AI Summit: Why are some industries struggling to deploy AI?
Two experts at the AI summit during London Tech Week examine industries which are flourishing with AI and those which are struggling – plus how best to solve the data problem.
AI Summit: Why are some industries struggling to deploy AI?
From the self-scanner in the supermarket to driverless vehicles and chatbots, there are already working examples out there where artificial intelligence (AI) is replacing humans.
However, data is proving to be a huge barrier to entry in some industries, either because it’s siloed, hard to access or too difficult to translate into AI.
Speaking at the AI Summit in London last week, Richard Self, senior lecturer at the University of Derby with a focus on researching AI, and Ravinder Singh Zandu, a modernising technology programme manager at the Cabinet Office, discussed how they see AI working in industry.
Currently, in his course at the University of Derby, Richard Self looks at the governance of advanced and emerging technologies, looking at how businesses take on board new technology effectively and deliver value from it, including AI.
One report he’s working on is AI’s success and failure, which he says mirrors the IT research firm Standish Group’s 1994 ‘Chaos Report’ of what success and failure looked like for ordinary IT.
Currently, even after almost two decades, the university’s study found that only 35% of IT projects are successful for enterprise, while only 12.5% of AI projects are successful.
While research is still ongoing, Self observes that what is becoming clear is that analytics AI is working best in science-based sectors such as technology, engineering and manufacturing where there is a clearly defined problem and a clearly defined outcome.
The panel agreed that manufacturing, which has undergone huge swathes of digital transformation, has taken to AI seamlessly because different manufacturers share common problems that are solved by using a similar approach, meaning that AI can learn from shared datasets.
However, in academic terms, industries related to the social sciences are struggling with AI, “because the data is based around loose perceptions, words, and diagnoses.”
He used the example of visiting a doctor to obtain a diagnosis. If a patient were to see three or four different doctors, their notes would certainly be worded differently because each doctor will see the issue from a different perspective and there are multiple ways to phrase the issue.
Richard Self explains that AI technologies cannot cope with this subjective data that humans generate. Which covers not only GP surgeries but also HR systems, where the data is not simply numerical.
Self warns that business decision-makers should not look to AI to solve their solutions because it is “the shiny new technology”.
Now, “everybody’s diving in on it because it’s the latest thing”. However, if there’s not enough data, sometimes instead of looking into an AI solution, “it might be bog-standard, ordinary, anything else technology”, which is best to solve the issue.
According to the panel, having a continuous flow of updated data is also important otherwise the AI model is obsolete. Which is something that 89% of CIOs know, the panel voted. However, 52% of CIOs fail to know what the correct data is that they are even looking for, according to Zandu.
A general issue for industry is that many cannot access the data they need to create AI tools in the first place.
Zandu, speaking on behalf of the Cabinet Office, said that the UK government is promoting open data to help solve this, and trying to open government data for small-to-medium enterprises (SMEs) to work on and build new products and services for UK citizens.
Subscribe to our Editor's weekly newsletter