We snagged a campsite with an AI bot. Here’s what it means for impossible campground bookings | Canada Voices

Ted De Jager is well acquainted with the annual spring ritual that often precedes summer camping at national and provincial parks for Canadians across the country.

Every year, in early May, the Abbotsford, B.C.-based retired RCMP officer arranges what he jokingly calls a mission control-style set up. A few minutes before 7 a.m. PT, he and his wife each log into the BC Parks website, which handles campsite reservations for provincial parks, with both a computer and an iPad.

One phone is also signed into the campsite booking platform – as an emergency back-up – while a second phone is running FaceTime to allow real-time co-ordination between the couple and two similar set-ups run by their longtime camping friends.

The moment of truth comes some time between 6:59:59 a.m. and 7 a.m., when online campground reservations open for dates starting three months ahead on a first-come, first-served basis. Will the friends be able to reserve enough sites for their annual get-together at Manning Park in August?

With several devices active, the group usually manages it. But even Mr. De Jager, who’s had his share of thrills in his career as a police officer and 15 years in the army before that, finds it all a bit too nerve-wracking.

“This is a whole different kind of stress.”

Yet Mr. De Jager’s stress – and his odds of getting a campsite, along with those of many other avid campers – could get worse. The reason is artificial intelligence.

Tech-savvy consumers around the world are increasingly relying on AI agents to make digital reservations for anything from restaurants to shuttle services to the airport. The Globe and Mail set out to see if it could design an AI-powered bot that can make campground bookings exactly when reservations open. We did, and it was easy.

The success of our experiment raises the concern that users deploying AI bots will be able to snag camping spots faster than even the fastest-clicking humans on government-run campground reservation platforms, which mostly rely on first-come, first-served systems.

For years, frustrated campers have speculated on social media about bots snatching coveted sites at popular parks. But evidence of the issue remained scant. Several provincial park agencies and Parks Canada, which oversees national parks, have said their platforms keep bots at bay by preventing bulk reservations and requiring users to set up secure individual log-in accounts, among other tools.

AI promises to unleash a different kind of bot. Not the so-called scalping bots programmed to make mass reservations to resell at higher prices, but bots that use an individual’s account to simply make a booking on their behalf and automate a time-consuming task.

This sort of bot is much harder for booking websites to detect because it mimics a human consumer’s behaviour, experts say. And it poses a particular challenge for reservation systems that award spots based on user speed.

What to know about Ottawa’s push for more data centres

Even before AI, the first-come, first-served model tended to favour savvier, faster, better organized people, said Nick Arnosti, an associate professor at the University of Minnesota in the department of Industrial and Systems Engineering.

“Now, if you’re writing a bot, basically you have an even bigger advantage over other people,” he said. “It absolutely would make the issues of first-come, first-served, I think, even more extreme.”

But there’s a potential silver-lining. The rise of AI booking bots could accelerate efforts already afoot, in both Canada and the U.S., to find a better system to award reservations for the limited number of campsites at government-run parks – or to allocate any scarce supply of public goods, for that matter. A system that’s fairer, less stressful – and where bots won’t have an edge.

It took The Globe only a couple of hours to build a campsite-booking bot using the Claude AI model. The AI assistant, which is trained to attribute importance to factors such as fairness and rule-following, it told The Globe, initially expressed reservations about building such a tool, which it likened to a scalping-grade bot.

But Claude eventually agreed to complete the task as a proof-of-concept meant to aid The Globe’s journalistic endeavour.

Experts call for ‘mystery shopping’ of AI chatbots to test if they are advising on self-harm, child bullying

We tested the bot on the Ontario Parks camp reservation page, which relies on a booking platform similar to the one used by BC Parks and Parks Canada. We scheduled the bot to reserve a campsite at Algonquin Provincial Park’s Mew Lake campground at 7 a.m. Eastern Time on a recent July morning for reservations opening five months later in early December.

While competition for camping in winter is much less intense than for the summer months, the trial run was eye-opening. The bot clicked its way from the log-in portal to the payment page in around 15 seconds. (The Globe then released the spot before completing the booking to avoid competing with actual campers.)

Dave Hunt, a spokesperson for Ontario Parks, said the agency uses a variety of safeguards to protect against unauthorized or large-scale automated activity. But AI-assisted tools such as the one used by The Globe, “reflect a broader industry challenge” for reservation platforms, he said in an e-mail.

BC Parks and Parks Canada provided similar responses to The Globe’s requests for comment on the AI bot.

Fake reviews are multiplying. Here’s why it will cost you

One way to remove the advantage of AI bots is to allocate campsites through a random draw. That’s the system Parks Canada already uses when reservations open at its most popular campgrounds.

Users join a virtual waiting room shortly before bookings open for the camping season at a particular park. Then, at a set time, the system randomly assigns everyone in the room a position in a digital queue. As people get to the top of the line, they can reserve sites and dates based on availability – until everything is booked.

Entering the virtual room early doesn’t grant any advantage, but arriving late lands you at the very back.

Virtual queues are already widely used for anything from concert ticket sales to product launches as a way to distribute small inventories of sought-after goods in an orderly fashion when demand far outstrips supply.

Another benefit: transparency. When consumers queue up in the physical world, “people will stand in that line and they will see some kind of order,” said Jesper Essendrop, chief executive of Queue-it., a Copenhagen-based virtual waiting room software provider that has worked with clients including Ticketmaster, Cathay Pacific and the Tokyo Metropolitan Government.

Assigning a number in a queue to users online recreates a similar experience, he said. And in a randomized line, a bot holding a spot for a single customer doesn’t affect other people’s chances.

“It should be part of the lottery and it should have a place in line,” Mr. Essendrop said, referring to this type of bot.

The tricky part is figuring out which bots have legitimate intentions and which ones are up to no good, he added. Until recently, bot detection was all about distinguishing between human users and automated agents, which were generally assumed to be bad actors that should be blocked.

But with people now using AI agents to do their shopping and make their bookings, virtual lines need the digital equivalent of a bouncer at the door who’s able to tell a well-behaved bot from a mischievous one, according to Mr. Essendrop.

Those authentication standards, which vet a bot by verifying its mandate, are still emerging, he said.

“It will be a huge challenge for e-commerce companies, ticketing companies and public institutions,” he said.

But if lottery systems neutralize a bot’s speed advantage, they also generate their own kind of frustration.

Roy Kuhnlein and Betty Ann Lough know that feeling well. The couple, who live in Calgary, have been trying their luck with Parks Canada’s random queue to camp at Lake O’Hara in B.C.’s Yoho National Park for four years – so far without success.

The turquoise lake surrounded by pristine alpine valleys is one of Canada’s most sought-after camping destinations. Parks Canada now relies on a random draw for local campsite bookings and has even set up a lottery for tickets to the shuttle bus that is the only way to access the area for day trips.

Mr. Kuhnlein and Ms. Lough used to camp at Lake O’Hara in the 1990s, back when one could reserve a site with a simple phone call. Today, they feel shut out of the place that provided the gorgeous backdrop for so many family memories.

They got to camp there again for a single night in 2024 – but only thanks to a cancellation by a friend who’d gotten a spot. Their own attempts have gone nowhere, year after year.

A randomized virtual line may be a fairer, better system than first-come, first-served, but it still doesn’t seem quite right, Mr. Kuhnlein said.

“It’s pretty disheartening when you see you’re number 9,037.”

It’s why some parks are adding a twist to the random draw approach. California, for example, is switching to a weighted lottery system for its most popular camping destinations. Every time users try the lottery with no luck, they receive one point to be used in future draws, which improves their odds. Once they win a reservation, their balance resets to zero.

The idea of tilting lotteries in favour of people who have tried and failed before isn’t new, said University of Minnesota’s Prof. Arnosti. It’s a system commonly used to allocate, among other things, hunting licences and registrations in some popular ultramarathons, which attract far more runners than they can accommodate, he said.

The biggest advantage of this system is that it, essentially, evens out users’ luck across the years, he said.

“It allows you to make the system more fair across time.”

Opher Baron, distinguished professor of operations management and statistics at the University of Toronto’s Rotman School of Business, likes that approach. But he suggests an additional tweak: Reducing credits for users who misbehave, such as campers who fail to show up on their reserved dates and don’t cancel their bookings.

This would be to discourage undesirable behaviour without unduly penalizing users for a one-off infraction, he added.

Of course, the hard work is figuring out exactly the details of how to assign the weighting. And then there are ethical questions about who deserves to receive preferential treatment. For example: Should local residents be prioritized over tourists?

“There’s not going to be one perfect system,” Prof. Arnosti said. But weighted lotteries have the potential to be a rather good one, he added.

That likely holds even in the era of AI bots.