How AI Could Really Transform Transport Outcomes
Key Takeaways
- AI will make transport policy, strategy, and planning faster, cheaper, and, for under-resourced agencies, higher quality, but these process gains alone won’t transform transport outcomes.
- Many transport strategies already say the right things; the real bottleneck is the political decisions that follow them.
- Public opinion and the car lobby keep pushing politicians towards bigger roads, and better strategy documents won’t change that.
- Politicians are already using generic AI tools for policy advice, and that advice is currently poor.
- Specialist AI advisors built on local data and the latest research could give decision-makers advice as good as an average human advisor.
- Unlike some human advisors, a well-designed AI advisor won’t soften frank and fearless advice when a politician pushes back.
- AI advisors could introduce politicians to politically smarter ways of delivering change.
- Instead of strategies being set in stone for years, AI could support living policy that flags when targets are slipping and suggests what to do about it.
- Giving the public access to the same AI analysis as decision-makers could close the accountability gap between what strategies promise and what gets built.
- The specialist AI advisor model could apply across many areas of government, from transport to housing and urban planning.
What Next?
Should you be aiming to build a high-quality AI advisor?
Introduction
AI is all the rage at the moment, and transport is no exception. Government agencies and consultancies are starting to use it to support transport policy, strategy and planning, with the promise of producing work faster, cheaper and to a higher standard.
Those gains are real and worth having. But I don’t think they will deliver significantly better transport outcomes. Most of our transport strategies already say the right things. What holds us back is the decisions that follow them, and producing better documents more quickly won’t change that.
There is a bigger opportunity, though. Politicians and the public are already turning to AI for policy advice, and right now that advice comes from generic tools that aren’t up to the job. If we built specialist AI systems that could act as genuine advisors, they could improve the advice decision-makers get, help them find politically workable ways to deliver change, and close the accountability gap between what strategies promise and what actually gets built. Getting there means rethinking how we use AI in government, not just in transport.
In this blog, I’ll look at what AI is currently expected to deliver in transport policy, strategy and planning, why that falls short, and where it could really help to transform transport outcomes.
What AI is currently envisioned to deliver
Most of the discussion about AI in transport policy, strategy and planning (PSP) focuses on improving the process. Broadly, the benefits fall into three buckets: speed, cost and quality.
Speed
AI should significantly speed up PSP. It can bring together disparate data sources and analyse them much faster than humans can. It does need appropriate access, and getting data into shape will take time and resources. It can then use that analysis to quickly produce the outputs needed throughout the process, including a first draft of the strategy itself.
Cost
Much of the work AI does will replace effort currently done by people, either in government agencies or consultancies. Agencies could use those savings to cut costs, to deliver more with the same team, or some combination of the two. My guess is that in most cases it will be the latter.
Quality
Where PSP is done by non-specialists in under-resourced agencies, such as many local governments, AI could significantly improve the quality of what is produced. For larger agencies, such as city, state or national departments of transport, I think the gains will be marginal. There will be improvements. For example, it should be easier to incorporate the latest research. But in my experience, AI has not yet grasped the subtleties of how transport systems work in the way the best transport policy planners and strategists do (like the ones who read this blog). Today’s frontier models still miss these subtleties. I expect them to get better, but I don’t expect that to change the outcomes we deliver fundamentally.
Why not? Because in most cases, our problems aren’t about the quality of our strategies. They are about the decisions that get made once those strategies are written.
The Real Reason That Outcomes Are Held Back
I read a lot of transport policies, strategies and plans, and many of them end up in the newsletter. A lot of them are genuinely impressive and have the potential to be transformative. Yet in the vast majority of cases, the outcomes they deliver are underwhelming, to say the least.
They talk about mode shift, sustainability, inclusion and equity, and every other motherhood-and-apple-pie aspiration under the sun, but the reality on the ground doesn’t change.
The bottleneck isn’t the strategy documents that get approved and published. It’s politicians and public opinion.
Public opinion is still heavily weighted towards bigger roads as the answer to traffic. Anything that takes road space from cars, such as bus or cycle lanes, gets painted by the car lobby as a war on motorists. So politicians keep promising more and bigger roads to fix congestion, because that is what wins votes.
Producing PSP documents faster, cheaper and to a slightly higher standard won’t change this dynamic. The real question is whether AI can change the decisions that follow them. I think it can, but only if we use it differently than government usually envisions. We need to use it as an actual advisor.
The AI Advisor
The uncomfortable truth is that transport decision-makers are already using AI for advice.
Right now, that advice is fairly poor and a long way from what you would get from leading transport experts. That isn’t surprising, because politicians are using generic tools that weren’t built for this.
But what if we built specialist AI systems trained on local data and context and the latest research? What if they were designed to minimise hallucinations and could give advice as good as an average policy, strategy or planning advisor?
That doesn’t feel far from where we could be today.
I wouldn’t want an AI system to be the only advisor in the room. But I think it could help solve several real problems that stand in the way of better transport systems. The rest of this blog looks at four of them: the need for better-advised decision-makers, the lack of political implementation options, policy that stands still, and the accountability gap.
Better Advice for Decision-Makers
Most politicians don’t go into politics because they are interested in transport. Many barely think about it until they are given responsibility for it. So they arrive with the same views as much of the public: more roads, carrots like cheap or free fares to get people onto public transport, and more free parking are all good policies.
In theory, it’s the job of public servants to explain what good transport policy looks like, and why some of the politicians’ instincts won’t work. In practice, this runs into two problems.
The first is that it means public servants have to have difficult conversations with politicians. In the Westminster system, this is known as providing frank and fearless advice. The best public servants are masters at this and build the kind of trust that means politicians actually listen. But it is often done badly. Stories of angry politicians shouting at their advisors spread quickly, and many public servants decide the easier option is to go along with the politician. The advice becomes more pliant and fearful than frank and fearless.
The second is that too many people at the top of transport agencies don’t understand good PSP either. They often come from generalist backgrounds and are still working out for themselves why more roads, more carrots and more free parking aren’t good policy. I wrote about that problem here.
So how could AI help? People already turn to AI for advice on all sorts of things, and politicians are no exception. They are already using it to help form policy, including in transport.
Now imagine an advisor that a politician can consult as often as they like. It won’t change its advice because the politician shouts at it, and if it is designed well, the advice will be good. The design of the AI advisor matters, because off-the-shelf AI tools are built to please their users, not necessarily to tell them the truth. But if we get specialist AI systems right, the quality of the advice politicians get could improve, possibly significantly.
For example, a politician might ask the AI advisor what will happen to traffic in the years ahead. An honest and accurate answer might be that, because most transport projects are planned on a predict-and-provide model, traffic will keep growing. If they want to reduce traffic, a vision-led planning approach focused on mode shift is what will deliver it.
There are, of course, politicians (and public servants) who are ideologically committed to building ever more and bigger roads, and no amount of advice, from AI or humans, will change their minds. In my experience, though, they aren’t the majority of the politicians put in charge of transport. Bringing the rest up to speed on the subtleties of transport policy could make a real difference.
Political Implementation Options
AI could also help politicians find politically smart ways forward.
Many politicians don’t think beyond a top-down approach to transport policy. The government announces a strategy, then announces projects and consults on them, but by then the decision has already been made. If that is the only tool you have, and you have seen it provoke a backlash against controversial policies, you are much less likely to take those policies forward.
There are good examples of other ways to deliver significant change that may be more politically palatable, and most politicians are unaware of them. One is the bottom-up approach Milan has used to reclaim public space (see here for the recent Freewheeling podcast on what Milan has done).
Whether a human advisor knows about approaches like these is hit and miss. An AI advisor could help decision-makers find them more easily, and give them the confidence to pursue policies they would never attempt with a traditional top-down model.
When politicians can see a politically workable path to their objectives, they are far more likely to act than if they only have tools they believe will fail.
Living Policy
At the moment, PSP is periodic. We produce a strategy every few years, often after a change of government, and then it is effectively set in stone until the next one.
What we aren’t good at is living policy that adapts as circumstances change and new data comes in. We might be falling behind on our targets for cutting carbon emissions, increasing mode shift or improving road safety, and the strategy carries on as if nothing has changed.
An AI advisor could flag when the latest data shows we are off track and suggest what to do about it. Politicians have often chosen to water down targets they aren’t meeting instead of changing their policies, and some will keep doing that. But others may be willing to try new approaches to hit their goals, and an advisor that keeps pointing out the gap gives them both the prompt and the options to act on it.
The Accountability Gap
As I said earlier, many transport strategies say one thing while the reality on the ground is another. Why do politicians get away with this? Part of the answer is the accountability gap.
Major newspapers and TV networks used to employ policy specialists who could analyse what politicians announced, report on whether it was a good idea, and explain to the public why a difficult reform was worth doing. The media has changed enormously over the last 30 years or so, and those days are long gone. Most journalists now have neither the time nor the expertise to scrutinise transport policies and their implementation, or to hold politicians to account for them. The result is many motherhood and apple-pie strategies that are never delivered.
So how could AI help close this gap?
Transparency is supposed to be one way governments make scrutiny possible. In reality, it’s heavily restricted.
You can ask the politician, but they will often dodge an inconvenient question. You can ask public servants, but access to them is limited, and they stick to the lines the politician has agreed.
What you can’t easily get is the actual analysis behind a project, plan or policy. You can lodge a freedom of information request, but even if it’s granted, what you receive may be a sanitised version prepared in anticipation of that request, without the problems the full analysis identified.
What if the public had access to the same AI system that analysed the project or policy? The analysis would suddenly be open to everyone. Will this new road induce demand and create more traffic? How much will this strategy cut emissions or road deaths? How much will free fares reduce car use? The public would be able to get the same advice the politicians received.
That would make it much harder for politicians to sign off on a motherhood-and-apple-pie strategy and then keep approving projects that contradict it without clearly explaining why. Over time, this could lead to one of two things. Either strategies become more honest (“we will keep building roads even though we expect them to make traffic worse”), or politicians implement their strategies more consistently.
Will this kind of radical transparency ever happen? Freedom of information laws once seemed like pie in the sky too, and today they are common in many jurisdictions. AI systems will be used for policy advice and analysis. When they are, I expect a growing demand for that advice to be made public.
Together, these tackle both sides of the bottleneck: better advice and smarter implementation options help politicians who want to do the right thing, while greater transparency helps the public understand policy decisions.
What Could Go Wrong?
None of this is without risk. Here are the biggest ones.
Whose values does it reflect? If an AI advisor tells a politician that more roads will mean more traffic, some will say it is biased against cars. The answer isn’t to make the advisor neutral on everything. It’s to make sure it is grounded in evidence, such as decades of research on induced demand, and transparent about what that evidence says and how certain it is. A good advisor, human or AI, separates what the evidence shows from the value judgements that are rightly left to politicians.
Who controls it? The same incentives that produce sanitised FOI documents could produce a sanitised AI advisor. A government that doesn’t like the advice could quietly retrain the system, restrict what it can say, or simply not publish it. That’s why governance matters as much as the technology. One option is an independent body that oversees the system's development at arm’s length.
Will it actually be any good? Off-the-shelf AI tools still hallucinate and are built to keep their users happy. A specialist advisor would need to be rigorously tested against the judgement of experienced practitioners, and it would need to say plainly when it doesn’t know something. A confident but wrong AI advisor could do more damage than no advisor at all.
Does it undermine democracy? Some people will be uneasy about AI shaping public policy. But the AI advises, and elected representatives decide. If anything, giving the public access to the same advice as their politicians strengthens democratic accountability instead of weakening it.
Will it replace human expertise? It shouldn’t. As I said earlier, I wouldn’t want an AI system to be the only advisor in the room. The risk is that agencies use AI to cut experienced staff and lose the expertise needed to judge whether the AI’s advice is any good. The best outcome is an AI advisor that complements outstanding frank and fearless public servants, not one that replaces them.
Politicians are already getting advice from generic AI tools, with none of these safeguards. If they are going to use AI, it is better to improve the advice they receive from an AI tool.
Beyond Transport
This specialist AI advisor idea isn’t limited to transport. It could work across many areas of government policy.
Take urban planning. An AI advisor could explain to a politician what is holding back housing supply, show the public whether current policies will deliver enough homes, and set out how other jurisdictions have handled the politics of planning reform.
What I’m describing could become the standard way of doing policy, strategy and planning across many disciplines, not just transport.
Conclusion
AI is going to change how we do transport policy, strategy and planning. That part is almost certain. Strategies will be produced faster, analysis will be cheaper, and under-resourced agencies will be able to produce better work than they can today. These are worthwhile gains, but they won’t significantly change the outcomes we see on the ground, because the quality of our strategies has rarely been the main thing holding us back.
The bottleneck is decision-making. Politicians who come to transport with the same instincts as the general public need better advice. Many don’t know there are politically smart ways to get difficult reforms done. Strategies are set and then left alone while the data tells us we are falling behind. And the accountability gap lets motherhood and apple pie strategies sit alongside projects that contradict them.
This is where AI could make a real difference. Specialist AI advisors, built on local data and the latest research, could give politicians frank and fearless advice that doesn’t soften when it becomes inconvenient. They could point to alternative ways to deliver change and flag when policies drift off course. Most radically, they could give the public access to the same analysis as decision-makers.
From where we are today, you might think this is a stretch. But decision-makers and the public are already using AI to get policy advice. What I’m proposing is that we improve the quality of the advice they get, giving us an opportunity to deliver transport systems that work better for everyone.
