Sci-fi fans will recall Clarke’s third law: “Any sufficiently advanced technology is indistinguishable from magic.” HubSpot Research recently ran a global online survey querying consumers about artificial intelligence and determined that 63% of respondents didn’t realize they’re already using AI technologies. While you can argue that’s not magic, it is an interesting trick.
Those numbers may sound shocking, but they shouldn’t. When AI is correctly deployed and integrated into business, it ought to be inconspicuous. The whole point of advancing automation, enhancing analytics, and embracing machine learning is to ease human interaction with technology — to make our tools more effective and less obtrusive. As such, good AI is invisible.
But that doesn’t mean its impact is undetectable. Leaving aside trending arguments about bots displacing the human workforce, running amok, or prospective brain-machine interfaces, it’s time for a reality check. Practical real-world AI currently performs colossal amounts of computational labor — humdrum stuff that doesn’t make headlines, but is most definitely fueling a surge in innovation. It’s here and it’s working.
AI today includes a variety of tools and technologies — ranging from trusted decision management systems to virtual agents and deep learning platforms. I see Fortune 500 companies with sophisticated AI rules-based systems driving automated operations across billions of events per day. I also see new and novel machine learning applications detecting subtle patterns across complex engineering and customer data.
Speaking from a data scientist perspective, the advances in the machine learning ecosystem are exciting, and business applications are coming out of the woodwork in all industry sectors. Supervised algorithms that can identify patterns are good tools for problems where the conditions of interest are well-represented. As outlined by Andrew Ng, this includes applications in language translation, customer loyalty (e.g., targeted online ads, customer churn), loan repayment, equipment maintenance, and fraud. Unsupervised algorithms work well for anomaly detection when patterns are not well known or characterized (e.g., in the analysis of sensor data from industrial equipment). Combinations of machine learning with rules-based systems and process-control thresholds are providing new insights into business operations. These methods are already being deployed across applications in energy, utilities, manufacturing, logistics, and retail.
Popular explanations of current AI capabilities often reference image recognition (à la Google), autonomous vehicle technology (à la Tesla), or automated personal assistants (à la Alexa) — and those are great examples of AI’s power deployed for search, real-time response, and natural language processing. But while speech recognition, natural language processing, and natural language generation are a big part of the current hype on AI, the truth is that AI is increasingly utilized for more “everyday” purposes in industry — dedicated to supporting closed-loop systems of insight for business advancement. In these areas, running the business (operation) produces raw data, which is used for analyzing the business and producing business intelligence data, which in turn is used for action that improves the business (cycle and repeat). The current art lies in the zone between decision support and automated action.
Machine learning is also being used to make software smarter and more suggestive, thus accelerating both efficacy and TTM. In the enterprise, AI engines can be deployed to consume data from disparate sources, communicate with each other, apply business rules and automate activity, and even suggest to the software user how to better utilize the tool that choreographs the whole process.
And applications that are not necessarily intelligent themselves can be embedded with AI to deliver more meaningful information and enable intense personalization within an existing workflow. In a marketing department, for example, when sales associates log into their work application, they can automatically see their prospects ranked in order of likelihood to close a deal. In an engineering environment, a safety inspector can automatically receive color-coded visualizations of equipment issues continually triaged by level of urgency.
While the growing sophistication of such applied sciences shouldn’t be downplayed, I don’t mean to suggest that AI “just works” wherever and whenever an IT department might like it to. This is not plug-and-play technology and enterprise AI is still constrained by the systems in which it must reside. There is a reason why totally digital and data-dependent industries are overrepresented in AI use cases at present: Their infrastructures were already streamlined for its arrival. To work effectively, any AI solution must be highly customized to the individual organization and calibrated to that context. You can have a broad suite of AI-powered tools, but before you can do anything useful with them, your whole system must be tightly integrated and your data sources prepared to “feed the beast.” That takes work.
These interconnections of systems and data is of crucial importance, and is often both a starting point and a parallel track in any organization’s smarter digital journey. Thankfully, this work is getting simpler – with smart products combining messaging, integration and microservices for real-time data preparation and transformation in private, hybrid and cloud environments. For companies making the AI leap, these technologies speed the transformation process, with design-pattern-based approaches that enable construction of AI-ready data systems that can cut the total time to operational efficiency to weeks or months as opposed to years.
This increasing speed to interconnection and data that are ready for AI is one reason to expect more and more “everyday” AI on the horizon — as the bar to entry lowers, greater numbers and types of organizations are embracing the technology. We are also seeing business problem cross-pollination that is enabling the spread of AI-driven solutions. Consider anomaly detection, the automated identification of items that lie outside an expected pattern of items in a dataset. Industrially, anomaly detection can be used in a million different ways, from predictive equipment maintenance to network surveillance. Some of the very same techniques that are used for finding anomalies in equipment performance can be applied to, say, financial crime and fraud detection in the banking and insurance industries. Or consider natural language processing (NLP) and natural language generation (NLG): The same techniques that allow Alexa to tell you whether it’s going to rain before you head out to a picnic can be used to more efficiently fulfill documentation and compliance requirements in heavily regulated sectors. Accenture reports that virtual assistants can review a thousand legal documents in a matter of days — a task that would take three people six months to complete. We are also seeing the emergence of NLG to standardize communication of insights from analytics, enabling consistent assessments on the business, and avoiding alt-facts and needless spin.
While these capabilities may impact human paralegals, it’s not all bad news. Alan Turning’s famous summation in Computing Machinery and Intelligence that “we can only see a short distance ahead, but we can see plenty there that needs to be done” remains true to this day. Accenture also studied more than 1,000 large companies already using machine-learning systems and “identified the emergence of entire categories of new, uniquely human jobs.”
The AI genie may not be magic, but it is indisputably out amongst us and changing the nature and pace of business. And it’s not going back in the bottle. Back in 1959, John McCarthy, who coined the term “artificial intelligence,” proposed that the “ultimate objective is to make programs that learn from their experience as effectively as humans do.” In modern enterprise, this is already in action. Real AI is here, right now.