In last week’s edition of the dBrief, Paul Bonington wrote about the “accelerated disruption” that is forecast for 2018. There can be little doubt the technologies that sparked and shaped digital transformations over the past few years are intensifying and reaching further into all areas of the enterprise. The signs are clear—organizations that do not respond to or act on these disruptions will struggle to survive.
Reading Paul’s piece reminded me of all the coverage and 2018 predictions for artificial intelligence, and its close cousin, machine learning. AI is so pervasive, a recent Forbes piece covered “51 Artificial Intelligence (AI) Predictions for 2018.” One of the more interesting contradictions surrounding AI is how broadly (and invisibly) it has penetrated our daily lives and yet how much further it needs to evolve to make a significant impact in a majority of organizations.
At one level this contradiction can be assigned to the classic arc of technology diffusion. The trajectory runs from the point of innovation to early adopters and through early- and late-majority stages, until finally reaching its plateau of adoption. Another way of looking at the relatively small number of business-led AI implementations is to understand the nature of AI’s dependency on data. This is especially true of the “training data” necessary for machine-learning algorithms.
Labeling data and humans in the loop
Machine learning requires a lot of data to achieve a high confidence level in the conclusions it draws and predictions it makes. The volume of data alone is not a sufficient indicator of success; optimal outcomes are achieved when the data is labeled. This is not surprising when you think about the multiple sources of data and content that most businesses manage and the importance of categorizing and describing this information in order to make effective use of it. Sound like metadata? Well, it is.
There are automated methods for tagging and labeling data. In more complex applications, like annotating categories of content, an alternative to automating everything is to use the “human in the loop” design pattern. In these cases, data is analyzed by an ensemble of algorithms. Whenever there is disagreement or low confidence, the information is presented to a human who makes the decision. The same thing happens every day in customer service departments: Most of the operation is automated and a human is called on to decide what to do only when there is ambiguity or unclear context.
Gaining business alignment
Apart from the data challenges in applying AI and machine learning there are also significant organizational challenges. It takes a long time—typically two to three years—to integrate new technologies and develop them for an individual use case even when you have a positive business case for deploying them. The hard part is changing the way an organization operates: changing the culture, getting everyone onboard and aligned with the right way for the firm to use the technology and getting all the individuals adapted to using the new technology. The people stuff.
Changing the culture can be especially difficult when it comes to introducing leading-edge technologies like AI, machine learning, conversational interfaces and blockchain technology. It’s one thing to replace a system of record (CRM, web); it’s something entirely different to deploy new systems of intelligence.
As challenging as these technologies may be to adopt and deploy, both technically and organizationally, smart business leaders will recognize the power of these innovations to transform their business. The journey may be disruptive to business models and there will be failures along the way. However, as long as the vision is true and clear and the strategy is properly executed these businesses will be among the winners.