The interplay of AI, modern lives and literature
In September, I visited Berlin for the International Literature Festival, on a track titled Automatic Writing, which invited an international cohort of 30 science fiction authors, scientists, programmers, activists, philosophers and journalists to talk about Artificial Intelligence (AI). These cross-disciplinary conversations about AI have become increasingly common, and are currently one of the most exciting, but also widely misunderstood areas of public discourse.
AI is a familiar trope from some of our favourite science fiction. In those stories, AI is often a humanoid robot, engaging in interactions with people. Isaac Asimov published his fictitious, but widely quoted, Laws of Robotics in the 1942 story, Runaround. Alan Turing developed the Turing Test in 1950, involving a set of questions that distinguish machine intelligence from human intelligence. Do Androids Dream of Electric Sheep? by Philip K Dick, published in 1968, is a novel about the moral dilemma of treating human-like robots as fellow humans or less. The humanoid-AI trope has been overlaid with real-life social inequalities like gender, race and class, as well as philosophical discourse on the essence of being human.
But scientific research on AI had nearly dropped to a standstill by the late 1980s. Scientists were making machines do complicated tasks, but never beyond repetitive ones. Disparate items ranging from 3D graphics to mechanised sex dolls looked a little more realistic, but nowhere close to be mistaken for people. AI technology didn’t seem to be able to locate the breakthrough that would lead to those astonishingly human-like robots we were dreaming of decades before.
What revolutionised the field between then and now? In a turnaround few had anticipated, it was the Internet that rebirthed AI in the 21st century. Suddenly a word like algorithm, which I never imagined I’d encounter again after school, is back in our everyday vocabulary.
We’re all acquainted with this AI in our day-to-day lives. We exchange words with our Siri, Alexa or Google Assistant; buy new products from uncannily accurate personalised ads; let our phones autofill our words or Facebook recognise us and our friends in photos. While AI hasn’t turned out to “look like” us, now it knows more about us than even our real-life friends.
The human mind tends to impose humanoid models on everything. We imagine everything from gods to aliens to mermaids as variations of the human; we build cars with faces; we think our dogs are smiling. It’s hard to imagine disembodied networks of numbers, codes and abstractions, controlled by governments and large corporations, to be AI in the same way we had imagined robotic “people”.
The difference between good old-fashioned AI and modern AI is precisely the reimagination from the humanoid model to artificial neural networks.
Neural networks are also eventually a humanoid model, but instead of our appearance or social behaviours, they replicate the circuits of neurons in the human brain. The older generation of AI scientists failed to give humanoid robots the “spark of life” that would spur them to start learning independently; so, current scientists have gone deeper into our anatomy, emulating the way our brains process information at all.
In function, neural nets are sets of algorithms — mathematical formulae that can analyse previously existing data into patterns, and use those patterns to predict future outcomes. (that is, if a lot of people who enjoyed Book A also loved Book B, it’s likely that the next fan of Book A will enjoy Book B as well. Or when someone types p-h-e-n-o, they’re more likely to finish with –menon than –logy.)
The predictions made by algorithms become more accurate with the larger amount of data they can analyse. As highly advanced algorithms, neural nets cannot present tangible outcomes if their input data set isn’t sufficiently large. The concept of neural nets had existed since the 1940s, but it was only with the enormous amount of data freely available on the Internet in the 2000s that neural-net technology started showing outstanding results.
But here’s the catch: “Freely available data” on the Internet is often the personal communications of millions of people; users like you and me who didn’t even know that our data was being recorded and analysed by third parties. While scientific research has always been dictated by external factors like politics, marketability and so on, at no point in history has the research itself depended on turning millions of unaware people into research subjects. Neural nets are a delightfully efficient technology, but their biggest successes so far have all been opaque to the public, working for companies like Google, Amazon and Facebook.
So these days when we talk about AI, the conversation is not only of relevance to a small group of scientists and investors. AI is already inside our homes and lives, gently recommending the next thing we choose to do, eat, wear, read, watch, say or believe. There are new AI stories to be written, and my Smart Compose knows what happens next in them.