How much ‘human’ work can machines really get done for us? Can we ever make computers that will 'think like us'?
I’m lucky enough to be able to grapple with these problems as part of my day job here at Capgemini, exploring how automation and Artificial Intelligence (AI) can be applied to the challenges of Business Process Outsourcing.
We’re already used to seeing robotics in many core areas of industry (just think of a modern car production line – they’ve been using robotics en masse since the 80s) and it’s now gained a foothold in non-core back-office processes too (think along the lines of Accounts Payable processing). But just what ‘human stuff’ will machines be able to do for us in the near future, and how should we approach this fascinating challenge?
Being nearly human
Back in the early years of computing, Alan Turing foresaw some of the fundamental questions around AI that still preoccupy us now. The Imitation Game, the recent film about his work and life, takes its name from a test he devised to explore the problem.
Today’s well-publicised ‘Turing Tests’ still attract plenty of attention, as computers attempt to masquerade as humans and outwit a jury of real people. Although there's plenty of debate about how useful these are as a true test of intelligence, they're a good starting point here.
Even in Turing’s time it was clear that computers would struggle to imitate humans in a couple of key areas:
- Being just too perfect in their performance – in Turing’s words, “unmasked by their deadly accuracy”.
- Being unable to deal with the unexpected – struggling to innovate their way beyond the limitations of their original programming (although, this is perhaps not so different to a low-skilled human resource).
These days, we can create computers that are better and better at mimicking us but they still don't really work or learn like us.
To think practically about AI, we need to think about the way we humans solve problems at work.
Solving business problems the human way
When it’s something simple, we can follow basic rules to fix it. Think of inputting and processing details like those in supplier invoices (tax, due dates, matching the right documentation and approvals). This is already a relatively easy area to automate.
What about handling the exceptions, when there's not an obvious next step?
When something unexpected happens, as humans, we tend to ask our boss or a colleague with more experience for help. Most of the time they’ll use that experience to solve the problem. In some cases they may also be flummoxed and then ask someone else. If the answer still eludes them, they will then get round the table with others and try to ‘work out’ a novel solution for this novel challenge.
This collaboration applies a complex mix of human imagination and experience, and applies it to a new context. I suppose this is what we call ’thinking outside the box‘.
As humans, we’re good at this. Our brains are pattern hunters – but we also enjoy looking for and making unexpected connections and building new concepts in this way. It’s what makes us human.
But crucially, we’re not always right. Experimenting and getting things a bit wrong (while judging how much risk to take) is all part of the process. This is a world of grey areas; the solutions we arrive at in these cases are essentially subjective – we could argue for and against them, but we eventually agree to get on with it and give it a try.
Solving business problems with Artificial Intelligence
AI systems get just as flummoxed as people, and they also need to be taught how to deal with new circumstances. Perhaps by allowing them a little more 'Artificial Stupidity' and letting them act a little more randomly, we can give them the opportunity to learn from success and (controlled) failures as we do. Compare that to today’s Robotic Process Automation that simply repeats the same tasks to set rules, which leaves little opportunity to learn and evolve.
I don’t think we’re close to a time when computers can replicate the human intelligence needed for more nuanced business analysis and recommendations. This is probably a bit of a red herring anyway – even if it does raise fascinating technical and philosophical questions.
However, we are already seeing computers encroaching further on the ‘subjective layer’ of business process decision making. Almost all high-value work performed by humans today is augmented by increasingly complex tools. The technology is making more and more intelligent ‘suggestions’ which we then take a measured final decision on, based on all of our human experience and intuition.
In that case, perhaps we should think of AI in terms of a dual-pronged evolution – with humans a key component in the system. This will soon be far more common for medical diagnoses – and has already been done in chess, led by Garry Kasparov.
Putting it into practice
These are exciting times to work with AI. It's rarely far from the news or the big screen these days, while predictions for the long-term future range from the utopian to the apocalyptic. The reality may turn out to be a little less dramatic (hopefully!) but the critical point for business is that any approach must be intrinsically linked to value creation – whether that’s reflected in speed or quantity or other metrics.
I believe there is the potential for a really natural fit with modern BPO. Just how we and our customers go about getting the best from it must wait for another blog!