What's Stopping Us Achieving Artificial General Intelligence?

A. Efimov, D. Dubrovsky, and F. Matveev explore how the development of Al is limited by the perceived need to understand language and be embodied.

- A. Efimov, D. Dubrovsky, and F. Matveev

Over seventy years ago, Alan Turing developed the simple but powerful idea that any solvable mathematical problem can in principle be solved with a ‘universal computing device’. The type of device he described in his 1936 paper became known to researchers as a ‘Turing machine’. Ever since we have been trying to create artificial intelligence by programming electronic machines. Most of the current research in the field of AI is indeed just an acceleration of that first universal Turing machine. Turing is also responsible for another fundamental idea that has shaped research in this area. The Turing test makes us ask: if we cannot distinguish whether we are holding a dialogue with a person or a machine, then does it really matter what is in front of us – a machine or a human – since we’re dealing with intelligence anyway?

The Merriam-Webster Dictionary defines intelligence as ‘the ability to learn or understand or to deal with new or trying situations’. Turing’s idea of using language as a tool for comparing machine and human intelligence, based on how well a machine can pretend to be human, is both simple and profound. Thanks to this idea, such wonderful things as voice assistants and online translators have come to life.