Some differences between human writing and AI writing

There’s been a lot of talk lately about artificial intelligence (AI) and its ability to write like a human. But what exactly are the differences between AI writing and human writing? Let’s take a closer look.

The history of AI writing goes back to the 1960s, when Joseph Weizenbaum developed the ELIA chatbot at the Massachusetts Institute of Technology (MIT). Since then, chatbots have continued to evolve, with Siri being created in 2010 and Facebook Messenger Bots following in 2016. In 2017, the Allen Institute for AI launched ELMo, a Large Language Model (LLM) that can deeply contextualize word representation. LLMs are autoregressive (any current ‘value’ is informed by the immediately preceding value), meaning they predict the next word based on the patterns of the previous text. And just one year ago, the research laboratory OpenAI – founded by Elon Musk – launched GPT-3, an LLM with more than 175 billion parameters, and training on all digitised text up to 2019, giving it immense predictive power..

The rapid progress of LLM capability has set alarm bells ringing among professional writers and commentators. The ludicrous mistakes and nonsensical sentences of the past are rapidly declining. In the words of Melanie Mitchell, the Davis Professor of Complexity at the Santa Fe Institute and author of “Artificial Intelligence: A Guide for Thinking Humans”, LLMs can now produce text that is ‘just astoundingly good’.

A Nobel-prize-winning machine?

LLMs are helping machines to get closer to being able to pass the Turing Test. First conceived by Alan Turing, the Turing Test is a way to determine if a machine’s natural language output is indistinguishable from a human’s. If a machine can fool a human into thinking that it is also human, then the machine is said to have passed the Turing Test. Some commentators even suggest they already have. But as we progress, inevitable questions begin to emerge:

Will copywriters and content writers lose their jobs?

Will this hamper human creativity?

Will the Nobel Committee one day consider a robot for the world’s foremost literature prize?

The short answer is not yet.

Consider for a moment the opening paragraph of the novel The Last Gift by Nobel Laureate Abdulrazak Gurnah:

Abdulrazak Gurnah – The Last Gift

Then compare it to what an LLM produced when asked to write the opening paragraph of a novel about an immigrant from East Africa living in England who reflects on his past after he has a stroke:

It was only after he had the stroke that he began to reflect on his past. He had never been a man to dwell on the past, but now, as he lay in his hospital bed, he couldn’t help but think about his journey from East Africa to England. He remembered the long journey by boat, and then the even longer journey by train and by foot. He remembered the cold, the hunger, and the fear. But he also remembered the hope. The hope that he would one day make a new life for himself and his family. And now, here he was, an old man, lying in a hospital bed in England. He had made it. He had built a new life for himself and his family. But now, as he looked back on his life, he realized that he had lost something along the way. He had lost his connection to his roots. He had become a man without a country, a man without a home.

Not bad, eh? But certainly not Nobel Prize-winning, probably not even publishable. Plausibly written by a human? Yes.

So, what are the differences between AI writing and human writing?

Consider again the above two opening paragraphs. The first sets the scene, generates excitement, and is permeated by the author’s life experience, literary style and influences. The second is grammatically perfect (perhaps a bit too much so) and logically coherent, but also shorn of creativity, with too many clichés.

This is the heart of the current difference between AI writing and human writing. LLMs are trained on every digitised word ever written (the internet – Common Crawl, Wikipedia, digital books, etc) – that is everything from Nobel Prize winning novels to the darker depths of Reddit. It is therefore hardly surprising that the aggregation of the totality of human knowledge, as expressed in language, can produce hackneyed prose with the underlying biases that permeate the culture of the English-speaking world.

AI’s role – what does an AI writer do better than a human?

There are many benefits to using AI writers, including the ability to produce content faster than human writers. AI writers can quickly generate large amounts of content, making them ideal for businesses that need to produce a lot of content quickly. In addition, AI writers can often produce content that is more accurate and consistent than human writers, making them ideal for businesses that need high-quality, accurate content.

An AI writer can also come up with ideas for writing that a human writer may not think of. AI writers have access to a much larger pool of information and can generate ideas based on this information.

LLMs as tools for humans

Despite its current perceived lack of emotion and flavour, there is clearly a role for AI writing – yet this is where the content and copywriter should rejoice rather than worry. LLMs are a tool for human use, rather than a replacement for human ingenuity. Recent improvements in LLM capability mean they are now great for producing short content, generating ideas, expanding those ideas into paragraphs, creating base text and placeholders for structure and flow and correcting grammar. The dreaded blank page will quickly become a thing of the past. The opening paragraph of this blog, and the section on what an AI writer does better than a human, were written using a LLM.

For business, the cost and time saved by AI written content will drive productivity – from generating ideas for tenders to producing first drafts of social media posts to summarising lengthy annual reports. In the hands of suitably skilled users, LLMs will speed up the iterative writing process, bringing skilled human writers quickly to a first draft before they inject their humanity into the AI-generated prose.

Let’s give an LLM the final word:

‘Human writing often contains more descriptive and figurative language than AI writing. This is because humans often rely on metaphors and similes to describe their experiences, while AI relies on more literal language. As a result, human writing can often be more flowery and poetic, while AI writing can be more direct and to the point. Human writing can be more changeable and fluid than AI writing. This is because human writers can more easily revise their work, making changes as they see fit’.