Griffin on Tech: Aussie Government’s mixed Copilot results, tech titans get political
If you’ve been wondering how our bruised and battered public servants have been using AI, so have I.
Apart from the highly publicised pilot of GovGPT which appears to have avoided hallucinatory embarrassments thanks to Callaghan Innovation employing retrieval-augmented generation (RAG) to enhance its accuracy and reliability, AI use in the public sector is a bit of a black box.
The New Zealand Government is a Microsoft shop through and through - Callaghan Innovation was a rare holdout in the Google camp, but this year fell in line and began logging into Microsoft 365 and Teams just like everyone else.
I’ve heard from many friends who work in ministries that Microsoft 365 Copilot isn’t getting blanket use, but licences have been issued to people deemed to be able to make the best use of it. So are policy papers and press releases being drafted with Copilot? Are financial projections and budgets being used to analyse spreadsheets in Excel? Dunno.
But the Australian Digital Transformation Agency (DTA) has just finished an assessment of a 6-month pilot involving nearly 6,000 Copilot licences across 56 agencies of the Australian Government and came up with some interesting insights.
A survey of around 2,000 participants in the trial found that only around one-third of them used the AI assistant every day, but 69% of them found they were able to use Copilot to work more efficiently, saving up to one hour a day.
Summarising information seemed to be the most useful use case for Copilot, which isn’t surprising given the vast amounts of paperwork generated in government. Executives, mid-level managers and IT staff got the most value out of Copilot.
So far, so positive for Microsoft. But it wasn’t always smooth sailing when it came to adoption.
“There are key integration, data security and information management considerations agencies must consider prior to Copilot adoption, including scalability and performance of the GPT integration and understanding of the context of the large language model,” the DTA pointed out.
Access barriers prevented Copilot's use in Outlook, which is arguably where it could be put to best use helping bureaucrats deal with the blizzard of emails they are having to respond to on behalf of their recently farewelled colleagues.
Training in prompt engineering, establishing clear communication and policies around the security of Copilot, and “adaptive planning” to reflect the rolling feature release cycle of Copilot were needed. Cost was raised as an issue too, with the DTA advising that agencies would have to carefully consider the cost of rolling out tech that’s still in its infancy.
Still, the pilot generally makes a good case that AI assistants can make our public service more efficient and productive, if its users are properly trained and supported to use them. More than 80% of survey respondents indicated they’d like to keep using Copilot.
Musk, Bezos and an election hanging by a thread
Americans go to the polls next Tuesday US time as a particularly toxic presidential election cycle draws to a close. It seems like years ago that President Biden froze up and lost his way during the first presidential debate against Trump, and the former president narrowly avoided death at the point of an assassin’s bullet at a rally in Butler, Pennslyvania.
What would a Trump victory mean for tech-related policy in the US, and given the global reach of Big Tech, the world? Michigan State University’s Anjana Susarla does her best to unpack what Trump and Harris would each likely do in areas of tech such as AI, cryptocurrencies, data privacy and anti-trust action.
“Overall, the Biden administration’s efforts at antitrust and technology regulation seem broadly aligned with the goal of reining in technology companies and protecting consumers,” she writes.
“It’s also reimagining monopoly protections for the 21st century. This seems to be the chief difference between the two administrations.”
What’s been striking in this presidential campaign has been the influence of tech bros, chiefly the world’s richest man Elon Musk, who went all-in endorsing Trump after the assassination attempt. The conflicts of interest inherent in Musk having anything semi-official to do with Trump’s administration and his mooted Department of Government Efficiency (DOGE) are mind-boggling.
Then we have the Jeff Bezos (2nd or 3rd richest man on any given day) edict to the management of the Washington Post, the paper which he owns, not to publish their planned endorsement of Kamala Harris for president. That has, according to NPR, resulted in at least 200,000 WashPo readers cancelling their subscriptions, a staggering blow to the publication with the motto ‘Democracy Dies in the Dark’.
If Bezos had made that demand as part of his conditions for buying the paper in 2013, it would have been understandable - I’m a bit queasy about papers making political endorsements. But coming as it did days before the election, it reeked of fear of a Trump win and the consequences the Amazon could face.
The tech titans, with their massive political donations, lobbying power, and in the case of Musk, his staggering social media reach, have had more influence on the democratic process than ever before during this presidential election campaign.
We may find out next week what the implications of that are for an election that will likely be decided in a handful of swing states where thousands of votes in each state may make the difference.