Thanks to the latest advances in machine learning, artificial intelligence (AI) is currently rocking the markets like the most revolutionary technology of the Fourth Industrial Revolution. Everyone in the business sector is talking about it like it's going to change our world forever, and in many ways, it already has. Recent studies show that 67 percent of business executives look at AI as a useful means to automate processes and increase efficiency. But it is seen by general consumers as well as a potent instrument to increase social equity, with over 40 percent of them believing AI will expand access to most fundamental services (medical, legal, transportation) to those with low income.
However, the speed at which this incredible transformation of the automation processes could be even higher, and there are a few issues that are currently bogging it down. Which ones are the most important roadblocks that are stalling the adoption of machine learning? (Deep learning is another form of AI that companies are beginning to adopt. For more, see These Pain Points Are Preventing Companies from Adopting Deep Learning.)
Lack of Organization
A company, especially a larger one, is a complex creature. Just like a mythological hydra, it has many heads that often need to make the same decision, such as the chief information officer (CIO), the chief digital officer (CDO) and obviously the chief executive officer (CEO). All these officers run their own departments, which are supposed to drive their AI efforts together, at the same time and with the same effort level. Suffice to say, in real life, this rarely occurs.
Clarifying who "owns" the machine learning project and is thus responsible for spearheading its implementation within the company is the first step. In organizations where several well-established data and analytics teams need to sync their operations, it is not infrequent that many of them just end up diluting their work on a myriad of smaller projects. Smaller pilot projects may contribute to the overall understanding of the machine learning science, but will often fail at achieving the automation efficiency needed by the core business.
IT service management (ITSM) might represent one effective solution to address this issue by helping the various IT teams understand, for example, those sectors of the business that generate significant revenues where automation can improve margins or reduce the percentage of errors.
Machine learning is an old yet new technology. Rudimentary AI dates back to the early '80s, but the recent development of modern deep learning algorithms helped this tech take a quantum leap forward. True specialists working in this field that possess a sufficiently in-depth knowledge are, indeed, very scarce, especially since Google and Facebook scoop up 80 percent of machine learning engineers with Ph.D.s.
Many organizations know their limits and no more than 20 percent think their own IT experts possess the skills necessary to tackle AI. The demand for machine learning skills is growing quickly, but those with the expertise and talent needed are true rock stars nowadays. However, many of those who possess sufficient training with deep learning algorithms may lack a formal qualification such as a masters degree to show that. Remember: This field is still new – many who are pioneering it today are old-time programmers from an era where Ph.D.s in machine learning simply did not exist.
Many human resources professionals must now struggle with the difficulties of hiring a proper candidate for jobs whose complexity may be beyond their own expertise. Today, even telling the difference between the competencies of a machine learning engineer, a data scientist and a front-end developer is a complex feat for the non-natives. Eventually, though, AI-powered recruitment will probably pose as its own solution to assist all HR managers.
Inaccessible Data and Privacy Protection
Before they can learn about anything with their cutting-edge machine learning algorithms, AIs need to be fed with data. Lots and lots of data. However, most of the time this data is not ready for consumption, especially when it comes in an unstructured form. Data aggregation processes are complex and time-consuming, especially when the data is stored separately or with a different processing system. All these steps need the full attention of a specifically dedicated team composed of a different kind of experts. (For more on data structure, see How Structured Is Your Data? Examining Structured, Unstructured and Semi-Structured Data.)
Data extraction is also often unusable whenever it contains vast amounts of sensitive or personal information. Although obfuscation or encryption of this information eventually makes it usable, additional time and resources must be devoted to these burdensome operations. To solve the problem upstream, sensitive data that needs to be anonymized must be stored separately as soon as it is collected.
Trust and Believability
Flexibility is not a trait that all humans possess. And when a deep learning algorithm cannot be explained in a simple way to a person who is not a programmer or an engineer, those who may wish to bet on AI to harness new business opportunities may start dwindling. This is especially true in some of the more traditional brick-and-mortar industries. Most of the time, in fact, historical data is practically nonexistent and the algorithm needs to be tested against real data to prove its efficiency. It is easy to understand how in some industries such as oil & gas drilling, a less-than-optimal result may lead to substantial (and unwanted) risks.
Many companies that still lag behind in terms of digital transformation might need to revolutionize their whole infrastructure to adopt AI in a meaningful way. Results might require a long time before they're visible, as data needs to be collected, consumed and digested before the experiment bears fruit. Launching a large-scale machine learning project with no guarantee that it is worth the investment requires a certain degree of flexibility, resources and bravery that many enterprises simply might lack.
In a curious turn of events, many of the roadblocks that still slow or stall the advancement of AI are linked to human nature and behaviors rather than to the limits of the technology itself.
There are no definite answers for those who still doubt the potential of machine learning. This is a path that has never been trodden, and field experimentation is still needed during this development phase. Once again, it is our turn to leverage one of the characteristics that helped humanity achieve its most extraordinary heights: our ability to adapt. Only this time we need to teach this skill to our intelligent machines.