So, AI can write a novel. Can it autocomplete my chores?

We don’t know exactly when human beings started speaking, but we do know that we’ve been writing things down, in one form or another, for thousands of years. At some point, though, it seems we got fed up. The human mind was not enough for us – we needed to find someone or something else to communicate with; something to take our words and spit them back in a new way.

As AI technology gets better and better at writing, why are we using it to generate tweets and short stories when we could be outsourcing the dull parts of our jobs?Credit:Matt Davidson (digital treatment)/AdobeStock

Recently, it has felt like text generated by artificial intelligence is everywhere. Programs like ChatGPT allow users to enter a question or a prompt and be met with rapidly generated content. Farewell speeches, the great Australian novel, even a sitcom scenario starring your favourite computer game characters – if you can imagine it, the program can deliver, though the quality may vary.

We did actually give the novel thing a go. Here’s what it came up with:

“The sun beat down on the dusty outback of Australia, baking the red earth and casting long shadows across the landscape. In the distance, a lone kangaroo hopped across the horizon, its powerful legs propelling it forward with graceful ease.”

So, based on that, it looks like our novelists might still be in work for at least a few more months.

Communicating with machines is nothing new. “The granddaddy is a program called ELIZA that was written back in the 1960s,” says Toby Walsh, professor of AI at the University of NSW.

Designed by Joseph Weizenbaum to act like a psychotherapist, “it was actually a very simple idea – it would just take whatever you said and turn it back into a question”.

Though ELIZA did manage to fool a lot of people, technology has improved since then – yet the recent leaps and bounds have taken Walsh, who has worked in the field for 40 years, by surprise.

The state of affairs

One of the most prominent text-generating programs is ChatGPT. “It’s autocomplete, like on your phone, but on steroids,” he says.

For autocomplete, “they’ve taken a dictionary of words and their frequency and trained it to what is the most probable letter and word that’s going to finish what you’ve typed. ChatGPT and GPT3 are very similar, but at a much greater scale – and they’ve not taken the dictionary, they’ve taken the whole internet. So it can complete not just the word, it can complete the sentence, the paragraph”.

Wielded correctly, using programs that can generate natural-sounding text should mean that we can skip over life’s boring bits. “We’ve always used tools to amplify ourselves. What’s new is that instead of having tools that can amplify our muscles, we’ve got tools now that can amplify our minds,” says Walsh.

How is it being used already?

When I told a colleague about this article, his first question was “but won’t students use it to cheat?”. The short answer is: yes, of course. The longer one is: they already are. With simply a prompt, these programs can generate convincing essays, rendering the last-minute all-nighter potentially a thing of the past. There are programs designed to detect this kind of cheating, by picking up regularities in the words used and the rhythms. “The problem here is it’s always an arms race,” says Walsh. “And you’re never ultimately going to win the arms race.” As soon as a checking program is developed, it’s used by the teams working on the text-generating programs to improve the text it puts out. And so on and on it goes.

But maybe this is an opportunity to improve things. “It’s about thinking well, do we have to stop asking people to produce essays that could be written convincingly enough by computers? Or should we get into something that’s more meaningful and more testing of their intellectual capabilities?”

Are the machines coming for our jobs?

In December 2022, business consultant Danny Richman tweeted about using GPT3 to help “a young lad with poor literacy skills who is starting a landscaping business”. The man would type in a simple string of text and the program would convert it into formal business language.

“That’s a wonderful example of how ChatGPT can be used in a positive sense. It’s not putting someone out of work, actually, it’s creating work,” says Walsh.

If we follow the likely trajectory of these programs, however, it seems probable that jobs will be lost. Much as work for typists dwindled when people started to write their own letters, as programs for AI-generated text improve and chatbots become more sophisticated, roles that rely on repetitive text-based tasks will likely disappear.

As AI gets better at taking on repetitive tasks, maybe we need to rethink what “work” looks like.Credit:Matt Davidson (digital treatment)/iStock

It’s a red light on the horizon for many people. Chatbots are already used in some customer-facing applications, something you’ll already know if you’ve ever experienced the infuriatingly polite loop that comes with trying to resolve a basic problem online outside of company hours. As the technology improves, they will be deployed more and more. Writers of all kinds – copywriters, creative writers, even journalists – will likely see these programs creeping into their fields, at least for the tasks that are repetitive and formulaic and don’t require human flair.

But maybe that’s a good thing. In theory, outsourcing work that is dull and repetitive – work that Walsh argues should never have been done by humans in the first place – should be cause for celebration; should mean that people are freed up to do other things in life, things they want to do.

There’s a tricky road ahead, which will require some shifting perspectives. “There’s no shortage of work, there may be a shortage of jobs,” he says. Looking after children and the elderly are key examples Walsh cites. “There’s lots of caring jobs that we don’t pay people for that we could afford if we had income being generated by machines.”

He also highlights the market for artisanal goods – that people will pay more for “things touched by the human hand”. Mass-produced goods are more affordable, but “we’ll pay extra to have handmade bread or artisan cheese”.

Alright, let’s outsource some drudge work to AI

Angus Thomson, one of our journalists, turned to ChatGPT to create a cover letter.

“I’m drawing towards the end of my 12-month traineeship … which means I will have to prove to senior editors I deserve a full-time contract. I asked ChatGPT to write a persuasive essay convincing Herald editor Bevan Shields to hire me as a full-time journalist, using examples of my published work,” he writes.

“The bot seems to have a good grasp of cover letter conventions, describing me as ‘a passionate and dedicated writer with a strong track record of producing high-quality, engaging content’. But when it comes to talking about my previous experience, it can get a little loose with the truth. It boasts (on my behalf) of working for the Herald, ABC News and interning at the Guardian. That all sounds great, except only one of them is true.

“Whether the AI software incorrectly scraped information from LinkedIn, or is outright lying through its virtual teeth, it seems it has a way to go before learning a key pillar of good journalism: telling the truth.”

With every update and improvement, it becomes harder to tell what is written by AI and what is written by humans. Credit:Matt Davidson (digital treatment)/AdobeStock

Truth be told…

How these programs handle truth is one of the major dangers of AI-generated text. “The fact that these programs will just invent stuff, and do it in a very convincing way, is problematic,” says Walsh. It’s fun when it’s a short story or a tongue-in-cheek attempt at a cover letter, but in the wrong hands can cause a lot of harm.

These programs work quickly and rely on prompts or questions. As a tool, they can be used to personalise phishing scams to increase the chance of someone clicking a link or giving out information, Walsh explains.

Then there is also the problem of fake news. Using these programs, it’s easy to fill websites with lies. “If you’ve got a favourite conspiracy theory, you can get to write some very plausible content and post lots of tweets to back up and support your conspiracy theory.”

Is it better at being us than we are?

Last week, I saw a post by Australian Twitter user Jamie Moffatt: “[the above] tweet was generated by training an AI on my tweets and getting it to suggest new ones in my style. jAImie is scarily funnier than me and I don’t know how to feel about that.” They’d used Tweethunter.io, which generated 10 different options – and the one they chose is hard to pull out as different from their regular posts.

Early on in our conversation, Walsh raised the idea of using AI to write letters of complaint. He then flagged that eventually there would be so many of these letters, a program would be developed on the other side to process them. We’d end up with computers just talking to other computers, with no actual need for the formal human language we’d originally programmed them to use. Where does that leave us? Standing politely at the sidelines while 1’s and 0’s are used to bicker over why the postman always shoves magazines in the letter slot, I suppose.

Our novelists are probably safe. While in time, AI could probably spit out a very entertaining formulaic story, Walsh thinks that, much as with art, they won’t be capable of engaging with deeper themes. “A machine is never going to have those human experiences, never going to talk to us in the way that a fellow human will talk to us in a novel.”

It seems strange that we have this technology and we are using it to generate things like creative texts or tweets, though perhaps that is just curiosity. In an ideal scenario – with many caveats – we use these tools to skip over the boring bits, and give ourselves more time to do the things we care about and want to do. But we have to always have an eye on the bigger picture, the natural endpoint.

“Like all technologies, especially artificial intelligence, there are positive and negative uses,” says Walsh. “There are useful things that we can do with these technologies. But equally, we should also be mindful of the risks that come with that.”

With Helen Pitt and Angus Thomson

A cultural guide to going out and loving your city. Sign up to our Culture Fix newsletter here.

Most Viewed in Culture

From our partners

Source: Read Full Article