The hype around AI these days is ubiquitous. Name an industry and you can find plenty of material on how AI is going to revolutionize it’s landscape. In some respects the hype is warranted, there are areas in which AI and machine learning (ML) truly are changing the foundations of business. But on other dimensions, the hype is just that: overblown promises fueled by outsized expectations and the hope of a panacea.
In this article, I’m going to break down a few thoughts on areas where AI can be a winning strategy, some stumbling blocks, and some places where the technology (and our techniques) are not quite mature enough yet.
The Good
Despite the hype, AI has made some promising advances both in terms of technical capability and in our shared learning on how to apply it to business. Here are just a couple areas where AI and ML shine today:
Platforms Are Making Core AI Capabilities Accessible
Gone are the days where any foray into AI required an incredible amount of technical capacity: computational power as well as highly-specialized software engineers and statisticians. Today, AI-as-a-Service offerings from platforms like AWS, IBM Watson, Microsoft Azure, and others are making serious AI capabilities accessible. These services make take a lot of the heavy lifting on for you and allow smart engineers to tackle complex problems without specific training.
Prediction and Classification in Narrow Domains
One question I get a lot is: what use cases are ideal for AI and ML today? The first answer I always give is: wherever your employees spend time classifying or predicting outcomes in very specific ways. For example, when you have manual (possibly paper driven) processes which require sorting and categorization at scale. Or, when needing to predict the amount of a particular fast-moving commodity product to keep in stock. Or perhaps you have a lot of sensors that monitor equipment and want to enable early failure detection. These scenarios will likely have substantial amounts of discrete training data available and the algorithms to perform such operations are relatively mature a this point.
Strides in Alternative Interfaces
Almost all of us walk around with a powerful speech recognition platform in our pockets. Massive advancements in speech-to-text and voice interfaces continues to change the way we navigate our lives and work. The AI breakthroughs that power these experiences are also massively successful in image/facial recognition, gesture detection, and other alternative interfaces. These technologies are ready today but require significant thought in terms of UX to make them work for customers in a natural and intuitive way.
The Bad
While there is a lot of good to focus on with AI these days there are also some things to watch out for when beginning to explore the possibilities.
Hype Leads to Failing Early Moves
Almost every buzzword-worthy technology these days comes with a huge amount of excitement which is amplified by social media, and quite honestly, vendors like Tandem. But our approach has always been one of pragmatism: avoid the shiny object while focusing on the objective. We have seen several clients, who after making heavy investments in AI, started from scratch or significantly pivoted their projects due to overly ambitious goals. Keeping your goals extremely narrow and clear will help you stay grounded in your expectations and will allow you to take incremental steps rather than expend the energy necessary for a giant leap.
Data Maturity Holds Us Back
Most of the success to be found in AI today is driven by large quantities of data. But, this data must generally be clean and contain the necessary signals that would lead to valuable insights. The lack of “AI-ready” enterprise data in most organizations means that investments in AI will at best only be achieving mediocre returns. Cleaning up raw data often involves extended manual (human) time which is costly. Prior to any major AI initiative, organizations should first engage in a data-readiness program to ensure they have adequate, accurate, and non-siloed training data.
Regulations And Privacy Concerns Abound
As of the writing of this article the tech world is swirling with leaks and hacks which have lead to a hyper-focus on privacy and security. It shouldn’t take these kinds of extraordinary events for us to consider these key social and business issues. However, the waters become even murkier when thinking about how user data and personal information might be used in AI and ML scenarios. The opaque nature of many of our algorithms means that even seasoned engineers may not know exactly how an AI is utilizing the data that it is fed. It is critical to be very measured in how you use personal information with AI, especially so in regulated fields such as healthcare, education, and banking. This particular issue will likely never recede.
The Not Quite Yet
Artificial Intelligence and Machine Learning are still in their infancy despite recent headway. There are most definitely some areas where continuing maturity will drive the technology and techniques forward over the next few years.
AI Lacks Sufficient Definition in the Boardroom
Like many truly transformational technologies (the assembly line, flight, telephony, TV, the Internet) AI must have executive champions that truly understand its capabilities and nuances. Without this kind of understanding and support AI will remain a misdirected buzzword. We encourage our clients’ executive teams to spend some time studying and understanding AI and ML before they make investments. This will give them the confidence to give their support to winning strategies instead of those driven by hype.
Data Bias Is an Continuing Problem
Most of today’s AI and ML algorithms are relatively ineffective at detecting widespread bias in data sets. And, most humans are too if we’re honest. This means that until data becomes extremely wide and encompassing or our algorithms become more adept at weeding out bias we will always run the risk of having biased outcomes. A classic example of a biased data set was a public works AI that was trained by smartphone accelerometer data to detect potholes in the late 2000’s. At the time most smartphone users were on the upper end of the socio-economic scale. Therefore, potholes in wealthy neighborhoods received disproportionate attention.
People Are Still Integral to the Equation
Lastly, we are limited today by a world where generalized AI is still rather distant. We have made huge strides in developing domain-focused algorithms but not those that could be applied to a rather generic problem set. What this means is that smart humans (designers, software engineers, data scientists) are still the main driving force behind successful AI projects. These talented individuals now have a powerful new tool to force multiply their efforts, but they must still be in the driver’s seat.
—-
There is no doubt that the future of technology in general is wrapped up in the future of artificial intelligence. But there is a tremendous amount of nuance in using this revolutionary tool to benefit our companies, customers, and society. Taking a careful and practical approach will help you safely discover the ways AI can be the most impactful for your business.
