Kill Decision, is the 2012 bestseller by Daniel Suarez, It is a fast, fun, and accurate introduction to robot swarms that also covers the ethics of lethal autonomous weapons.
Most non-roboticists think of swarms as just lots and lots of robots, what we call more precisely multi-robot systems. It is true that swarms consist of many robots, but in robotics a swarm is a very special case that emulates insect swarms or animal flocking:
a collective of identical (homogeneous) robots who are not very smart individually,
accomplish tasks by sheer dint of numbers where the desired effect emerges, and
the robots don’t directly communicate and may not even be aware of other members of the collective.
Note that we use the term “collective” because there is no real team work, as a team requires being aware of others on the team and their goals and capabilities. The Borg collective on Star Trek: The Next Generation originally captured the idea of a collective swarm but the show undermined the scientific verisimilitude when it introduced the Borg Queen and centralized control. When a roboticist says “swarm,” it almost always means that they are using a model of insect swarming as the organizing principles for their algorithms. Though it could mean flocks of birds or schools of fish.
Conversations between roboticistis working in swarms goes something like this:
“I’m Dr. X and I work with swarms.”
Dr. Y looking politely unimpressed replies “Ants?” (That’s a good guess because ants are really popular given that they are such a successful species and have been extensively studied by entomologists.)
Dr. X sighs because indeed ants have become passe’ in robotics, having been replicated in robotics simulations since the 1980s and replies, “No, I’m using the schooling behavior in fish to maintain formations in the presence of obstacles.”
Dr. Y becomes more appreciative, “You’re duplicating the lateral lines on fish that detect if they are next to another fish?”
Dr. X nods enthusiastically, “We have a new short range sensor! Just by knowing the closeness to two other fish, the school can maintain formations around obstacles with only 30 lines of code.”
Then Dr. Z jumps in and says “I’m doing something similar but each of my robots has GPS and uses wireless to communicate its GPS location to every other member of the swarm and then they optimize their location.”
Dr. X and Dr. Y look horrified. “That’s direct communication!” says Dr. X.
“No, no, no, just think of GPS and wireless communication as a type of sensor,” Dr. Z explains.
Neither Dr. X or Dr. Y are buying it. “Being called on a cellular phone is not the same as you seeing someone is standing next to you,” Dr. X. “Besides, that’s a lot of lines of code!” says Dr. Y.
“But it’s optimal!” Dr. Z protests.
“Living things don’t have to be optimal to be successful,” Drs. X and Y say in unison and walk off. Dr. Z eats lunch alone.
Kill Decision is set in current time. Some member of the Industrial-Military Complex is combining recent breakthroughs in computer vision with models of weaver ant behavior to create highly aggressive autonomous unmanned aerial systems. One small UAV or drone targeting a person or group for assassination is scary. Imagine thousands of those inexpensive white DJI Phantom quadcopters acting like a swarm of ants or angry wasps. Now imagine that they have advanced computer vision recognition abilities onboard so that they can hunt for and find a person without having wireless communication back to the Cloud. Now go further and imagine that the quadcopters act just like ants, that when one hunts and finds food, the others are recruited through stigmergy to join in. Just like being attacked by fire ants, a person could be surrounded by drones in minutes. Or be shot to pieces by weaponized drones in minutes.
Let’s get back to the term stigmergy, which is one, but not the only, type of coordination in a swarm. It means that any communication is done through placing cues in the world. Ants, famously, leave pheromone trails that other ants sense and start to follow. People do this too, we will break branches and place rocks to mark a trail. There is no direct calling out to another member of the collective, and no awareness of whether there is another member even anywhere nearby (but with the huge numbers of a swarm, statistically there will be). Most importantly, there is no need for wireless communication to duplicate this in a robot. And if there is no wireless communication from a robot to other robots or to a controller, then it is very hard to detect drones. And so you have stealthy drones programmed to do evil.
Suarez imagines swarms of small, cheap drones that use electrical-chemical sensors to detect chemicals, duplicating the pheromones of ants. The result is killer drones that impossible to detect until it is too late.
But Suarez knows the solution to killer drones: a ruggedly handsome leader of a special operations squad teamed with a beautiful, feisty entomologist. Love wins over swarms every time!
Seriously, Kill Decision is a great combination of hard science and adventure, combining Michael Crichton and Tom Clancey into an exciting, and accurate, techno-thriller. It is much better than Crichton’s Prey (see the RTSF review here) which was also about swarms.
If you want to learn more about swarms, the section on swarms in chapter 53 in the Springer Handbook of Robotics, second edition, is a good start. Also Suarez has a wonderful TED talk on lethal autonomous weapons at https://www.ted.com/speakers/daniel_suarez
Buy the book or audio book at Amazon by following the below link:
For more information:
- Suarez’ TED Talk on Lethal Autonomous Weapons
- Chapter 17 MultiRobot Systems (MRS) in Introduction to AI Robotics, 2ed
- Chapter 53, Swarm Robotics, Springer Handbook of Robotics, 2ed