In 2014 I had already tried and failed at buidling a company aimed at providing psychographic targeting for the advertising industry. Later that year, thoigh, I arranged a dinner together with my friend Maria at her home to discuss an application of psychographics within the political domain. This was a new twist of the same old ideas in the main project I called Mindalyzer, that I’d been working on for many years. By the turn of events lately I now conclude that the science community involved in psychological stage theories must produce free annotated datasets and artist, intellectuals and politial activists in the metamodern movement are the most likely people to get it and put it to good use. Read on for the full story of how I reached this conclusion.
Maria and I invited eight people from our overlapping networks to that 2014 dinner either because they were, and still are, specialists in strategic and digital communications or engaged in civic initiatives involving psychological and existential health. All of us, I would say, could be described as concerned citizens. That evening turned out to be one of those really important moments of my life, even if it didn’t turn out the way I hoped. I remember I had overcome a lot of fear about putting my actual thoughts out there in front of people I knew were a lot smarter and more successful than myself. How we react differently upon fear is really one of the most interesting topics in life.
At the time of that important dinner I had just barely learned how to use Python and regular expressions to address the problem of psychological text analysis. I had performed an initial experiment looking for (by me) pre-defined keywords categorised by the six first stages of spiral dynamics in tweets. It was all very rudimentary and I could not say anything about any results, only that this could be developed technically by using more sophisticated methods. The main point of the evening was to throw out the question to this illustre group of minds and hearts – would it be a development project worthy to pursue together somehow?
After the dinner I held a short presentation, maybe 15 minutes and demoed the script I had written. I remember there were no real conversation about my question taking off at the dessert afterwards. But, a few of the guests looked very closely at the tiny word lists I had created for the presentation and then the evening took a surprising turn. Instead of a constructive discussion about the potential to further develop this embryo a couple of the guests held an very learned two hour long lecture mainly on why Spiral Dynamics is a defective theory and why linguistic analysis of cognitive stages would not work.
It was indeed a stunning and very informative talk, but I couldn’t help but feel that he just killed off my possibility to get the external help I knew I so badly needed in order to get something off the ground. I was pretty pissed off to be honest and felt they had stomped on a small plant before it had the opportunity to get some sunshine and water. Today I would call that attitude typical postmodernism – intellectually impressive but stifling for practical progress. Maybe that frustration fuelled me when wrestling with at least trying to test the idea myself.
”If modernism was the absolute belief in the building of Babel, and postmodernism was the realisation that it cannot be built, metamodernism can be tought of as the engagement in building it with the knowledge that it is a flawed or unachievable task”
— Alejandro Alvarez (@aleztriplea)
Maybe this new data science algorithm word2vec would help?
Fast forward to July 2016. I had learned some more programming and dipped my feet into data science and produced a word embedding, a statistical representation of how words and short phrases are used in relation to each other from a pretty impressive dataset of more than 200,000 mostly Swedish speaking Twitter accounts over a full year. Maybe, I thought, that word vectors could be used to create somewhat less subjective word lists for the different psychological models I was looking to analyse? So I set off, still on my own. I published a simple presentation of what I was about to do called Can word embeddings be used to improve psychological text analysis? Part 1 – setting up shop.
It was a painstaking work to select and test different seed words. In fact I believe that process took most of the actual cognitive time spent on this project. I had to go back to reading about spiral dynamics, and re-visit the book Spiral Dynamics: Mastering values, leadership and change, but also Igniting Inspiration: a change manual for cisionary leaders, both of which had set me off on this strange journey.
After about a year I could start to see some indications of where this project was going. I noticed that even though the word embeddings was somewhat helpful in choosing related words for the word lists, I had to manually toss a lot of obviously non-related words out, apart from manually selecting only the adjectives found in what the algorithm suggested as similar words. I called the selection process pruning, since it reminded me of the work of a gardener more than anything else. By pruning more and more words I felt like I was digging an ever deeper hole for myself. How could I ever ask external people to reproduce the pruning of words and come to any intersubjectively verifyable word lists? After more than a year of work I felt like I stood on so shaky scientific ground that I could just as well kill off the project. I ran the wordlists over a selection of twitter accounts and had a brief look for obvious correlations like a feeling personality and mentions of peoples names, but couldn’t find any.
My final conclusion was that the approach of using word lists was probably too simplistic. I suppose that having very limited time available and no external cheering on or expectations to motivate me, I had (and have) a hard time finding the energy to pull through over years at a time.
And so I buried the project and got on with life. I didn’t even publish the code, until tonight. Instead I got help in analysing the large Twitter dataset collected for this project from a much more scientifically rigorous perspective using network analysis and clustering techniques together with Mikael Huss. That felt like a fresh new beginning and really improved my learning in data science and statistics and also broke the long isolation I had felt. Our experiments turned into fun and play in the shape of the Swedish Metamemetic Society (formerly known as the Swedish Memetic Society). We also had the opportunity to share our datasets with students and researchers that might take on the quest, even though I could not find the time, energy or skills needed to actually do something proper out of it.
All of that was a lot of fun and has taken the cognitive surplus I have been able to muster apart from being a father, husband and spending time in nature since. However, the fact that postmodern critique hangs like the sword of damocles over the project and also the fun and games in the Swedish Metamemetic Society has decreased the level of fun just lately. Another factor is that eloquent diss of psychological text analysis in 2014 has accompanied me subconsciously and was awakened by meeting postmodernism again when relation to the Memetic Society and what had been going on there.
Some fresh air opening up for a possible future continuation
When I shut down most of my digital and massmedia intake in the beginning of this year to try to get some focus and new routines into my life, I got a lot of time to read books. So I finally read the book The Listening Society: A Metamodern Guide to Politics by the pseudonym Hanzi Freinacht that came out already in 2015. It turned out to be a very clear and learned (how often does that happen at the same time?) overview and critique of all the theories, models and thinkers that have guided me along all these years and who kind of ended up in the project to use word embeddings to analyse human language.
It was so much more succinct and even the idea to put this to use for political ends (in a positive and humane way, I’m sorry to have to add that after the well-known Cambridge Analytica scandal) was put into some kind of intellectually rigorous order. I highly recommend to read the book. Of course, only future will tell whether or not people will start using the ideas proposed in it for actual political activism and policy advocacy. Either way I think it is at least an inspiring vision of what society could look like and the first comprehensive formulation of the main actual problems of our time and also an convincing argument for why the current political system is unable to handle them.
Still, after reading the book I am not fully convinced that it is impossible to predict what Hanzi calls effective value memes (which is a re-worked model of spiral dynamics) from human language. I just can’t see how it should be done and how I could put any more effort into the project. I’ll just give it all a pause now and see where the further developments in psychological text analysis – heavily fuelled by the Cambridge Analytica scandal (evil and good has the same mother, as the saying goes) and the continual development in text analysis, which continues to advance and be let free by companies such as Google to the benefit of everyone interested in the analysis of human language. I’m prepared to leave all of this where it is and wait perhaps ten years so matters have time to unfold.
How and why it could be done from now on
I learned the most important lesson when analyzing Jungian cognitive functions from blog texts together with Mikael Huss; one needs annotated data to do machine learning experiments. So I hope that scholars understand the value of producing such datasets and releasing them into the public domain to enable the use of collective intelligence to solve the problems involved. I actually think the world might become a better place by people doing just that. Probably the smart and ethically concerned people in new political movements like the Swedish party Initiativet would be able to understand the whole thing and make good use of it.
An important factor when you do experiments like this is that you can’t be economically, prestigiously or otherwise invested in the results of an experiment like this. Therefore it’s hard to take the first probing steps. Therefore I don’t think it’s a good idea to do further experiments in the form of a business operation of any sorts. The risk would be to great that profit motives would influence the experiments or verification of results.
It could maybe be done within an academic setting in a true interdisciplinary and open-minded setting. If any academic funders could be convinced that the metamodern philosophical ideas promoted in The Listening Society are anything to take seriously and that continually updating machine learning methods could prove to be a nice merger of modern and postmodern methodology, then maybe. Because any of the approaches I can see looking forward will be dependent on lots and lots of annotated text samples that would need meticulously devised surveys about e.g. MHC stage of the writers.
Maybe in the civic sector, but is Hanzi credible enough?
But if Hanzi Freinacht’s ideas are not ready for academia, then it would probably be better to do it as privately funded research involving the same type of skilled researchers. Since the money would be coming from wealthy individuals, or maybe companies using it for CSR. I think that would probably be the most effective way.
The crux with such a continuation, is the same as when I tried to ask for help with this project – to find the right people with the right motivation for such an endeavour. And maybe it’s all just a pipe dream anyways. No results guaranteed and maybe there are philosophical reasons, as hinted but not clearly stated in the book, why this would never work. Another possibility of failure is that if one could find statistical relations between state, depth, stage and symbol stage that sums up in an approximation of the effective value meme – those relations might be as weak as the slight signals we did find when predicting Jungian type from text samples. Being so small it is hard to believe that they can in the end be very useful in making correlations to something else that would be more useful from e.g. a political or wellness perspective.
I really don’t know, but I believe this is bound to happen – if not else by the research frenzy that has taken of around the possibility of psychographic targeting being misused. Another friend, Jonas Andersson Schwartz pointed me to a well-researched article that shows that the alarmism around psychographic targeting is most likely unfounded.
So, there seems to be time and no need to panic around any of this. I think I’ll take a long and well-deserved break from all of this and enjoy the fact that I came all this way. Who knows what happens next? 😀