You’d probably be right if you said the IT industry has itself to blame. But I sometimes worry that Artificial Intelligence has been given a sort of needless mystical sheen. It’s as if AI is somehow beyond the capacity of mere mortals to understand or leverage. This sort of bamboozling might be helpful to somebody in the short term, but in the long term, it's doing everyone a disservice.
I was thinking about this after taking part in a recent panel discussion about the monetization of data and AI as part of the Data Innovation Summit 2020. Given the theme of monetization, we were definitely focused on the real stuff. From my perspective, there are four down-to-earth issues that are unnecessarily holding back many successful AI implementations.
- No data, no AI
First, there can often be an underlying expectation that AI is a magic wand that can solve any issue. That's simply not true. Successful monetization depends on having the right quantity, quality, and location of data. If those things aren't right, then it's going to be a struggle.
Take, for example, a recent project we worked on for a major retail hospitality chain, where Fujitsu has been a long-term partner. The last thing any retailer wants is a bottleneck in accepting money. We identified an AI-led predictive maintenance project intended to reduce cash register breakdowns. We had relevant data going back over several years which meant it was possible to identify which error codes were reliable predictors of future equipment failure. With the right data to hand in enough quantity to make reliable predictions, this was a quick-win AI project for our customers. The result was that it reduced maintenance costs and improved cash flow, as well as wider positive implications for backup and audit trails.
- AI is not just about AI
Even when good quality data is readily accessible in adequate quantities, it’s not necessarily going to be enough. There is a bigger picture that can sometimes be a blindspot for technologists, who can be guilty of believing AI is all about data and systems. That’s a recipe for potential failure because the wider business context is likely to be more critical.
For example, we were working on an AI app at a major investment bank to minimize the many false positives that bedevil current approaches to fraud detection. The success of the project triggered a request to look at AI for the automation of insurance underwriting. On the face of it, you would imagine that’s a slam dunk, with all the quality data on hand you could possibly hope for. And yet the project never got off the ground – because it was simply too disruptive for the organization. By not addressing the political dimension, viable AI improvements can be destined never to get off the ground.
- AI ideation is not difficult. Just ask
Coming up with suitable ideas for AI projects is not as difficult as it's made out to be.
So, where does AI ideation come from? Some organizations need little or no help here – they are setting the pace and we know who they are. Others, mostly smaller enterprises, have no likely need in the foreseeable future – hairdressers, for example. But the rest – a very large group in the middle – often do not know what data they've got, where it is, or what on earth to do with it. They are often actually frightened by data and when you consider some of the regulations in play that’s perhaps not too surprising. New German data laws, for instance, require an individual’s data to be portable to whichever jurisdiction they are currently in. Work out how to do that, if you can.
The upshot is that there are a lot of organizations stuck in the muddled-middle, with a strong sense of AI fear of missing out, but unsure where to start. There’s absolutely no need for this. We can easily pull the rug out from underneath the fear-mongering that ideation is something only cognoscenti can pull off. The simple fact is ideas that work comes from people who work on the relevant systems and processes day to day. Get them together with the people who know how to build AI systems and you will generate viable, monetizable AI projects. And it’s particularly important now that "water cooler moments" are in such short supply because everyone is working remotely.
Fujitsu's co-creation approach recognizes this reality. It’s something we have been advocating for many years now – because it works. One of the most powerful things I have seen in my 40-year IT career was in a co-creation workshop we held in Stockholm a couple of months back. A young data scientist from a Polish start up developed – in a single afternoon – an app to read architectural blueprints. Prior to sitting down, she knew nothing about architecture or technical drawing, but by inviting her into the same room as people who did, she was able to deliver a huge advance very quickly.
- People who need people – aka the war for talent
I’ve left the most difficult until last. By far the most intractable AI issue is the lack of people who know what they are doing.
What sort of skills do those people need and where do you find them? In fact, the term “AI” is a complete misnomer - it's neither artificial nor intelligent. AI is just statistics. In the absence of a reliable stream of graduates coming out of universities with usable AI skills – and there is a real gap here – you should therefore turn to people with solid mathematical skills. Except they are thin on the ground too and unlikely to be plentiful in the accessible pockets of IT literate employees.
There are, of course, gradations here. You don't need people with deep-math skills for every problem, nor do you need a strategist for lots of others. In the longer term, organizations should be developing strategies for team structures to fill these skills gaps – following the example of my colleague Rupert Lehner – who has been working on recruiting a new generation of mainframe engineers for several years now. This is a longer-term play because finding the right people in the short term is only currently possible by working with partners. In fact, given the disparate strands that come into play in AI projects, this is very likely to mean multiple partners, making a reliable ecosystem – and ecosystem management skills – highly relevant.
Fujitsu is known to be really strong in these areas – co-creation and ecosystem management – and can be relied on to provide a common-sense perspective on whatever project you are considering, or possibly have not yet considered considering. If that’s the sort of conversation you’d like to have, check out Fujitsu approach on Data-driven transformation.