To date, cyber-attacks have been initiated by, and ultimately controlled by, humans. This is akin to the television combat competition, Robot Wars, where the robots are effectively the technical tools we use for offence or defence activities. However, we may be moving into new territories; artificial intelligence systems attempt to take advantage of modern microprocessors speed to run software designed to search through large data sets and “learn” patterns and develop ad hoc rules based on those discovered patterns. This can be done at speeds that simulate or exceed the human brain’s capacity – which is the real objective.
As an example, in 2016, the artificial intelligence / machine learning system ‘AlphaGo’ beat the world’s top ranked player at the game of Go. AlphaGo was trained in the rules of Go and given the details of 1000’s of real human-played games to learn from. Through this, the system was also able to produce generalisations which could be used to develop and perfect new strategies. In the subsequent year, AlphaGo’s successor, Alpha Go Zero beat its predecessor 100 times in a row; unlike AlphaGo, the system received only the basic rules of Go, without data from human games, and thus worked out how to play Go from the rules and by playing itself at Go. Alpha Go surpassed the playing (and winning) capabilities of all previous version of Alpha Go in 40 days.
As the technology continues to develop, instead of being directed by a human controller, the robots in Robot Wars will one day be able make their own decisions and act autonomously. The same technology is perfectly capable of being the sole agent in the next generation of cyber offence and defence – that is, launching attacks and conducting defensive actions without human input. Indeed, as AlphaGo and AlphaGo Zero have demonstrated, once an organisation successfully develops AI driven cyber capabilities, it is highly unlikely that humans would be capable of competing against the speed and knowledge base of AI systems. We may be entering a new era, wherein instead of being just a part of our toolset for launching (and defending against) cyber-attacks, the technology will be the standalone vector.
The field of AI is developing at an exponential rate. Machine learning is already a mainstay of networking and endpoint security solutions; it is only a matter of time before AI is incorporated into the arsenal, potentially driven by improved offensive capabilities. Once the starting gun has been fired, the subsequent combat and arms race is likely to be exclusively between machines. Could we be at the cusp of true Robot Wars?
Dr Wendy Ng, CISSP, CCNP; 20th November 2018