Machine Learning for Everyone: Responsibly Embracing AI
Business entities looking to be competitive must embrace artificial intelligence (AI). After recently attending Tableau’s Data Day Out and IBM’s Disruptive Innovation Forum, in addition to discussing all things data science with fellow data scientists, it is evident the hype is real. This especially is the case regarding machine learning (ML), an application of AI.
With vast amounts of data and a desire for forward-looking insights, ML is seemingly everywhere. At IBM’s event, I attended a friend's talk which focused on data lakes, featuring a machine learning application for a floor tiling company located in the southwestern United States. IBM also hosted several displays of applications with Watson, the IBM AI system. Watson was demonstrated in mining, healthcare, baking, and banking. It appears that everyone wants to get their hands on ML, and that they are now able to do so.
Tableau’s event featured speakers from a variety of groups including Walmart, Deloitte, and the University of Toronto, as well as product demos showing future releases of the software. These included functionalities with Alteryx, a software enabling users to deploy “drag and drop” ML, in addition to natural language processing (NLP), where the software creates a story for the data.
While the accessibility and availability of ML/AI to an audience beyond data scientists/engineers is a nice concept, it is also potentially dangerous. I saw a headline the other day along the lines of “Managers Need to Embrace AI”. Fair enough, but we need to define “embrace”. It is not enough to give a blind go-ahead to ML tactics. It is also not enough to simply use interfaces which eliminate true understanding of models being deployed.
In order to embrace AI, there should be a required minimum level of understanding. Someone responsible for overseeing the implementation of ML or another AI application should have a high-level understanding of the input data, requirements, output data, and models being trained on data. They should know if a decision tree versus a neural network versus ridge regression is being used, be able to provide a high-level explanation of the model, and justify why that model is being used. A person responsible for designing and implementing the model should know everything about the model to the most granular level of detail, regardless of whether they are coding or using a “drag and drop” method.
As AI is utilized in decisions which impact real people, it is imperative that we educate ourselves on the topic. It is not something to be feared, but something that needs to be understood. Embracing AI is necessary; doing so responsibly is vital.