Doug Bonderud

Jul 12th 2017

Battlecode: Banking on Bot Brilliance?


What’s the value of adaptive artificial intelligence (AI)? That’s the idea behind Battlecode — MIT’s long-running code competition which pits teams of AI-controlled units against each other on a virtual battlefield. Each year, developers tweak the rules to improve balance or introduce new elements but the premise remains the same: design AI that can outthink and outmaneuver the opponent to win big. But what’s the big-picture impact of this bot-based battle?

Battlecode Basics

The MIT competition got off the ground 17 years ago. The concept is simple: Players write code for teams of virtual robots, designing AI which allows units to coordinate attacks, develop defenses and ultimately own the battlefield. In addition to the once-a-year competition, Battlecode is also an MIT course (number 6.147) which offers a walk-through for beginners and a newbie-only tournament. Advanced competitors (in teams of one to four) compete throughout January and February in Sprint, Seeding and Qualifying tournaments for the chance to play live on stage during the finals and share more than $50,000 in prize money.

Evolving Iteration

At its most basic, the competition has teams creating AI code which tells units how to act under specific circumstances and respond to emerging threats. Historically, all units were able to “speak” freely to one another and share “sight” of the map at large. But in recent years developers have limited this ability, changing the rules so that only specific units can send out messages — the more complex the message, the more time it takes to send — and units can only see their immediate environment.

Recent iterations of the competition also saw the introduction of a “third” faction: zombies. These units attack anything that comes into range, and if AI-controlled troops are killed, they respawn as zombies. This change prompted the development of a new strategy. Team Foundation, which placed fourth in 2016, used low-health, high-speed scouts that pulled zombies to the enemy base and then died, causing mass chaos for opponents.

Advanced Tactics

So why Battlecode? With so many other coding competitions now available for tech students looking to hone their skills — such as CyperPatriot or Google’s Code Jam — what’s the appeal? As noted by one former winner, Cory Li, the “experiences and skills learned through Battlecode are hugely relevant in software engineering, startups and life in general.” And according to GCN, these “war games” can also help improve decision-making abilities of agency managers, financial analysts and cybersecurity staff.

Consider the new strategy designed by Foundation. According to “that one blog,” a postmortem account of the 2016 competition from first-place team Future Perfect, the dominant tactic in a zombie-infested landscape was “turtling” — using turret-based defenses to hold off enemy attackers. Other competitors designed mobile-turret strategies but Foundation’s “very scary” zombie-based attack quickly overwhelmed even top-seeded teams in record time.

Team 1064CBread, meanwhile, discovered that it was possible to leverage the game’s messaging feature to win battles. Here’s how: Since robots could receive messages from both enemies and allies, 1064CBread built an army dedicated to spamming the virtual airwaves. Enemy units with their receivers turned on were effectively jammed turn after turn, making them easy prey for zombies, while 1064CBread’s bots ignored the messages and survived the assault. Eventually, developers were forced to patch the strategy by reducing the number of messages units could send each turn.

Bottom line? While Battlecode’s bot brilliance doesn’t yet directly inform real-world applications, it serves a critical purpose: empowering the next generation of coders to think outside the box and create never-before-seen strategies. From enterprise efforts to military applications, taking a pass on predictability is critical in obtaining a competitive edge and increasing ROI, or outmaneuvering opponents to minimize losses and reduce total resource commitment.