Thought of the fortnight: AI/Chatbots dos and don’ts

We have just launched a very successful AI-powered assistant called Nova. It is 98% accurate and incredibly impressive. It scrapes information from our internal and external websites, so the key to making it work – and where most of our effort was spent – was making sure all the information on our websites was accurate. An AI assistant is only as accurate as the information you feed it.’

‘We have just launched a very successful AI-powered assistant called Nova. It is 98% accurate and incredibly impressive. It scrapes information from our internal and external websites, so the key to making it work – and where most of our effort was spent – was making sure all the information on our websites was accurate. An AI assistant is only as accurate as the information you feed it.’

‘With chatbots – know when they need to give up to direct someone to a real person. If it doesn’t provide the answer after a couple of goes, it will just add frustration. And a chatbot should be upfront that it is not a real person! Find a feedback mechanism to tell you if it is hitting the mark.’

‘We designed a chatbot and spent huge amounts of time working through its personality, how sophisticated its understanding of increasingly complex and interlinked processes might be, and more. The reality was we should have started with a simple, rule-based chatbot on a redacted range of signposting but with hybrid functionality for other chatbot types meaning we could have later built out the more sophisticated potential. Had we got the basics right first, we would have implemented it quicker (with less resistance), experienced lower error rates, and really sharpened up the quality of a core set of functions. We could have done the more sophisticated work to ‘build out’ the functionality and personality later having established confidence.’

‘In truth, we have all reached a point where we know AI will feature in our future operations but we struggle with the capacity and financing to drive the real exploration of how it might be embedded in our work and to what impact. We know we want it to reduce our administration cost per capita and enhance the quality of operations but delivering that needs upfront investment of time and money that we are struggling to create.’ 

‘The level of data in universities is vast and the majority of it is unstructured. AI should be the perfect way to tap into the goldmine of information that exists which we previously didn’t know how to manage and we can use it to start to understand what it tells us about our strategies and operations. The biggest challenge in making that leap probably isn’t the technology but our organisational competencies and capabilities.’ 

‘Never believe artificial intelligence can replace natural human intelligence. It can’t but it could add to, complement, enhance and manage the human capital our institutions have. It was when we started to explore how AI could enhance our existing teams and work streams rather than replace them that we unlocked some more practical, realistic and higher impact opportunities.’

‘We all know AI will be an incredibly important feature of the future and yet we feel a long way behind other industries and businesses in exploring its potential to shape our organisations. This seems to come from a good place, our determination to think about the ethics of AI before adopting it unvetted, but unless we can solve that quickly we will be left behind and miss the opportunities it offers.‘

‘I wonder how different our needs would be from any other university and, if there are some processes, systems and decisions that follow similar pathways, perhaps we should be jointly commissioning the decision of sector chatbots rather than all trying to create our own?