then tried to ossify a previously contested social system into the rigid caste system. Under that people were classified, and then that classification was used to shape their future lives and opportunities. The agency of the many was taken away, and redistributed to the few.
We must guard against the same happening again now. Not just through an over-reliance on machine-learning or AI, but also more fundamentally through a depletion or degradation of human to human transmission of agency. The transmission of a spark that can lead to seemingly ‘heavy’ correlation curves fundamentally changing, as people change what they feel is or is not acceptable.
But this will require lots of people with money and managerial influence to really ground themselves in humility, and to relinquish agency to people they may not agree with. How many will have the courage to do that? But unless that happens, it will all still be different classification systems competing against each other — labelling people and preferences as “good” or “bad”, as “sensible” or “foolish”, as “progressive” or “bigoted”, as “leader” or “populace”. Classification should not be allowed to become destiny.
