- Machine learning (ML) algorithms are incredibly powerful, and companies like Google, Microsoft, Amazon, and Salesforce.com realize that – hence their intense interest in operationalizing ML and DL tooling.
- But, those algorithms alone are no guarantee of value. Whether you’re predicting the weather or optimizing a delivery route, AI lives or dies according to the humans within whose care it finds itself.
Can we truly know whether or not we’re living out our lives as a part of a simulated, holographic model of the universe as proposed by mega-entrepreneur Elon Musk? Should we even care about such things? If you’re at all concerned about the weather – about the expected path a hurricane will take, let’s say – then the answer is a resounding ‘yes.’ I would argue in fact that we are living out our lives based upon countless simulations.
When your phone showed you the best route to work; when your bank notified you of a potential fraudulent charge; when your streaming service recommended that you spend the next ten hours binge watching Game of Thrones: those were all driven by simulations using historic data to predict future outcomes. These are so pervasive and the underlying AI technologies like ML algorithms are so highly publicized that we as consumers take them for granted.
Enterprise IT has likewise built up an alarming familiarity with AI and an expectation of immediacy and simplicity surrounding predictive analytics, where any programmer can, with a few lines of code, create and deploy a working predictive model. With Microsoft IoT Hub and Azure Machine Learning, for example, a developer can quickly stand up an IoT app that can predict the weather based on collected temperature and humidity data – no muss, no fuss, and only a few select lines of SQL code.
Tools like Microsoft Azure Machine Learning are amazingly accessible, bringing AI to a much wider audience of enterprise consumers. But, in doing so, these products also paint a picture of AI as just another tool in the developer’s toolbox, something we can put into action quickly and with immediate effect. That is a very dangerous assumption. How do we know the supportive weather data is correct or appropriate to our needs? How do we know if our model will remain accurate over time? How do we even know if we’re correctly interpreting the predictive outcome within the context of the resulting application?
We should, therefore, view the current ML gold rush as a warning and a reminder that the application of AI in predicting business outcomes is an endless cycle where collaborators seek to answer such questions. More importantly, it reminds us that AI is a team sport, requiring the cooperation of many specialists — not a single, PhD-equipped data scientist, mind you, but instead a small, diverse team of pragmatic experts capable of reliably employing tools like ML algorithms.
A great case in point comes from IBM. While at the company’s Analytics University conference recently, I learned that at least one IBM product group was selectively deploying what it calls ‘Data Science Elite Teams.’ Separate from professional services and pre-sales, these squads are assembled to match customer requirements and typically include experts in data engineering, ML and deep learning (DL), optimization and prescriptive analytics, and communications (someone who can tell compelling, accurate stories).
To this mix, the vendor will work with its internal integration services group (IBM Global Business Services) to add a business/vertical market expert. But, it’s telling that IBM would choose to figuratively bundle people and software together in this way, mostly sidestepping an additional integration services engagement. Clearly IBM feels that that, in order to help its customers succeed in deploying complex predictive models using its analytics software, it (and its customers) must invest in the human side of AI.
ML algorithms are incredibly powerful, and companies like Google, Microsoft, Amazon, and Salesforce.com realize that – hence their intense interest in operationalizing ML and DL tooling. But, those algorithms alone are no guarantee of value. Whether you’re predicting the weather or optimizing a delivery route, AI lives or dies according to the humans within whose care it finds itself.