3 days ago
NZ can't afford to be careless with its AI strategy
Opinion: The Government's new strategy for AI was announced last week to a justifiably flat reception.
As far as national-level policy goes, the document is severely lacking. One of the main culprits is prominently displayed at the end of Science, Innovation and Technology Minister Shane Reti's foreword: 'This document was written with the assistance of AI.' For those with some experience of AI, this language is generally recognised to be a precursor to fairly unexceptional outputs.
The minister's commitment to walking the talk on AI, as he says, could have been seen as admirable if the resulting output was not so clumsy, and did not carry so many of the hallmarks of AI-generated content.
To be blunt, the document is poorly written, badly structured, and under-researched. It cites eight documents in total, half of which are produced by industry – an amount of research suitable for a first year university student. It makes no effort to integrate arguments or sources critical of AI, nor does it provide any balanced assessment.
This same carelessness is exhibited in the web version of the document which has scarcely been edited, and includes a number of errors like 'gnerative AI' as opposed to generative AI. It also contains very little actual strategy or targets. It reads more like a dossier from Meta, Open AI or Anthropic and is filled with just as much industry language.
In short, it is entirely unsuitable to be the defining strategic document to guide New Zealand's engagement with what it accurately defines as 'one of the most significant technological opportunities of our time'. Especially not in a global climate where there is an ever-growing appreciation for the potential harms of AI, as seen in the growing number of class actions in the United States, or resources like the AI Incident Database. AI harm and job displacement are very real and important problems. Yet, in the Strategy for AI they are described as dystopian scenarios being used by the media to compound uncertainty.
The problem is not necessarily that AI was used to assist the production of the document, it is the extent to which it was used, and how. AI has a number of useful applications such as spellchecking, assisting with structure, and providing counter-points which can help further flesh out your writing. However, it is inappropriate to use generative AI to produce national-level policy.
What is particularly alarming is that anyone with a ChatGPT licence and about a minute of spare time could very easily produce a document similar in content, tone and structure to the government's strategy.
Thankfully bad policy can be improved, and hopefully this will be eventually. But, by far the most damning aspect of the strategy is the underlying notion that generative AI should have a key role in developing policy in New Zealand. There is an unappealing hubris in thinking that New Zealand's public servants, many of whom are phenomenally skilled, deeply caring, and out of work, could be replaced or meaningfully augmented by such a ham-handed and poorly thought out application of generative AI.
Unfortunately, it is likely that the strategy's fast and efficient rollout will be seen by the Government as a success regardless of the quality of the output. This will no doubt embolden it to continue to use generative AI as an aid in the production of policy in future. This is a real cause for concern, as it could be used to justify even more cuts to the public service and further undermine the function of our democracy.
Use of generative AI in the development of policy also raises fundamental questions as to what our public service is and should be. It would seem imprudent to employ our public servants on the basis of their care, knowledge, expertise and diligence and then require them to delegate their work to generative AI. A public service defined solely by the pace at which they can deliver, as opposed to the quality of that delivery, is at best antithetical to the goals of good government.
