700 words (8 minutes reading time) by Tim Whistler

In the world of artificial intelligence (AI), the concept of model collapse is a well-recognised challenge. Now it is potentially reaching into my everyday life if my Spotify-generated daylist starts to select songs from a previous daylist! Yet, my concern isn’t confined to personal playlists, and it should be echoing through the corridors of local government and raising questions about what performance measurement actually tells us.

Carole Parkinson’s question at the end of her footnote resonates with me: How can we detect model collapse in local government? It may well be a moot question, and model collapse is simply an explanation of the current situation, as one of Carole’s colleagues suggested.  This should become apparent if ChatGPT’s methods to identify model collapse in AI are accurate. Five seem particularly relevant for the local government context:

1. Assessing the Quality of Outputs

2. Variety of Outputs

3. Data Consistently Ignored by the Model

4. Cross-Validation

5. User Feedback

These methods revolve around performance measurement and business intelligence systems. A pivotal question comes to mind: Are councils capable of accurately measuring their performance over time in order to detect model collapse?

The answer, unfortunately, seems elusive. Despite the technological advancements of the 21st century, performance measurement in local government appears limited to responding to State directives, presenting a significant opportunity for improvement. Understanding current performance is crucial before initiating changes to ensure a grasp of both intended and unintended impacts.

Reflecting on the major reforms of the 1990s, interventions that hindsight suggests were triggered by State perception of model collapse, one can’t help but wonder about their effectiveness in improving council performance. Perhaps investigating this could be a worthwhile project for someone in the future?

Another way to detect model collapse is major failures. In Victorian local government, such failures trigger interventions by the Minister for Local Government, typically involving the appointment of Monitors, and/or dismissal of councils and the installation of Administrators. Conduct panels and investigations by the Local Government Inspectorate can be seen as other formal responses to failure. So what has been happening?

The Local Government Victoria (LGV) website reveals interventions at 16 out of the 79 councils since 2003. These interventions peaked between 2016 and 2020, with 8 instances, and have continued at a significant rate, with a further 6 instances since 2020. The reasons for intervention span CEO relationships, governance concerns, interpersonal issues among councillors, interactions with council staff, councillor role understanding, and occupational health and safety (OHS) concerns.

A pattern emerges when examining the types of councils involved. Rural councils have borne the brunt of interventions, with 4 regional cities, 4 large shires, and 3 small shires requiring intervention.  As FinPro has shown, financial sustainability pressures created by the rate cap, which has now been in place since 2016, disproportionately affect these councils. Urban councils, even though they are almost half the councils in the state, have had less interventions, with only 3 metropolitan and 2 interface councils requiring intervention.

The concerns identified by the Minister point to councillors under pressure, leading to conduct issues, blurred lines between strategic and operational decisions, and OHS concerns. Breakdowns in the relationship between councillors and the CEO are evident, with the division of responsibilities and the CEO’s control of council staff creating tensions.

The introduction of new powers for councillors and CEOs in the Local Government Act 2020 might contribute to this strain. The explicit delineation of roles between strategic and operational decision-making could still be undergoing refinement. The citizens’ advocacy group, Council Watch, has attributed some of the problems with the behaviour of councillors to their legislated roles and a power imbalance, which they believe needs to be reviewed.

Various factors, including the circumstances of the most recent council elections during the Covid era, add complexity to the situation. While some argue that campaigning difficulties and online inductions have prevented team building and strained councillor relationships, the rise in Ministerial interventions predates these challenges. Something deeper may be at play, prompting the question: What has changed across all councils over time?

The quest to unravel the mysteries of model collapse in local government continues. As we navigate this intricate landscape, the need for introspection, collaboration, and a renewed commitment to effective performance measurement becomes increasingly evident.

What do you think?