22 July 2019

AI | More Than Human - Top take outs

A few weeks ago, a group of us visited AI: More Than Human at the Barbican. It was a journey on the history of AI; from the creation of the first neural network in the 1940s, which copied the brain’s processes and developed into machine learning, to examples of how far AI has come today, including DeepMinds AlphaGo, the first computer to defeat a professional in the Chinese Strategy game ‘Go’ in 2016.

A few take outs…

AI only works if there is enough robust and unbiased data to start with – There were a few really interesting examples of AI having bias, including facial recognition technology not recognising the faces of underrepresented genders and races. As the software hadn’t been trained with enough diversity, it could only recognise certain ethnicities and face shapes, therefore drawing incorrect conclusions. Take a look at ‘AI Ain’t I a Woman’ below.

The key learning is that the more data is shared, and the more robust and representative testing is, the more accurate AI can be. This is something that Brandwidth are well versed in. When creating our Voice App for the Housing Sector we acknowledged that the depth of knowledge and ability to understand user commands would only be good enough if we involved a large representation of people to test, as invariably different people can ask the same question in different ways.

AI: Ain’t I a Woman

The power of AI in healthcare is huge – Using ‘tech for good’ is an area that I am always excited about and there were lots of examples of how AI has the power to analyse data and interpret trends far quicker than the human eye, therefore having the ability to potentially predict diseases a lot earlier and ultimately save lives! AI software can diagnose cancer risk 30 times faster than a human doctor with 99 per cent accuracy. Plus, developments like Organs-on-Chips technology – a microchip lined with human cells and tissues – open up new understanding of how different diseases, medicines, chemicals, and foods affect human health. By directing the latest advances in AI to streamline drug discovery, there is potential to significantly cut both the time to market for new drugs and their costs.

We can’t always tell if AI has been used (for good or bad) – At the exhibition there was a game in which the aim was to guess whether extracts of text were written by a computer or human, and it proved surprisingly difficult to tell the difference. There was also a video which showed how analysing clips of Obamas speeches and applying a learning algorithm meant that a new, entirely fake video could be edited together whilst looking and sounding extremely real.

Fake Obama created using AI Tool

As AI becomes ever more sophisticated, in some instances it will be difficult to tell what is real and what is not. Can you tell which face is real? You could argue that in some cases you don’t need to know AI is being used, such as if the use of AI is improving your experience (think ‘smart replies’ in Gmail or Spotify music recommendations). On the other hand, services like this money saving app actively promote the use of AI to monitor spending patterns and help you save you money effortlessly.

The key is not to mislead the end user where there could be doubt. At Brandwidth, we believe that when AI is being used in voice, chatbots or any other engagement interface it’s important that the user knows they are not talking to a person. We encourage our clients to be open about the fact that a chatbot is not a customer service representative and although the software should be as helpful as possible, users shouldn’t believe that they are talking to a person as this opens up areas of risk as well as the potential for expectations to not be met.

The exhibition finished with ‘What a Loving and Beautiful World’ – a digital installation that responds to visitors shadows and was a fun way to end the evening. Unfortunately we didn’t make it to the Robotic Bar, instead opting for drinks outside on a sunny evening, but we did leave feeling inspired, proud of the work that Brandwidth is already doing in this field and excited for what is to come in the world of AI.

Will a humanoid robot be able to win a game of football against the human FIFA world champion in 2050, a goal set by the founders of RoboCup? Personally I am less interested in that, but more how AI will continue to transform organisations and our daily lives for the better.