[LINK] Australian Government Trial of Generative AI for Law, Education, Health, and Aged Care

[David]

On 8/03/2024 8:28 am, Tom Worthington wrote:
> On 7/3/24 12:53, Roger Clarke wrote:
>> Why not ask Copilot what could possibly go wrong with a 6-month trial of Copilot by 50 Australian Public Service (APS) agencies?
> I asked Copilot what could go wrong with a trial by the APS. It gave me a reworded version of the APS media release, saying how wonderful the results would be.

> [later...]  "In summary, while Generative AI offers immense potential, its implementation requires thoughtful planning, ongoing evaluation, and a commitment to responsible use within the APS ."
Would it make sense to ask Copilot which tasks AI would be best suited for in order to achieve budget savings responsibly?  That's possibly an unfair question but it would be interesting to see the result.

> But when I asked about "Generative AI", rather than "Copilot", I got a still mostly positive, but more more critical answer:
> [...]
> "Lack of Context Understanding:
> AI models lack true understanding of context and intent.
> Copilot might misinterpret nuances or fail to grasp the broader implications of its output.
> Human oversight is necessary to ensure context-aware communication."

Exactly.  AI (currently) detects language associations but without any understanding of meaning.  And that implies an inability to think creatively, or with insight, and (especially) with anything like empathy or sympathy.  It's interesting to see how often this simple point is not appreciated in public discussion.  Lack of understanding then leads to:
> "Legal and Accountability Issues:
> APS agencies must clarify accountability for AI-generated content.
> Who is responsible if an error occurs? How do we handle legal implications?"
Clearly not the AI system.  But if Agency Staff carry ultimate responsibility it's hard to see where any efficiencies and budgetary savings will come from.
> "These questions need clear answers."
You bet they do!

_David Lochrin_