Like Mary Shelley’s Frankenstein, Romesh Ranawana could be an unassuming creator of software capable of being commandeered by an artificial intelligence overlord against human opponents.
At his Colombo office, Ranawana is developing software that will be able to control military battalions, fighter aircraft, artillery and other warring paraphernalia in a computer-simulated environment against human opposition. These systems are being used to train soldiers in warfare in a virtual environment as an alternative to the far more costly live exercises.
On a computer, the battle simulation systems Ranawana develops appear like well-designed computer games. But, unlike games where a machine’s reactions to human players are hardwired, some of Ranawana’s war simulation software are intelligent.
They are capable of improving tactical skills on their own to outflank human opponents. Stereotypical creators of monsters too powerful for human control are diminutive in stature, have pleasing demeanor and are uber geeky: a fitting description of Ranawana. Attired in jeans and a casual white linen short-sleeved shirt, Ranawana contends that dystopian views of artificial intelligence are overblown, for now.
“People imagine a terminator robot coming to kill you. That Skynet sort-of super-intelligent AI, better than any human at any task, is a century away.”
The company he co-founded, SimCentric Technologies, to develop war simulation software to be used for training soldiers, however, is in a transformation that will build intelligent battle simulation software able to perform some tasks normally requiring human intelligence.
“I want to build software that controls opposition forces with machine learning,” says Ranawana about battle simulation software. Such a system will make it possible for human soldiers to train by engaging a virtual enemy. The machine learning technology he refers to is an application of artificial intelligence (AI) that provides systems with the ability to learn from experience without being explicitly programmed.
Commandeered by an AI overlord like Skynet, the super-intelligent antagonist in the Terminator movie series battle software may, one day, take over an army of drones and robot soldiers in a real-world war against humans: that’s the dystopian view of artificial intelligence in the future.
AI’s effect on future jobs and whether it will destroy humanity altogether are the two dominating debates as the technology is deployed in many industries.
Although the potential of artificial intelligence was known for decades, it wasn’t useful until a few years ago, when leaps in computing power have made it possible to deploy AI in everyday gadgets like smartphones, digital home assistants and even cars.
Intelligent machines are now good at things like visual perception: cars that can read traffic signs, and speech recognition and language translation available on smartphones. Steady improvements can soon hit a threshold from where it may become possible for machines to suddenly take on many more tasks previously limited to humans.
Naturally, there is growing apprehension of how such an accelerated transformation of AI use in the military, where machines may end up making life and death decisions, will impact humans.
On the other hand, military training and warfare are also undergoing a technology-led upheaval. Machines are taking over some offensive roles, others are changing beyond recognition and new ones are emerging. For enterprising defense suppliers, how future armies will be trained and wars fought are full of opportunity.
S
ince founding eight years ago, SimCentric Technologies has emerged as a formidable supplier of battle simulation software for the world’s major militaries. Its main products train soldiers to target artillery, call for precision airstrikes, add vehicular traffic and people to virtual city streets, plan and execute attacks, and analyse the aftermath.SimCentric’s virtual traffic and people-generating product and battle planning, executing and analyzing software are its newest products.
Instead of costly live artillery firing and air strikes exercises, armed forces in some 30 countries including the US, the UK, other European nations and Australia use growing hours of simulation software training for their soldiers.
Building a machine that can train not just troops but senior commanders on war strategy by engaging them in complex virtual battles can become a game changer for military training worldwide. Its potential can even extend beyond military colleges and training simulators to real-world conflicts when commanders can contest a machine that will role-play the enemy.
SimCentric’s research and software development team of over 100 people is based at Orion City, a campus housing many Sri Lankan technology firms based north of downtown Colombo. Its sales team and chief executive are based overseas.
Despite growing its portfolio, SimCentric’s software covers only a handful of warring operations. New products under development will train soldiers on last-mile battlefield logistics convoys and firing range safety training. Ranawana, based in Sri Lanka, is chief technology officer, while UK-based co-founder Adam Easton is SimCentric’s chief executive. To be successful, they have to overcome a bunch of challenges all based on how realistic and interactive a training environment their technology can create for training soldiers.
They are attacking the challenge from a number of directions. The first is to build an own operating system that can propel the firm towards a contained ecosystem.
S
imCentric’s software requires a widely deployed operating system called VBS. In 2008, when the company’s co-founders, Ranawana and Easton, started developing their first product, the now widely used artillery-targeting software called VBS Fires, it made sense to opt for the industry-adopted VBS operating system. (SimCentric is just one company designing software that runs on the VBS platform.)When the pair found it was challenging to integrate their software to the VBS operating system, Ranawana coded what is called an API (Application Programming Interface), a layer between the operating system and independent software, to make communication between the two seamless. As a result, the company ended with two products, obtaining early orders as many Western military organisations were in emergency procurement mode in 2008 and 2009, as their armies had been deployed in many conflict zones.
Eight years since the launch, both the API, called VBS Fusion, and the artillery- targeting product Fires are among its most successful.
Successfully hitting a target that is many miles away from the artillery gun is dependent on many variables, like the exact distance, the accuracy of the coordinates, wind speed and direction, and relays from forward observers closer to the target. The software is used to train gun crews on the nuances of targeting and the scouts on relaying information to accomplish the task.
Virtual training is also advancing with technology. Soldiers now don virtual reality headsets to train patrol in unfamiliar cities. In some facilities, the environment is projected in a planetarium-like dome so soldiers can patrol without the aid of a computer screen or VR glasses.
Since a year ago, SimCentric has been building its own operating system, which is now nearing completion. If it’s able to convince major customers – which Ranawana says are limited to armed forces in around 40 countries – to adopt its yet-unnamed operating system, it’s market position will be much wider than now. Due to its success, the two founders, who are its only shareholders, have been able to fund development from cash flows. Ranawana says annual revenue is now around $4-5 million, and growing at a steady 25%. “My focus is now to plan the next generation of products,” says Ranawana.
Haridhu Abeygoonaratne, who was SimCentric’s fourth employee, has now been promoted to general manager, taking over some of Ranawana’s former responsibilities.
“My development roadmap in the next five years is going to be very AI-focused,” reveals Ranawana who is holds a DPhil, equivalent to a doctorate, from the University of Oxford. It’s during his time at Oxford that Ranawana met his co-founder Adam Easton who was an Australian army captain who shared an office with Ranawana when they were both engaged in the same course of study.
That friendship and determination to set up a business following their university stint resulted in the two spending a year in their homes coding the first two products SimCentric took to market, the API for the VBS operating system and the artillery product called Fires.
“In our minds, at the time, we were qualified and cocky enough that if SimCentric failed, we had what was needed to get a job. That’s why we took that risk and spent a year at home developing our first product.”
For the next phase, SimCentric is training and establishing a half dozen-strong machine-learning team. As AI is a buzzword and tech investors understand its potential, many companies incorporate AI algorithms that are downloadable from the internet on to their products, even when they add no real value. “There are only few people who understand what is happening inside those algorithms. If something goes wrong, they are getting wrong data and they wouldn’t know what to do.”
Deploying more machine-learning algorithms is SimCentric’s top strategy to address its two core challenges: greater realism and interactivity.
SimCentric’s product that adds traffic, people and activity to virtual streets, called Ambiance, already uses machine-learning algorithms. “Machine learning is good for optimization; meaning, when you want the computer do something that you haven’t told it before.”
When a soldier trains on the Ambiance product wearing VR goggles by simulating a troop patrol on a street in Kandahar in Afghanistan, the system generates people and traffic. For the soldier, who may be preparing for deployment to these very streets in days, the experience would be realistic if people and traffic responded in the way they do when a heavily armed street patrol is encountered. Similarly, if the solder decides to talk to any passerby on the street and that virtual character would be able to provide a relevant response, the system may be considered interactive. “A normal computer programme can’t do that. This is where machine learning is useful.”
Often, soldiers train in simulators not offering realistic and interactive feedback. If a drone strike is being practiced, chances are there won’t be traffic or people who bring realism and force the trainee to avoid real damage. Adding traffic and people to virtual streets is called ‘pattern of life’ in the industry. SimCentric has secured the contract to develop the global ‘pattern of life’ standard, which they will unveil soon.
“Others can take this core engine and develop their own ‘patterns of life’.” Ranawana emphasizes that SimCentric isn’t developing new AI algorithms, but is beginning to conduct some fundamental research.
His benchmark is the Google-owned company DeepMind’s ancient Chinese board game ‘Go’ playing AI software that beat one of the game’s grandmasters a couple of years ago. To achieve this, the computer, named AlphaGo, used a machine-learning technique called reinforcement learning. A machine learning to master a board game like Go will play itself millions of times, learning from its mistakes each time it loses. Unlike unsupervised learning, which requires loads of data, the system gleans to gain an understanding, insights and learning; reinforcement learning requires no data.
“Reinforcement learning applied to the battle simulation will master tactics to outwit their human opponents, leading to an intensive training experience. With simulations, for us, it’s the ability to not have to explicitly programme something. Instead, we create a system that keeps adapting according to how the simulator works. Anyone can hardcode; for us, it’s about creating the environment where it will keep adapting; that’s AI’s potential.”
[pullquote]AI’S MAJOR LEAPS OVER THE LAST FEW YEARS ARE RAISING QUESTIONS SIMILAR TO THOSE THAT AROSE AT THE DAWN OF THE INDUSTRIAL AGE[/pullquote]
M
any westerns militaries are stretched, and their best tacticians and trainers are deployed in conflict zones across the world. Training recruits in military strategy and simulator practice before deployment are challenging due to these constrains. Military planners are eager to procure intelligent battle simulations to overcome these shortcomings and constraints. For instance, a live artillery firing drill could cost up to $30,000 a time, whereas time in front of simulator can minimize the number of live drills necessary for training.For SimCentric, the emerging challenge has been the slow sales rate as militaries, no longer on emergency procurement mode, take years to sign off on a deal. SimCentric running out of customers to sell the same product to is driving its development of new products.
Like most technology companies, SimCentric used to also pursue technology for its own sake. “The big lesson we learned later was to start from the end state,” admits Ranawana. How is the customer going to use this? How is he going to download this and install it? How is he going to learn about it in the first 10 minutes? “First, sort out all those, and then work backwards to see what tech we need to achieve this.”
Ranawana says he now challenges developers with the premise; they have a one-hour window to impress a customer who has downloaded and installed the product. If they encounter problems within that hour, they will switch off and judge the company based entirely on that experience. To get them interested again is difficult.
AI’s major leaps over the last few years are raising questions similar to those that arose at the dawn of the industrial age. Fears that machines will replace humans at work proved to be overblown then. Human-like ability called general intelligence, experts say, is beyond machines’ capability so far. However, computing power is advanced enough for machines to better humans in narrow areas and take over mundane tasks.
Ranawana may not yet have created a set of war tech products that may suddenly get out of human control, at least not for some days more he thinks.