This is Letterfit’s featured article for autumn / winter 2018. Please don’t copy or re-distribute it without permission. You can comment by using this link – thanks for your interest.
‘…like a mannequin in sunglasses’ *
Python isn’t exactly like other programming languages – in its wider use or in its on-going use at Letterfit. It is an accessible way into working with basic levels of AI and Machine Learning.
Put more boringly (perhaps), it’s object-oriented and it’s useful for coding new typefaces, amongst other things. Its task-specific pre-coded strings accessed in data-processing libraries make it somewhat similar to jQuery, which means that leaner code generates more functionality – which, in many respects, can make you feel like a mannequin in sunglasses when you venture into its deeper levels of mind-boggling complexities.
‘The arts are neglected because they are based on perception, and perception is disdained because it is not assumed to involve thought.’
E H Gombrich, from Art & illusion, (2002)
‘Science manipulates things and gives up living in them.’
Maurice Merleau-Ponty, from L’oeil et l’esprit, (literally ‘The eye and the spirit’ or ‘Eye and mind’), first published in Art de Frana 1, no 1 (1961)
At Letterfit, we have great respect for art and artistic methods as well as science and scientific methods. Some of our clients have devoted their efforts and minds into making things work better within science – and clearly, science and technology aren’t the same thing. We believe in living with and owning all of this, with re-thinking and re-working the shared contexts of art, science & technology that we’re capable of contributing within.
What does this have to do with ‘tekoäly‘?
‘Tekoäly’ is Finnish for ‘Artificial Intelligence’. We first became aware of this language shift at Letterfit during 2013-18, when it was evident that some of AI’s automated algorithms, when attempting to auto-interpret human perceptions expressed as political views, slang, emotional outbursts, jokes and certain forms of emotional intelligence (the restraint, empathy and timing of ‘getting’ context & sensitivity within awareness and ethics) had gone unexpectedly awry. Digging further into this led us to tech-science articles about Finnish views of AI – and their views of ‘Havainto’ (‘Perception’).
We all need people like Jimmy McDonough (pictured above) to keep us in check. When we last visited him in 2018, we mentioned that our studio was interested in this. He said, ‘You’re not going to start wearing those weird round glasses, the kind they only make in Finland, are you?’
To which I responded, ‘No.’
‘Because that’s what happened to this group of design students I couldn’t stand. Their professor wore these round glasses that were only available in Helsinki. He had this annoying habit of half-smiling and staring off into the distance, like he knew something most of us didn’t. Some of his student-followers got the same stupid glasses – so they could wear these and walk around with the same half-grin, staring like Moomins, as if they knew something ironic none of us could possibly know.’ 1
Human perceptions : links – or barriers?
Though it’s a bit unfair to what’s mentioned above (about ‘Perception data‘) and to Jimmy, whose words can only be paraphrased – in spite of these and other practical concerns, we’re allowing ourselves to perceive AI as ‘tekoäly’, which (to us, at least) is less insurmountably complex and artificially superior than ‘AI’.2 Basically, we’re using Python programming to look for hopeful clues that AI can help us to work a bit smarter, not feel distanced or ill-fitting, in our adaption to it (since none of us are naturally-gifted mathematicians). Python as a programming language gets maths functions working for interests in language, the arts – and even our studio’s advocacy for multi-lingualism. This research employs Python’s sorting of natural language data-sets for ‘sentiment analysis’. 3
It’s difficult, if not impossible, to be a practitioner of art or design for long and not be aware that sentiment, perceptions, contexts and tastes affect meanings that go beyond what’s literal, factual or scientific. In short, each of us interprets some meanings intuitively. Less obvious is an awareness that the same words in their translated approximation of meaning(s) – apart from their visual style, treatment or association – can also be perceived very differently. 4
Near-human (rather than inhuman) perceptions
There are growing cultural expectations for AI aided by ML to better reflect the emotionally-aware and perceptually-intuitive responses of human variations. Whether or not this gap can be filled by ML remains to be fully realised in 2018. Regardless, what’s expressed in words by people (inarticulately, articulately or ambivalently) is unlikely to be free of unexpected shifts – or be emotionless.
‘Language is a body, a living creature… and within this creature’s home is the inarticulate as well as the articulate.’
John Berger, from ‘Confabulations‘, first published in 2016
Translating ‘a living creature’ is perceptual
Umberto Eco once remarked that ‘translation is the art of failure’. He compared the words traduttore, traditore (translator, traitor) to say that he thought he could only be, at best, regarded as an ‘admirable traitor’ for his translation work. In direct contrast to Eco’s views of these limits, The World Economic Forum has speculated about the possibility of Machine Learning translations opening up new possibilities for trade – in effect, lessening or ending the language barriers of the Tower of Babel analogy, once and for all.
These are intriguing times to live and work in – whatever you might think about all this.
We are admittedly designers more than coders, but this Python-enabled research will support an adventurous multi-platform publishing project that we’re completing. Coding in Python can be simple or complex, but its mix is proving to be accessible to us. Since this is being done to support an unpublished text, it wouldn’t be right to go into its premise or plot, but it can be said that perceptions, articulations and sentiment affect this short story’s outcome. 5
This text was authored by Jim Jackson, principal designer at Letterfit.
* Someone in the studio said this when asked how it felt to be coding in Python. We know we’re not alone in this (there’s a lot in the news about how popular Python is). We’re in it for the long haul – the project that this Python work will apply to will be going on for the next three years.
1 Thanks to Mr McDonough for allowing us to share this anecdote. It’s somewhat ironic that
‘Take care when gathering information. We live mainly on information. […] ‘
…were thoughts first published in 1647 (quoted here). Some of the functions we’re using aren’t that unknown or overly complex, either – some adaptations are partly based on the original Python-enabled functionality (now done with Amazon’s proprietary coding) that once scanned their databases to tell you ‘customers who bought this item looked at / bought these items.’ Who knew that would apply to narrative design?
2 We don’t assume to be data scientists, but we tend to agree that specific questions about language use and clearer meanings are for all to consider, within the means that are open to us. It’s been pointed out to us that perceptually replacing two longer words in English with one shorter word in Finnish is arbitrary and doesn’t change its meaning. It’s evident that the use of ‘teko-‘ in Finnish does signify something artificial or synthetic, the suffix ‘-äly’ means ‘intelligence’ or ‘capability’. Even so, perceptions of words aren’t made null and void by their arbitrariness (dog to chien, etc). Native Finnish speakers may or may not agree, but this less-cumbersome word diffuses some of what’s perceptually daunting about ‘Artificial Intelligence‘ to us. Some professional and pragmatic objections (‘don’t criticise AI, it’s the future, it helps me to pay the bills’) may be expressions of hyperbolic fears rather than language-based insights.
3 Diego Marani’s novel New Finnish Grammar is a literary exploration of the links between Finnish and Italian language sentiments. Python’s reliance on measurable values coded as integers or floating numbers to process natural language data means we have accept its limits alongside its advantages. This can help us interpret some general perceptions of AI and ML that are dependent on which datasets (and which languages) are accessed. We need to look beyond the marketing-based overstatements about AI and ML to see what can be done – after all the hard work of data cleaning and data processing are put in place – for machine learning to be implemented that could support these interests.
4 These factors are what makes human perceptions interesting rather than annoying to us: their changeable states. It’s impossible to exhaustively pin down human perception as a value-specific state (in precise terms that Machine Learning can algorithmically identify), since these perceptions vary and fluctuate during the same day and to the same person’s perceptions. This isn’t about rejecting AI’s ‘impressive micro-smarts’ that can be used to solve big problems, either (methodically narrowed down in ethical terms by us) faster and more accurately than any of us could do otherwise. This is about accepting that perceptions and emotions play into any human desires for clearer communication about what matters in all of this – meaning the changes in AI & ML applications that will happen in the near future.
5 We do take Python programming seriously at Letterfit, we’re committed to on-going learning and development to do this well. It’s good to lighten up enough and know that those we trust as friends, authors and colleagues and those who do work for the massive tech-giants have similar problems with getting words to translate universally. Using Python to code anything new and useful is more about showing than telling, so nothing more will post here (as a link to new work) till it’s functionally ready. ML research, like all good things, takes time.
Leterfit’s code-learning recommendations
CityLit’s Introduction to Machine Learning (well-organised & insightful, worth the cost)
Python for Data Science (a free beginner’s cheat sheet from DataCamp)
Tour of The Top 10 Algorithms for Machine Learning Newbies (authored by James Le on Medium)
Code academy’s intensive 7-week course for using algorithms (recommended by a knowledgeable & experienced colleague)
Direct link for downloading Anaconda navigator to install Jupyter notebook
Intro to Reinforcement Learning (authored by Thomas Simonini/Free Code Camp)
Seaborn.pydata (data visualisation tools)
Elite Data Science’s free intro tutorial to using Keras to build convolutional neural networks
The Open Graph protocol (useful for some Python-enabled visualisations)
Code mentor (one duplicate link on this list, but others that are useful, too)
Machine learning in Python (an introduction)
Machine learning mastery (a ‘what works’ free e-book)
List of of machine learning algorithms (New tech dojo)
Github for beginners (five years old, but still valid)