Simple AI can achieve significant results
Today I’m inspired by a recent Harvard Business Review article entitled “Competing in the age of AI (artificial intelligence) : How machine intelligence changes the rules of business”. It provides an analysis of AI lessons learned from digital firms with no limits to scale. One quote resonated strongly with me:
“Oddly enough, the AI that can drive the explosive growth of a digital firm isn’t even all that sophisticated… AI doesn’t need to simulate human reasoning (strong AI) ….you only need to be able to perform tasks traditionally handled by people – what is often referred to as “weak AI””.
That got me thinking about a lot of the time I have spent in functions that involve analyzing and manipulating data. A lot of that work doesn’t need strong AI – a lot of that work could be characterized as manipulating data and presenting information to governance forums to drive and influence decision making. Often times the information we need exists in silos or in systems that do a poor job of presenting information in compelling ways.
While the future of AI is indeed inspiring; for those of us currently sitting at the keyboard – we need to shape our own digital data destiny. One effective way to do that is to try it yourself.
Current Apps offer huge potential to augment our abilities – even without AI
Inspired by my 9 year old daughter who has been learning coding using an MIT developed language called Scratch, I started to think about some of the analytical challenges I faced in my line of work. I dived into VBA programming – leveraging library books, online searches and the help of a friend who had written macros. Within a month – I went from zero knowledge of Excel VBA macros – to something approaching ‘good’. I was able to automate something that would have taken me 30 mins to do manually on each occurrence and instead do it within 1 second (more to come on that in a subsequent article).
This helped me realize that for many of our challenges, we don’t necessarily have to rely on six or seven figure software solutions; or waiting for the Strong AI that can read our minds. Many of our challenges can be improved upon using existing tools that themselves are rapidly advancing. New analysts in some companies take python courses. We can all learn these concepts ourselves.
Here’s the result of my first serious piece of coding in VBA illustrating some of the sophistication that can be achieved with low cost tools today. It takes less than one second to covert static data about risk into a compelling visual representation. Radar bands represent degree of immediacy of the potential risk. Data used is hypothetical – not attributed to any organization.
Your own personal Lexicon Search
Another use case I built was to create my own function in Excel to search for key words I defined. Effectively my own lexicon analysis tool.
Organizations spend millions of dollars licensing lexicon search tools for email. The so called “unstructured data” problem. However doing this yourself with VBA is actually quite easy – and then you can apply it to any unstructured data problem you come across.
Tired of reading thousands of words in a typical annual report or 10k filing; or even just analysing the thousands of controls a typical organization runs. Then enhance your abilities (#AugmentedBrain) by developing your risk scoring algorithm.
My own example here, helps me hone in on those areas of 10k’s which may be more important for my review. As you can see – I’m able to generate a risk score based on my subjective assessment of that risk. In my tool, “Material Weakness” gets the highest risk score. That helps guide me to the paragraphs I’m more interested in.
I’m currently well into this book which dispells many of the myths of AI today – especially the misleeding way AI progress is reported today. Check out the book by Gary Marcus and Ernest Davis.
Firstly AI to date is really good at looking at correlations from large volumes of data. This means great search results (as in from Google) as well as huge progress on Google Translate (which is trained by looking at millions of examples of the same source material in multiple languages).
AI is also great at playing rule based games millions of times. That’s why AI can beat world champions at Chess, Go and even Jeopardy. This is brute force learning across millions of games.
AI is not great at those things that are not statistically frequently occuring. Or things that haven’t been seen before. AI today doesn’t actually understand text (or for example the characters in the text). Unlike humans, it can’t read between the lines.
All of the above makes me realize that in my field (Operational Risk) – there’s not much danger from AI today on the high value work we do, which is to focus on the possibility of remotely occuring events – and even events that haven’t happened yet.
But there’s huge potential for technology (including AI) to automate and reduce the time we spend on low value add bureaucratic elements that come with the territory. As this article begins with – there’s a lot of low hanging fruit from not just simple AI but non AI automation.
Tell me more about your experience in automation
For those of you on this path – which tools do you use that have the potential to transform at low cost? What have you achieved with these tools? What’s the art of the possible in the risk landscape today!
In the year 2020 – do you have a 20/20 vision to shape your personal digital destiny? How are you using these tools to lift your capabilities to do more, something I call #AugmentedBrain