I recently read the article The Simple Economics of Machine Intelligenceand the book Prediction Machines. It got me to think – the technology geek in me was of-course shocked at the over-simplification. And while I accept that the current state of AI is nowhere close to the myth of super intelligent machines, I still think that the reality is somewhere between the two extremes.
Five years ago, it was all about data and insights. Data mining algorithms were designed to determine patterns, correlations and convert data into insights and advanced machine learning algorithm predicted value (through ranking, relevance etc.) and recommended actions (using classification, context and value). Quality of Prediction was highly dependent on the richness and relevance of data and the buzz was about transforming data into informationand ultimately improving the decision making process. Better insights improved the context, and as the accuracy of predictions started to improve, it lead to more holistic perspectives– all this started slowly to make the difference, and the conversations gradually shifted to predictions.
If you sift through the AI hype, most of the products today sit in this space. No wonder the economists are reframing AI as cheap prediction. They argue – the economic shift will centre around a drop in the cost of prediction (and an associated increase in the value of judgement) – which will in turn lower the cost of activities that were historically prediction oriented (like inventory managements, demand forecasting), as well as open up opportunities to use prediction as an input for things for which we never previously did (like navigation which now has moved out of the rule based programming framework to a machine learning based prediction framework that predicts human response to changing environmental data).
But limiting the view to predictions (and keeping actions triggered on judgement) out of the purview seems like an incomplete and a shortsighted view. It is like saying that we will replace bits and piecesof our existing processes by newer technology. It’s an oversimplification of AI and risks losing the bigger picture – the opportunity to reimagine business processes – not just with digital intelligence – but also through self-learning and self-organising workflows.
Of-course it is true that now prediction is becoming a commodity. It means that morecompanies can make betterdecisions, and fastertoo. But decisions do not necessarily translate into actions.
Some argue that automation of workflows is helping to achieve this. I hold a slightly different view. Workflow automation has been practiced for decades now – and even I agree with reasonably good results. But these workflows are not easy to setup. The design of a workflow is a slow, definitive process and requires careful planning and understanding of all possible scenarios and outcomes. This imposes a structure and boundary to the workflow and limits the flexibility. Changes to the workflow are slow to introduce and require both re-design and re-implementation followed by re-training. This often makes the workflows out-of-date in todays fast changing dynamics.
In the age of cheap prediction, new scenarios and outcomes are constantly discovered. Even if these predictions are fed into automated workflows the actions and outcomes are confined to a pre-determined set – the set that the designers anticipated as the range of scenarios and implementers programmed as rule based decision algorithms or multitude of if-then-else type statements. Conventional approach to workflow implementation and automation is simply too costly, low on productivity and offers limited flexibility to adapt to changing needs.
This is where I believe AI will have the biggest impact – the potential to take the predictions (or even simply data or insights) and change the workflow dynamically, the possibility to use the prediction to not just determine existing action paths but to learn new action paths and set them up automatically (by-passing the slow painful process of large IT projects with their 6-12 month cycles of consultancy, design, implementation, integration, staging and deployment).
AI has the potential to introduce this fantasy – the ability to dynamically create, manage and execute workflows – a paradigm shift into a constantly evolving and dynamic workplace. That is the vision that I am working towards and our early success makes me an optimist – our use of machine learning algorithms today allows us to correlate events with dynamic storyboards (thus setting up the full context), and with a simple application of prediction we are mapping intents to a multitude of actions. This is the first step away from the realm of data and insights into the world of real time actions and our machine learning driven framework is slowly evolving into a learning framework of self-organising and dynamic workflows and action paths.
AI will have a profound impact on the workplace – be it revenue or cost or customer experience – it will be more far reaching than the cheap prediction scenario envisions. The opportunity stands before us to turn business models upside down, to create new categories and overturn existing market structures. I can’t say if it will change everything, but it sure can change a lot of things. People and organisations that accept this and participate (beyond the hype paradigm), i.e. for whom AI is not simply a badge of innovation, will reign as future leaders.
And if you still do not believe me, take a moment to look at Amazon– AI is central to their workflow even today – some of this is highly visible (and an edge for the future), such as the autonomous Prime Air delivery drones; the Amazon Go convenience store that uses machine vision to eliminate checkout lines, but then there’s so much more behind the scenes – some to drive more traditional predictions for process optimisations like demand forecasting, product search ranking, recommendations (product, deals), detecting anomalies and exception handling (e.g. frauds, asset management) and others that are innovating business models and personalising customer service… and many more that we are still to hear of!
This article was first published on @LinkedIn October 23, 2018.