Human beings have long been awaiting the predicted rise of artificial intelligence. Since John McCarthy first coined the term at a Dartmouth College conference in 1956, the world has been enthralled by the possibilities – and dangers – posed by the advent of thinking machines with greater intelligence than humans.

After recent advances made towards achieving true AI, particularly in the last few years, we now seem to be on the cusp of achieving technology that is adaptive, self-learning and intuitive.

Legendary futurist Ray Kurzweil, who accurately predicted the demise of the Soviet Union and the defeat of the best human chess player by a computer, has also predicted that by 2045 the growth of AI developments would be so “astonishingly quick” that humans would be unable to keep up without enhancing their own intelligence with the intelligent machines they created.

This may or may not be an accurate prediction, but we already have Siri and Google’s self-driving cars, and this year we saw Google’s DeepMind computer beat Korean grandmaster Lee Sedol at the complex game of Go!

Now, popular apps and platforms that use computer algorithms to translate languages, such as Google Translate, Skype and Facebook, recently announced that they would be converting their existing translation systems over to ‘neural networks’ to increase their accuracy.

Since our field is technology PR, this term is one that is not new to us, as we work with a number of companies making advances in a variety of different fields that will shape the artificial intelligence of the future.

Neural networks: ‘learning’ systems

Artificial neural networks (ANNs) are computer systems modelled on the structure of the human brain and nervous system. Unlike traditional computer systems that are connected in series and refer to large volumes of data to answer a question, ANNs can ‘guess’ answers and learn from their mistakes much the same way as a human might. Instead of executing programmed instructions as a traditional system must, ANNs respond in parallel to the pattern of inputs they receive.

An ANN can be ‘trained’ to learn something, such as handwriting recognition, whereas a traditional system must be programmed precisely using language inputs, which leaves programmers at a distinct disadvantage when it comes to describing things that can vary greatly from one example to another.

Additionally, neural networks do not have separate areas for storing data, but rather store information in the overall ‘state’ of the network. This greatly reduces the size of the network needed, meaning ANNs can be programmed into an embedded chip, such as might be found in a mobile phone. This has already been proven to be fast and highly accurate in applications such as guessing age and gender from human facial images.

As it might be some time yet before we can get reliable, accurate real-time translations as we talk to someone from across the globe, it is still advisable to use multilingual human PR professionals, such as those we have right here at Pinnacle, before trying to communicate to potential customers in other regions.

Even though the pace of improvements may be slow, neural networks are steadily being rolled out across platforms we use every day in our homes, offices and on our smartphones, learning from the things we say and do in order to build the artificial intelligence of the very near future.

Anyone for a tinfoil hat?