麻豆传媒

A KILLER hangs in the crystal clear waters around the Great Barrier Reef. It鈥檚 not the box jellyfish, although they are present, along with sharks and stingrays. This predator carries a highly effective poison and can roam for 8 hours at a stretch, terminating up to 200 marks in that time. This is one of the world鈥檚 most advanced killer drones.

Thankfully, humans aren鈥檛 in its cross hairs. Instead, it hunts down starfish. But not just any starfish. These threaten the very existence of the Great Barrier Reef, and this robot鈥檚 mission is to protect it.

Yet the robot is crossing a boundary that many find uncomfortable: it makes the 鈥渒ill decision鈥� autonomously, taking away all input from humans. Is that a safe price to pay to save the reef?

Crown of thorns starfish (COTS) are among the largest starfish, with up to 21 arms bristling with spikes. In a healthy reef they are a positive force, nibbling on the most rampant corals, keeping them in check and giving slower-growing species a foothold.

The problem comes when their numbers run out of control. It鈥檚 not easy to know their numbers precisely. One way to find out is to tow a diver behind a boat for a series of 2 minute intervals and ask them to count starfish. Obviously they miss a few. But as a rough guide, a healthy reef might have about one crown of thorns starfish per hectare, an area about the size of one and a half football pitches. It鈥檚 considered unsustainable when that rises to 15 per hectare.

On the Great Barrier Reef there have been four outbreaks since the 1960s. The latest began in 2011 near Cairns and is likely to spread (see 鈥淐oral catastrophe鈥�). 鈥淲e are seeing the beginnings of a fourth recorded wave,鈥� says , who leads a reef monitoring programme at the Australian Institute of Marine Science.

One cause of the mushrooming numbers is fertiliser run-off from farms, which flows in ever-increasing amounts into rivers and then sloshes onto the reef. This creates algal blooms, which provide a feast for starfish larvae if they occur during the spawning season in October.

鈥淭hese animals produce 65 million eggs,鈥� says in Queensland, who studies the starfish. So under the right conditions, their population can explode. And that does serious damage. that 42 per cent of losses to the Great Barrier Reef between 1985 and 2012 were due to the pests.

In 2011, the Australian government launched a programme to reduce the numbers of the starfish. Divers scanned the reef armed with long poles they could use to inject them with poison. The problem was it took 20 or more injections to kill each starfish and the divers could only cover so much ground.

Enter the killer robot. Designed by a team at the Queensland University of Technology led by , COTSbot visually searches coral reefs, identifying crown of thorns starfish and injecting them with poison.

Clean kill

It鈥檚 remarkably good at its job. 鈥淲e鈥檙e approaching 99.9 per cent accuracy,鈥� says Dunbabin. Using machine learning, the robot has been trained to identify the starfish so well that it has outgrown its stabilisers. 鈥淲e trained it with 3D-printed COTS initially,鈥� says Dunbabin. But these days, it can tell they aren鈥檛 real and won鈥檛 inject them.

So, in late October, Dunbabin was on a boat at Maori Reef near Cairns testing the robot on a coral reef with real starfish for the first time. The next step will be to test what effects, if any, the robot might have on the wider reef ecosystem. This will be done by roughly February, says Dunbabin, and if all goes well the plan would be to eventually deploy it for real pest-control missions.

This drone is one of the most advanced killer robots in the world, in that it鈥檚 able to make a kill decision independently. Dunbabin isn鈥檛 sure if that鈥檚 been done before, besides possibly in a for South Korea鈥檚 border with North Korea.

The bot was designed in 2005, numbers more accurately than the diver surveys. There wasn鈥檛 much hope of turning it into a killer back then because the ineffective poison was still a limitation. 鈥淵ou had to inject every tentacle,鈥� says Dunbabin. 鈥淓ven for today鈥檚 technology, that鈥檚 probably beyond what we could do.鈥�

That changed when Rivera Posada was relaxing on the beach and had a brainwave. He was in the midst of studying the bacteria that live in the starfish, and so was encouraging them to multiply by putting them in a culture medium, a mixture of nutrients in which bacteria thrive.

鈥淲e were just drinking beer on the beach and I thought maybe this culture medium could induce an imbalance in the animals,鈥� he says. His initial idea was that perhaps the medium would encourage some particular bacteria to grow and cause an infection. But when he tried it out, the effects were much more dramatic. 鈥淚t just melts the starfish, it鈥檚 incredible.鈥�

鈥淲e trained it with 3D-printed starfish, but now it knows they鈥檙e not real鈥�

It turned out that the chemical responsible is bile, the yellowish substance your liver produces to break down fats. It鈥檚 produced by almost all vertebrates, but not invertebrates like starfish. It鈥檚 foreign to them and produces a violent allergic reaction that proves lethal.

Incidentally, there are other places with a COTS problem besides the Great Barrier Reef: reefs near Japan, the Philippines and the Pacific island of Vanuatu. Rivera Posada is now helping divers tackle some of these using bile. They simply get the bile from slaughterhouses, he says, where it would normally be tipped down the sink. Then they swim around injecting it into starfish. Since it occurs in all marine mammals anyway, there鈥檚 no risk it could cause any environmental problems, Rivera Posada says.

Back in Australia, the bile discovery turned COTSbot into a killing machine. Dunbabin鈥檚 team designed a unique pneumatic injection arm to deliver the fatal shot. Just is enough to kill 200 starfish. But given the ravaging hordes that can crop up during an outbreak, Dunbabin reckons we will still need an army of bots.

No mistakes

Even then, the drone doesn鈥檛 replace people entirely, as divers would still be needed to find starfish hiding under rocks. 鈥淲e see COTSbot as a first responder,鈥� says Dunbabin. But how comfortable should we be with a bile-spitting killer robot roaming the reefs?

鈥淚 am cautious about equipping any robot with weapons and autonomy,鈥� says Toby Walsh, an artificial intelligence engineer at the University of New South Wales in Sydney. He鈥檚 an active member of the Campaign to Stop Killer Robots, a group with the central aim of keeping kill decisions in the hands of humans.

Walsh says it will be important for COTSbot to be immune to hacks, and it shouldn鈥檛 be capable of making mistakes or being repurposed by others. But, overall, he is rather impressed. 鈥淭he only good killer robot is one that kills pests,鈥� he says.

Dunbabin says these concerns have been thought about in the robot鈥檚 design. 鈥淲e set its threshold for detection so high that it鈥檚 not going to inject anything but a crown of thorns starfish,鈥� he says. It鈥檚 not connected to the internet, so there鈥檚 no way anyone could remotely hack it. Even if someone did, he says, its behaviour is so strictly specified there isn鈥檛 much mischief they could do with it. You might be able to hijack the hull and use it for your own purposes, he admits, 鈥渂ut anybody could get themselves a hull鈥�.

Killer robots do have one big point in their favour. 鈥淚n Australia, we鈥檝e had a few disasters introducing new species to kill pests 鈥� witness the cane toad problem,鈥� says Walsh. 鈥淭he advantage of introducing robots is that we can easily turn them off once they鈥檝e done their job. We鈥檙e not going to end up with a plague of crown of thorns-killing robots.鈥�

(Images: Yann Hubert/FLPA; Dr David Wachenfeld/Auscape)

This article appeared in print under the headline 鈥淭he starfish terminator鈥�