The deep learning experts at Alphabet-owned AI company DeepMind started by beating the world’s top chess grandmasters. That challenge swiftly met, they then moved on to tackling the more complex abstract strategy board game Go, vanquishing its most accomplished human players.
Rules-based board games are one thing but DeepMind more recently developed an AI that mastered the computer game StarCraft II. The real time strategy game was an altogether different proposition, involving a myriad of randomly generated scenarios with multiple possible outcomes and consequences. DeepMind’s AI has reached GrandMaster level.
The AI company’s most recent demonstration of what deep learning algorithms can now accomplish has been to beat top level software developers across a series of coding competitions. The AlphaCode system developed by DeepMind went up against 5000 human software engineers, among the best of their profession, and ranked in the top 54%.
DeepMind research scientist Oriol Vinyals says that the significance of the AlphaCode achievement goes beyond the clue it might one day be able to significantly reduce the number of software developers needed to create an application. It takes the company closer to the creation of a fully autonomous problem-solver.
Mr Vinyals said that the fact he had been a huge fan of coding competitions his whole life meant working on the team behind AlphaCode was a “dream come true”. The coding challenges AlphaCode competed in were set by the coding platform Codeforce, which ranks developers based on their scores across critical thinking, logic, algorithms, coding and reading comprehension.
Neither Mr Vinyals or Codeforce’s founder Mike Mirzayanov expected AlphaCode to perform as well as it did across the tests. The latter commented:
“I was sceptical because even in simple competitive problems it is often required not only to implement the algorithm but also — and this is the most difficult part — to invent it. AlphaCode managed to perform at the level of a promising new competitor. I can’t wait to see what lies ahead!”
Despite the impressive results, AlphaCode is not expected to replace software developers any time soon but will, believes Mr Vinyals, help “programmers and non-programmers write code, improving productivity, or creating new ways of making software.”
A DeepMind spokesman added:
“AlphaCode uses transformer-based language models to generate code at an unprecedented scale. It placed at about the level of the median competitor, marking the first time an AI code generation system has reached a competitive level of performance in programming competitions.”
“The problem-solving abilities required to excel at these competitions are beyond the capabilities of existing AI systems.”