The internet is leading the way.
There is a lot of AI-produced content already on the internet. Chatbots, algorithm-generated content, predictive text and customised feeds are de rigueur in big business. Now some brands — Northface, Hilton Hotels, Yahoo! Sports and Dominos Pizza for example — have curated company specific user-focused AI tools. Why? Because they save time, money and most importantly generate data. AI learns via data, and there’s already enough information on the web to deliver what browsers want or need to convert their curiosity into sales.
What kind of content can AI produce?
An AI tool can reproduce any number of musical notes, images, text and graphics, cartoons or avatars into a million different sequences, written or visual, static or animated. Say you want a banner advert prompting people to order your pizza placed on film-related websites. You feed the AI tool with the images, text and graphics you want and it will come up with several similar banner ads for your home delivery service. Instead of A/B testing, you can run one hundred banners and in 24 hours, know which one drives the most sales.
What are the practical limitations of AI written content?
According to Steven Pinker, humans are born with a grammar gene which allows us to understand how to arrange our thoughts into sentences without having to be taught the building blocks of language in a classroom or from our parents. Contrary to the hopes and dreams of SATS test developers everywhere, the logic of a phrase doesn’t always adhere to regular rules but is based on a set of contextual probabilities examined by the readers or listeners brain at lightning speed.
For example ‘Colourless green ideas sleep furiously.’ is a grammatically correct sentence that makes no contextual sense. Whereas ‘Buffalo buffalo Buffalo buffalo buffalo buffalo Buffalo buffalo’ makes little grammatical sense (it’s a noun string), but can be read and understood given context — if Buffalo is a place in New Jersey, and the verb to buffalo is used to mean intimidate, then (The) Buffalo buffalo (that) Buffalo buffalo (often) buffalo (in turn) buffalo (other) Buffalo buffalo’. This contradiction is how both legalese and the nonsense poetry of Edward Lear and Lewis Carol have become accepted and entrenched in our collective literary conscience.
So instead of strings of text in chains composed according to rules, AI tools need to predict their audiences set of experiences, understanding and education to pitch reading material at an engaging level. Although deep learning is a significant step for AI, machine learning automatically focuses on the most common instances, which is excellent if you want you writing to fit a particular mould, but not if you’re aiming to create innovative content. So far, no AI has been able to make something that is both original and makes sense. Although poetry, film scripts and a novel written by AI exist, and they’ve definitely excited the AI crowd, they’re hardly ready to be used for mainstream entertainment. Great writing can change views and behaviour. So while politicians use algorithms mined from social media to persuade and reach the voting public, we are not at the stage they would use AI-written communications, speeches and reports.
What written content can you do with AI?
AI can digest a vast amount of knowledge resulting in the production of multiple outcomes. So, materials like Adobes’ FrameMaker 2019 release can organise and memorise content which helps with translation, filtering details and multi-platform publishing. IBM’s Watson Real-Time Personalization (sic) can deliver personalised web-content for browsers based on their interactions with the site.
At the level of the individual, digital platforms like Grammarly can check any web-based text for basic written errors. It can even contextualise in a broad sense to either academic, business, technical, creative or casual domains. You can use re-writing tools like the Hemmingway App, which adapt your writing to the dogma of the lauded journalist and fiction writer, Earnest Hemmingway. Gmail predicts your generic replies in emails to try to shave response time. Even Macs have an inbuilt dictation function that uses voice recognising technology to help produce written content. At this point, however, none of these programmes can adhere to company style or brand voice guides, evolve to stylistic quirks and sometimes, they’re just plain incorrect.
AI – a utopia or dystopia?
If we are to believe Elon Musk, then the robots should be feared. Perhaps because he is mindful that the commercialisation of AI might make us dependent, as individuals, as a society, as states— whose power is often (especially in the West) related to productivity.
Who gets that power and what they do with it is what’s really at risk.
Much of the data fed into AI comes from one particular demographic — that of the computer engineer. Many of America’s computer engineers live in Silicone Valley and are white males. This has led to bias. For example, Apple came under fire in China when users found that face recognition software wasn’t responsive enough to tell them apart. Women have complained of not being able to set up or utilise their voice recognition software in cars because the data sets fed to the AI are mostly in male voices.
Alongside the political rhetoric of nationalism, there is fear growing that economic powers outside of the West will dominate the AI tech field. Although America is winning the race when it comes to claiming patents for AI technology, China has more engineers on the ground, better access to data (with a larger tech-using population), better government support and fewer business regulations. In 2018, 48 per cent of the venture capital money directed to AI went to China, whereas only 38 per cent went to the US.
But the real concern for both AI Superpowers is the loss of jobs. Automation in factories is a long-term trend, but robotics, drones and driverless vehicles will eventually replace human workers in some job sectors. However, PricewaterhouseCoopers argued ‘that AI would create more jobs (7.2m) than it displaced (7m) by boosting economic growth’.
Academic Nick Srnicek and writer, Alex Williams suggest that obstreperous nationalists needn’t be feared if the Left could change its ideology views on physical labour. In their book, Inventing the Future: Postcapitalism, a World Without Work, they suggest that the automation of labour will reduce the working week, diminish the presentism work ethic and still generate income. With the caveat of the state providing a basic income, they propound the need for a change in the political paradigm. For example, they suggest that Tube unions should embrace driverless cars, on the understanding that the benefits are spread evenly across the company hierarchy. The addition of a basic income would mean that traditionally unpaid labour (looking after relatives, running a household) would earn economic value. Essentially, we would all have less to worry about and more time to look after each other.
So, how will this affect the writing?
Either way, the robots are coming. Will they learn to be as dexterous with words as we are? Will the speed of their learning catch up with or maybe overtake our rate of language development and acquisition? Will the consumption of literature always be driven by imagination, emotive style and human connection? Who knows, but my prediction is this; humans will still need to drive strategy and review, if not produce, original content. Much like the cut-up technique pioneered by William S. Burroughs, humans will continue to commandeer the arrangement of the text, even if they don’t develop the words themselves. Gatekeepers involved in the proliferation of content, such as editors, marketers, or publishers, will still drive for either ideological spread or economical gain.
In the closer future, I predict there will be a split between the analogue and the digital. Avatar was a groundbreaking film in its use of CGI, and yet the desire for practical special effects has not diminished in the film industry as a whole. The music industry has weathered the digital revolution, with listeners tapping into on-line platforms but also harking back to physical records and going to live events: in 2018 vinyl record sales hit a 25-year high. Ebook sales are falling (by 10% last year) whereas independent publishers and bookshops are slowly increasing their market share.
This may be down to our desire for imperfection; the uncanny accuracy of digitalised content is unnerving. So once computer-generated content has grown in substantial use, people will feel suffused, attack its lack of humanity and seek out ink pens and papyrus scrolls in an attempt to find something tangible again.