Machines can now solve the Rubik's Cube

• United Kingdom

• Technology, Media and Telecoms - Disruptive Technology

05-07-2018

CNET recently reported that yet another milestone has been reached in the quest for ‘general’ artificial intelligence, as a machine engineered by a team of computer scientists from the University of California, Irvine (“UCI”), has successfully taught itself how to resolve a Rubik’s cube, without human intervention.

How does this work?

Unlike Chess or Go, the best possible next step to solving a Rubik’s cube is not immediately obvious and hence UCI’s machine needed to adopt a pioneering technique, known as “autodidactic iteration”. Instead of relying on step-by-step algorithms devised by humans (as per previous machines), autodidactic iteration enabled the machine to consider a completed puzzle and then work backwards (thereby establishing whether the proposed move would result in a configuration that was closer to the solution).

UCI’s machine achieved remarkable success, not only by solving every Rubik’s cube presented to it, but by doing so in consistently fewer moves than those relying on algorithmic techniques (including us, old-fashioned humans).

Solve complex solutions in fewer steps

Narrowly, the success of autodidactic iteration represents a significant step towards solving other, more complex problems, faced in circumstances where humans are unable to reliably and/or efficiently discover the optimal steps amongst a finite set of options. A simple application might result in a machine which determines, say, which taxis amongst a fleet should pick up particular fares. However, more broadly, this breakthrough represents an important step for AI and machine learning, by creating a framework for machines to make efficient choices which may ultimately pave the way toward the holy grail, of ‘general’ intelligence.