3rd August 2017

By Karl Roberts – Head of Propositions, GCI.

Waking up yesterday morning you would have thought that the machines truly had risen. In response to Facebook shutting down an AI experiment, The Sun quoted experts calling it “incredibly scary” whilst The Mirror stated how “Robot intelligence is dangerous”.

Facebook was seeking to push the margins and make chatbots more useful. Many chatbots used for customer services purposes use natural language processing to query knowledge bases – they’re useful but relatively limited in terms of functionality. By ‘teaching’ ‘Alice’ and ‘Bob’ to negotiate and trade hats, books and balls, Facebook was attempting to make chatbots more autonomous, intelligent, and potentially more useful to end users.

The bots were programmed to experiment with language in order to see how that affected their dominance in the discussion. However, this had the effect of the bots competing against each other and ‘shouting’ in out of context English, effectively reinventing a new version of the language that only they understood.

As the BBC suggests however, this was less about the bots inventing a new language but more an explanation as to how neural networks had modified human language for the purposes of more efficient interaction.

What have we learned? Firstly, it reinforces something that we know already – getting chat bots to think and act in natural language is hard. We’re getting there to some extent. For instance, we’re now at a 95% accuracy in terms of speech recognition but getting to the last 5% is the hardest part given the complexity and nuances of human language. However, once we get to 98% it truly becomes a ‘game over’ in terms of changing the old ways that we interacted with systems.

AI is currently poor at empathy as robots don’t have emotional intelligence. But experiments by Facebook and others will help us get them there. We should support their endeavours since we’re still at its infancy in terms of what’s possible in with AI but it’s already changing the world we live in. Look at IVR systems which have now become second nature to use. Siri, Cortana, and Alexa are helping us navigate our daily lives. Amazon is helping us guide our shopping purchases and Netflix, what films and TV to watch. Better AI helps us all. An overreaction that suggests that robots are about to ‘take over’ (or similar) potentially holds this important research back.

Last week GCI published a 90 Second Interview with Karl Roberts where he explained how artificial intelligence will herald the rise of the ‘super-agent’, also stating that artificial intelligence is still in its infancy in terms of what’s possible; clearly demonstrated in this article. If you’d like to read Karl’s full interview, please click here.

Karl Roberts is Head of Propositions at GCI.

He heads up propositions for GCI, and has been involved with Voice Communication and Customer Experience technologies for 25 years. Karl has digitally transformed the largest Cloud Contact CX platforms in Government, designed Europe’s first passive biometrics banking service and pioneered ideas around intelligent meta-bot interactive assistants using Echo and Skype as a personal tax assistant.