UN 205: Will GPT replace DTP? An AI derived bus network for Box Hill

An ABC item earlier this month says that AI-based large language models like Chat GPT could help plan cities. An Age article says that AI bots could help slim down the Victorian public service. That's topical given the state government is looking for savings with its Silver review due to report this month. 

So for fun I gave ChatGPT a task:  

Describe a reformed bus network for the Box Hill area with a list of main routes and the roads they travel on.

This is the response, pasted verbatim, including formatting. 

🚌 Proposed Key Bus Routes for Box Hill Area

---

Route 1: Box Hill – Deakin University – Burwood Station (University Link)

• Primary Purpose: High-frequency link to Deakin University

• Roads: Whitehorse Rd → Elgar Rd → Riversdale Rd → Station St (Burwood)

• Frequency: Every 10 min peak / 15 min off-peak

---

Route 2: Box Hill – Doncaster Shoppingtown via Station St

• Primary Purpose: Cross-suburban shopping and employment access

• Roads: Station St → Doncaster Rd

• Frequency: Every 10–15 min

---

Route 3: Box Hill – Blackburn – Forest Hill – Vermont South

• Primary Purpose: East–west connector through middle suburbs

• Roads: Whitehorse Rd → Blackburn Rd → Canterbury Rd → Burwood Hwy

• Frequency: Every 15 min

---

Route 4: Box Hill – Surrey Hills – Camberwell

• Primary Purpose: Local access and feeder to Camberwell Station

• Roads: Whitehorse Rd → Union Rd → Riversdale Rd

• Frequency: Every 20 min

---

Route 5: Box Hill – Mitcham – Ringwood

• Primary Purpose: East–west orbital route

• Roads: Whitehorse Rd → Springvale Rd → Maroondah Hwy

• Frequency: Every 15 min

---

Route 6: Box Hill – Donvale – Ringwood North

• Primary Purpose: North-east local connector

• Roads: Elgar Rd → Wetherby Rd → Mitcham Rd → Warrandyte Rd

• Frequency: Every 30 min (coverage route)

---

Route 7: Box Hill – Glen Waverley

• Primary Purpose: South-east orbital route, university + train link

• Roads: Elgar Rd → Highbury Rd → Springvale Rd → Kingsway (Glen Waverley)

• Frequency: Every 15 min

---

Route 8: Box Hill – Balwyn – Kew – Clifton Hill

• Primary Purpose: North–south route into inner suburbs

• Roads: Whitehorse Rd → Balwyn Rd → Harp Rd → High St

• Frequency: Every 20 min

---

🚏 Supporting Feeder & Local Routes

• Shorter, less frequent routes feeding into Box Hill and key train stations.

• Potential local shuttles to areas like Greensborough, Mont Albert North, Forest Hill Chase, etc.

---

🎯 Network Features

• Simplified, straight routes with fewer deviations.

• Clockface frequencies for ease of use.

• Improved weekend & evening services.

• Integration with trains and trams, especially at Box Hill, Glen Waverley, and Camberwell.

• Direct access to major trip generators like Deakin University, Box Hill Hospital, and shopping centres.

Chat GPT's map

You could request a map. The output gave me a Python script. I wanted to bypass this for an image I could present directly. That required me to log in to ChatGPT. 

The result was what you see below (click for a better view). There is no geographical accuracy. Eg Ringwood North is shown south of Camberwell and a long way from Ringwood while Glen Waverley is north-east of Doncaster. 

My map of 8 main routes

Using the written route descriptions I tried to make a map that was geographically accurate with regards to Box Hill, route termini and streets the routes would run along. That proved a challenge as some route descriptions lacked sufficient detail to make their paths unambiguous. My best attempt is below:  

As instructed the network is centred on Box Hill. Some routes are sensible. Others are sensible but could be longer to connect with other destinations (like existing routes currently do). Others you might not do.

Quick comments on each are as follows: 

1. Box Hill - Burwood station. Alignment is good between Box Hill and Deakin Uni, being similar to existing popular routes. However Burwood station on the quiet Alamein line is a weak terminus. Also this section of the route would overlap the longer and more frequent 75 tram so you wouldn't' do it. .  

2. Box Hill - Doncaster. An useful and direct connection this segment is currently served by the much longer 903 orbital. If it was linked to the Route 1 proposed above it would match a longer section of the 903. 

3. Box Hill - Vermont South. Has similarities with the existing 732 but is shorter and goes along more of Whitehorse Rd. 

4. Box Hill - Camberwell. A route with benefits for Surrey Hills but it inefficiently overlaps too much of the 70 tram on Riversdale Rd. 

5. Box Hill - Ringwood. Description here is ambiguous (with mention of Springvale Rd) but I've understood it to mean a direct Whitehorse Rd/Maroondah Hwy route. Not without merit despite it paralleling a rail line as there are a large number of destinations along the highway.  Existing routes like 271, 279 and 901 serve this section but not continuously. There may be a case to be splitting Route 901 at Ringwood and South Morang and rethinking what happens in between. That might include a Box Hill - Ringwood route on the highway. 

6. Box Hill - Ringwood North. The route description was not clear so what you see on the map is possibly not what was intended. This was described as a coverage route with a 30 min frequency (most of the others were every 15 min).

7. Box Hill - Glen Waverley. The alignment and the frequency is good in that, combined with Route 1, this would provide a 7.5 minute frequency between Box Hill station and Deakin University. The current network does not provide a direct Box Hill to Glen Waverley bus as one would not add a lot of new coverage. However it does link two major activity centres. This concept has similarities with a Liberal election promise in 2022 which may be where ChatGPT got the idea. 

8. Box Hill - Clifton Hill. Alignment again not clear, especially at Clifton Hill end. Parts have similarities with existing Route 302 (though this goes into the city). 

The above was the first network option that came up. Had I asked ChatGPT a different question or given it guidance the result will have been different. 

Conclusion

This (free) version of ChatGPT AI has a way to go when it comes to designing bus networks.

But more advanced paid types with more data input and other guidance may produce more sensible networks. 

Or, given the planning and scheduling software already used in the industry like HASTUS and Remix maybe an AI plug-in to enhance its planning function is how this will evolve in the future? 

What about other applications for AI in transport? AI can give reasonable answers for things like customer feedback. Maybe even journey planning. It might help with advocacy or submission writing, like I tried here.

However the AI you get when you do a Google search can mix things up. For instance when I did various searches on the Bus Plan the result conflated official and unofficial sources, such as from DTP and my blog. Personally I would love the day that a lazy official, unaware of this, produces a report that includes slabs of my blog as official policy. 

The old saying is 'Garbage In, Garbage Out'. Proper use of AI possibly means that its search be limited to official and verified sources rather than what just anyone says. Though DTP/PTV data can be wrong so an official source is not by itself a guarantee of correctness. AI risks just exacerbating existing errors. AI also needs a knowledgeable driver - that is people using it need to be learned enough that they can spot an error and correct.  

Comments on AI's potential role in transport are appreciated and can be left below. 

See other Useful Network items here

VAGO to audit Victoria's Bus Plan

The Victorian Auditor-General has included a performance audit of Victoria's bus planning in his 2025-26 annual plan tabled in state parliament last week. VAGO's review will examine the performance of Victoria's bus services, including if the Bus Plan is on track to achieve its intended objectives and targets.

Other transport topics to be examined include myki ticket modernisation, the progress of major projects, road maintenance and customer service of outsourced Vicroads functions like driver licensing. A follow-up of the audit on integrated transport planning is also pencilled in for 2027-28. 

An audit is timely

Back to buses. An auditor-general review of bus services is both desirable and due. One was proposed six years ago but got put off. However, with it being nearly 1500 days since Victoria's Bus Plan was announced sufficient time has elapsed for an auditor to form an opinion on its effectiveness or otherwise.

Not only that but the dollar amounts involved are, in auditor-speak, material. Victoria spends the better part of $1 billion each year to run bus services, mostly through payments to private bus operators. Unlike train and tram contracts (which appear to have got more generous over the last 20 years) payments to bus operators per service kilometre delivered appear to have been fairly constant relative to CPI.

Just because we're getting a reasonable amount of service kilometres per dollar for bus does not mean that these bus and driver resources are optimally deployed to be useful to the most number of people for the most number of trips. The key determinant of this is how well routes and timetables are planned and meet the public's travel needs. 

With routes and timetables for maybe two-thirds of Melbourne's bus network substantially unchanged for 15 - 40 years, network reform (the first priority of the Bus Plan) has a far greater bearing on the value we get from buses than certain other initiatives (eg fleet electrification that may have merit for other reasons). The 'health checks' I've done on the bus network are here and here, with only a slow rate of improvement between them. 

The extent and quality of bus network reform can be difficult to measure but is easy to obfuscate. This makes it vital for an independent party like the Auditor-General to properly examine the performance record here.  

link to count-up Bus Plan clock

Audit role

What might the audit cover? VAGO audits don't always cover everything some would like. It's worth taking a step back to understand why this is.   

Different independent institutions established by parliament to monitor the executive have different  roles and emphases as set out briefly below.

The deliberations of parliamentary committees like Parliament's Accounts and Estimates Committee are shaped by the politicians on them. The Independent Broad-based Anti-Corruption Commission (IBAC) is a specialist corruption investigator with quasi-judicial powers. The State Ombudsman is a 'last resort' complaints investigator, although they may also comment on policy, especially where it may be improved. Infrastructure Victoria is a major giver of advice on transport policy (including through the media where it is more open than DTP) but this does not extend to cases where the policy has been adopted by government. Thus you won't find IV commenting about the merits or otherwise of the Suburban Rail Loop (a project of this government). Indeed their reports tend to avoid even mentioning the SRL unless necessary. 

Then there's the Auditor-General, the oldest government accountability office dating way back to Victorian self-government in 1851. Being independent and being seen to be independent is central to how auditors general do their work. They are in the camp of institutions who do not analyse or comment on government policy. But they are very interested in it being well carried out.  

The pitfalls of those who are not ministers commenting on policy are explained here

 

The difference between the policy of administration (a political responsibility) and the administration of policy (a job for bureaucrats to do and auditors to review) is explained here

Potential audit topics and departmental responses

When asked about their record, the government and DTP can (and likely will) rattle off a long list of bus service upgrades that it attributes to Victoria's Bus Plan. They represent, with few if any exceptions, progress.

But something being good might not entirely satisfy a probing auditor if the methods used to determine priorities have not been robust. 

I have suggested that past departmental secretaries might have used counting methods of doubtful rigour to exaggerate their achievements in bus network reform when under scrutiny. Including from PAEC, which is the same committee the Auditor-General reports to. I would expect that the appropriateness of performance metrics would feature in a performance auditor's test program. 

An audit might also be interested in monitoring operational characteristics such as bus route patronage and acting on the data received to optimise resourcing. An examination will likely find that such data is collected (though one might query its accuracy). The department could give legitimate examples where resources have been transferred between routes (mostly involving Transdev/Kinetic routes) for an overall greater good.

Although they will probably cite new bus contracts as making this easier, DTP might find it harder to argue that this is being done in a particularly systematic way (eg tackling low productivity and/or duplicative routes first) even though the Bus Plan flagged resourcing/patronage productivity mismatches as an issue. 

Auditors (especially) hate things being funded without an implementation plan. Even if what gets funded is sensible, likely consistent with what a plan would recommend and proves successful in practice.

They're not fans of 'bait and switch' either. That is if a department would promise something (as being consistent with the plan), not continue with it but switch to doing something else. The party being audited may claim that that something else was consistent with the plan. Or they could just write a plan that is so vague that almost anything could count as being consistent with it, with the real detail coming later.  

DTP may have left itself exposed here as it:
(a) failed to produce the Bus Plan's promised Bus Reform Implementation Plan by its 2023 deadline (with a weak answer from then Secretary Younis when quizzed here),
(b) apparently stalled on the North, North-East and Mildura bus reviews promised before the 2022 state election, and
(c) ran trials that either had no apparent progress (eg Rapid Running being extended to more routes) or were reasonably foreseeable duds (eg FlexiRide) that distracted attention from beneficial bus network reform or service upgrades (such as Greensborough will finally get after flip-flopping on FlexiRide).   

Scale is important when planning transport in a big city. One or two little route upgrades might be successful but you need to replicate it across hundreds to have a metropolitan-wide effect. Has bus plan facilitated reform to the scale needed and that it aspired to? The record so far is that it has not, with the pace of reform both slower than other cities (eg Perth) and our own record in the 2006-2010 period. Achieving the aspired pace of reform requires changes within DTP as current processes have more in common with cottage industries than the mass production needed for large-scale delivery.  

The government releasing the Bus Plan without substantial funding in 2021 and the department's inability to argue its case for this in the 2023 and 2024 state budgets wouldn't have helped its standing either. Rather than being a substantive program or project (like the WGT or SRL), the Bus Plan has been demoted to be more a thinking approach to be applied if or when the government wants to do things with buses.

Essentially the Bus Plan has been an unloved orphan for most of its first four years. There are welcome signs of revived government interest in buses in the 2025 state budget, though many are more 'catch-up' growth area additions than established area network reform as envisaged in the Bus Plan. 

Conclusion 

The audit's findings are a matter for the Auditor-General after weighing all evidence.

However based on what is known publicly I would imagine its conclusions may be more than a 'good job keep it up' type result. After all VAGO performance audits almost always find something that can be improved, even if minor. 

Consequently the Bus Plan audit promises to be interesting with meaty recommendations that could give the department some worthwhile guidance on improved performance. Especially if laced with big helpings of reality from auditors who are not so remote as to forget to occasionally GOTB! 

Regional Rail Link turns 10 / Metro Tunnel testing

Today is a super important date in transport for two reasons. One related to the past, the other for the future. 

Regional Rail Link & reformed buses

Firstly regional Rail Link turns 10 today. This transformed transport in Melbourne's west and Geelong with the new Regional Rail Link starting, bringing rail to new stations at Wyndham Vale and Tarneit. We had built new stations and extended electrification but this was a completely new line on a completely new alignment.

While the intent was to provide a bypass to free up rail capacity in western Melbourne, the new stations became massively busy in their own right due to them serving huge growth area catchments popular with migrants seeking new affordable housing. Tarneit, for example, is now V/Line's busiest station outside Southern Cross. This 2010 Paul Mees ATRF paper, throwing shade on the Regional Rail Link, has not aged well given the undoubted patronage success of its stations; it is now impossible to see western Melbourne without it. 

The basic off-peak weekday service to South Geelong started as every 20 minutes, versus the hourly provided under the old alignment via Werribee. Patronage boomed with this link becoming crowded shortly after opening. Weekend service started at hourly on opening. However crowding forced that to every 40 minutes. That wasn't enough so some extra trips were slotted in to provide some 20 minute intervals. Finally last December a 20 minute 7am - 9pm weekend timetable was instituted with more trips extended to Marshall. 

The Regional Rail Link has one 'good' problem -  it has become a victim of its own success. There is massive demand for more than the original two stations. There's been talk of extra stations but none have opened in RRL's first decade. Although West Tarneit is under construction with an opening due next year.

Reluctance to add new stations may be due to a view that this will crowd trains and unacceptably slow travel for Geelong passengers. RRL's future is clearly a two-tier service, with an intensive (possibly electric) service to Wyndham Vale and express services to Geelong/Marshall/Warrnambool. There also needs to be a westward extension of Werribee Metro electrification with a new station at Black Forest Road.  

Wider network reforms - varying prospects

Regional Rail Link in 2015 was not just a V/Line rail upgrade project. As originally conceived it was also going to be the centrepiece of massive multimodal Metro train and bus network revamps across Melbourne. Planning for this had been done but their fates varied.

While the new Labor government pursued its infrastructure program with gusto, it resisted adding Metro service, especially if new timetables would affect travel for some in (then) marginal seats on the Frankston line. Thus the proposed 2015 Metro timetable was ditched, setting rail service reform back years, with elements not picked up until 2021 (and again likely late 2025). This delay meant that some anticipated benefits of RRL, that of freeing space for more Werribee line Metro services, were not realised as soon as they could have been, as the Auditor-General noted in 2018. A decade on we're still waiting for some, though last month's state  budget has funded some welcome Werribee peak uplifts.   

The government wasn't just shilly-shallying on Metro rail service reform. The proposed 2015 Transdev greenfield bus network was also scrapped . Although some reasons for not proceeding with some of this were sound, a promised later review did not happen and the cause of bus reform was put back years.  

Fortunately the radically reformed RRL-related bus networks designed for Geelong and Wyndham survived. Geelong's new 'greenfields' network featured simpler more direct routes running at higher frequencies. Apart from isolated and minor objections this simplified Geelong bus network has proved successful with operating hours its main outstanding concern. 

RRL could have opened without the Geelong bus changes. But the bus revamp enabled PTV to present a compelling multimode package to the public on Day One. This is a recipe Perth's Metronet routinely follows with new bus networks for Airport, Yanchep, Ellenbrook and Thornlie-Cockburn. Melbourne's record here is only sporadic; if we were as good as Perth now (or our own record with Geelong in 2015) our Metro Tunnel would get a radically revamped bus network between at least Watergardens and Dandenong along these lines along with reformed CBD trams. The indications so far are not strong for sweeping changes here, although some smaller bus reforms around Parkville happened last year. 

Wyndham also got a brand new bus network on this day in 2015. In contrast to Geelong that had to happen as otherwise there would have been few if any routes properly serving the new stations at Tarneit and Wyndham Vale. That would have caused massive parking pressures and jeopardised public goodwill towards the RRL.

The 2015 Wyndham bus network comprises a two-tier network with more frequent main road routes and neighbourhood style coverage routes. While service levels aren't yet as high as they need to be the network has been extremely successful with Wyndham bus routes ranking amongst the most productive in Melbourne. This has set up Wyndham buses on a path of continuous improvement, as opposed to the network atrophy seen in other areas (eg Melton town, Broadmeadows, Preston - Epping, Ringwood, Knox, Greater Dandenong, Frankston etc). The latest Wyndham bus boost will start in less than two weeks, with improved timetable for four routes starting on July 1.  

See my 4th anniversary write-up here for more about the Regional Rail Link. 

Metro Tunnel's full day test

This one's about the future. 

Rail officials and enthusiasts alike are out and about all day today to watch a full day's testing of the Metro Tunnel timetable. 

While it's a Saturday, Rail Express reports that a weekday timetable will be operating on lines that will use the Metro Tunnel. 

Passengers will need to change at Caulfield or Footscray if travelling through to the city.

This won't matter to observers as they can just find a seat somewhere, grab their watch and tick off train arrival times. While this is normally only an activity undertaken by signalling staff and hard-core gunzels, this time it's of much wider interest. 

This interest is partly because the state government has kept even basic Metro Tunnel frequency information under wraps. 

Despite adding relatively little Metro train service in its first decade, it has raised expectations of the service possible with the Metro Tunnel and other projects. This heightened anticipation has given rise to a lot of guessing and discussion including here.

Speculation on Metro Tunnel service levels intensified last week when PTV's website showed modified timetables for Cranbourne, E Pakenham and Sunbury, applicable for today, that could well be part of a Metro Tunnel weekday timetable.

Daniel Bowen analysed this here . He hoped it was only partial, with the real timetable having more trips. Such a consistently high all-day frequency is key to the sort of multi-directional changing all day activity that is needed to make the Metro Tunnel a patronage success.  

It is also desirable that best endeavours are made for the final timetable to avoid big 'holes' such as the peak period / peak direction 20 minute gap towards Cranbourne shown below.  

The 2012 Network Development Plan - Metropolitan Rail proposed a hierarchy of base all-week frequencies for the rail network, including a 5 min inner core interval. Peak frequencies would be higher. For example it was envisaged that Cranbourne, Pakenham and Sunbury (all Tier 2 stations) would have 5, 9 and 9 trains per hour respectively by 2022 (see Figure 5-8). 

The pandemic and its aftermath have reduced peak period train usage to less than envisaged in the NDP. However a 5 minute all day core frequency between at least Footscray and Caulfield remains essential for the Metro Tunnel to provide a 'big city' metro user experience, justify its construction costs and play its full role in the central area transport network (including enabling cascading tram network reform).   

It will be interesting whether observations people make today tally with published timetables or not. And if any media releases reporting on the success or otherwise of today's trials provide further information on Metro Tunnel service levels.

Any observations on today would be welcome and can be left in the comments below.