While schools are banning ChatGPT, Australian workplaces are already deploying it and other new kinds of artificial intelligence (AI) tools.
For these tech companies, law firms and architecture practices, AI isn’t just the future of work — it’s already here.
Launched at the end of November last year, ChatGPT is known as a generative AI, or an algorithm that can be used to make content. It generates text responses to text prompts, like a very advanced chatbot.
Other generative AI tools make other kinds of content, like audio, images and video. Many have launched in just the past six months.
For schools, universities and other educational institutions, the recent emergence of these programs poses the risk that students will ask a machine to write their essays or complete their assignment questions, rather than doing the work themselves.
Some have decided to ban ChatGPT. But experts say this is the wrong approach, and students need to learn how to effectively use these tools in order to be employable graduates.
In the midst of this debate, Australian workplaces have been quietly experimenting with generative AI.
Many who spoke with the ABC say that within a few years — by the time today’s students graduate — these new tools will be essential parts of the white-collar workplace, as much a feature of the office as dusty keyboards, water coolers and small potted cacti.
Here’s how they’re using them.
The architecture practice
Architects at the large Melbourne design practice Gray Puksand use an AI tool named Midjourney, which creates complex images from textual descriptions, using similar technology to ChatGPT.
Images like this one:
Midjourney produced this from the text prompt: “low rise parametric building biophilic reference sheet 3d plans sections elevations construction details annotations callouts axis dimensions visual coding”.
Jeames Hanley, the practice’s digital technology manager, said dozens of architects were using the tool to quickly realise and explore visual ideas.
“We have a paid subscription to Midjourney — we have 40 staff using it in a shared thread,” he said.
“We are seeing the architects’ imaginations being expanded as they can so much more easily visualise the design they are thinking of.”
Asked to create a plan for a 100-person office space with desks, focus rooms, meetings rooms and board rooms, plus plants and lots of light, Midjourney produced these variations:
Although Midjourney is older than ChatGPT, it’s still relatively new. A San Francisco start-up of the same name launched it in mid-2022, and has been beta-testing and updating it since.
“I don’t think it’s going to replace anyone, but in terms of efficiency it’s going to change our industry,” Mr Hanley said.
“Some of our staff will be freed up to work more on the work that excites them, which is the design.”
The law firm
Last week, Sunshine Coast litigator Kyle Kimball turned to an AI program for help.
“People ring up and say, ‘this has happened. Have I got a case?’ I have to explain, ‘it’s not that simple. We don’t know all the evidence.'”
So he asked ChatGPT to write an explanation, using the analogy of a doctor ordering an MRI and blood tests to diagnose the patient.
The result was good enough to use, he said.
As an experiment, he also recently asked ChatGPT to draft a statement of claim based on the 1992 Mabo case, and was impressed.
“If I’d asked a first-year lawyer to do it, it wouldn’t have been anywhere as good,” he said.
Mr Kimball manages a firm of nine lawyers based on the Sunshine Coast and in Sydney.
ChatGPT’s work contained factual errors and had to be carefully checked, but then so did the work of junior lawyers, he said.
“Find me a person that you’re paying to do a job that does the job 100 per cent accurately all of the time, and in the time this machine does it in,” he said.
“I think it’s very exciting.”
At the law firm Piper Alderman, chief operating officer Chris McLean has also been experimenting with ChatGPT.
“There are lots of things in law it’s not anywhere near ready for,” he said.
For instance, it wasn’t ready to give complex advice to clients, he said, but might be used to write case summaries for lawyers.
“For high-volume things like drafting wills and conveyancing, it would be incredibly useful,” he said.
Belgian start-up Clausebase has already developed contract drafting software and has recently been beta-testing a ChatGPT-powered module.
The consensus of feedback was that the tool was useful, but imperfect, said Senne Mennes, the company’s co-founder.
Lawyers needed to do a lot of work to check and “clean up” the work, but ultimately saved time.
“One person phrased it very beautifully. He said, ‘I already have the chisel and GPT gives you the marble when you don’t have any.'”
The events platform
For the past couple weeks, Melbourne start-up Muso has been using ChatGPT to embellish the prose on its live-entertainment booking platform.
Every month, the platform hosts thousands of ads for gigs: for each, the venue must write a brief description of the event.
Often, this description was terrible, said Simon Raik-Allen, Muso’s chief technology officer.
“They write something like, ‘great cover guy coming.’ They haven’t got time to do much else,” he said.
The volume of bookings was so high that any human copy writer tasked with fixing the prose would have been immediately swamped.
Now, a ChatGPT-powered tool embedded within the booking platform generates the copy, based on whatever facts the venue chooses to provide.
From the input “weekly trivia at the Railway Hotel”, it produced this description:
“Come join us every week at the Railway Hotel for a Trivia Night! Put your knowledge to the test while sipping drinks and enjoying good company. Whether you’re a trivia veteran or a rookie, everyone is welcome. We’ll provide the questions, and you bring your sharpest wit and the most challenging answers. So come join us in a relaxing and friendly atmosphere, and see who can come out on top!”
It’s not poetry, but it’s good enough for an event listing.
“It saves [the venue] time and encourages more people to come to the gig,” Mr Raik-Allen said.
The software developer
In Melbourne, computer science student Riley Underwood has been using ChatGPT to write code and help engineer software.
Previously, when faced with a coding problem, he’d go to Google and find links to potential solutions.
“Now I don’t go to Google. I’ll go straight to GPT,” he said.
Mr Underwood asks ChatGPT to write code, and uses its response as a guide.
“You have to have a bit of experience to say this is 90 per cent correct, and I have to make adjustments to it.”
But, as with drafting legal documents, this process of editing is still more efficient than writing code from scratch.
“I would say there’s probably a 30–40 per cent increase in productivity,” he said.
Mr Underwood and his business partner, Vance Chao, recently created a ChatGPT-powered tool that generates 250-word property listings from property facts for use in the real estate industry.
In some scenarios, ChatGPT can turn half a day of coding into two minutes’ work, Mr Chao said.
“It’s a lot easier to read something that someone’s given you and say it’s wrong or say it’s right than it is come up with it by yourself,” he said.
Steve Nouri, a Sydney data scientist whose titles include “chief data evangelist” for an Israeli tech company, broadly agreed with this assessment.
“I’m understanding the limitations before introducing it to a pipeline of software development,” he said.
But he noted that many “junior developers” were using it already.
“It’s 100 per cent quicker than writing code from scratch.”
The creep of AI into everyday products
For the moment, generative AI tools tend to be restricted to discrete products, like ChatGPT or Midjourney.
But increasingly these products will merge with other pieces of existing technology, and be harder to disentangle.
Microsoft and OpenAI, the San Francisco-based maker of ChatGPT, are working on a ChatGPT-powered Bing search engine, while Google is reportedly looking to launch an AI search chatbot.
Meanwhile, use of generative AI tools will probably become more expensive. For the moment, ChatGPT can be used for free, but OpenAI has plans to make users pay.
The present may turn out to be a golden era of free access to powerful generative AI, followed by a period in which start-ups try to monetise their products.
“I reckon that once they get out of trial, they’re going to radically jack up the price,” said Muso’s Simon Raik-Allen.
“It’ll be a huge money spinner and probably won’t be available to every regular person because of the cost.”
Even so, given the enthusiasm of its early adopters, and the apparent boost to productivity, it’s easy to imagine workplaces will pay for access.
And if they do, they’ll be asking for graduates that know how to use these tools.
At Gray Puksand architects in Melbourne, Jeames Hanley repeated a quote he’d read somewhere on Twitter.
“It goes along the lines of: We won’t be replaced by AI, but we will be replaced by people who use AI.”