Perhaps one of the most exciting (and complex) technological advancements being worked on today is AI, short for artificial intelligence. But with so many marketers claiming their software is "AI-powered", "AI-enabled", or "AI-driven", to mention just a few, it's perhaps no surprise that we mere mortals are left wondering what AI is and whether it has a meaningful place in our businesses today?
This article looks to decode the buzzwords, explain the core concepts, and consider whether AI and machine learning have a significant role in SMEs today.
What is AI?
Contrary to the dictionary definition of "ai", and somewhat disappointingly, we're not here to look at whether a "three-toed sloth" could yield any benefits to your business. In technical terms, AI is "intelligence demonstrated by machines, instead of intelligence displayed by animals and humans" . However, with "intelligence" itself being a pretty broad brush, AI is often broken down into various sub-fields, each of which is focused on particular goals.
Artificial General Intelligence (AGI) doesn't currently exist.
Before we go any further, it's important to point out that "Artificial General Intelligence", i.e. the ability for machines to understand or learn any intellectual task that a human can, doesn't exist . Yes, it's a long-term ambition of those researching AI, but it's currently only hypothetical, despite being perhaps the best-known face of AI in popular science-fiction. Instead, we're here to discuss "weak" or "narrow" AI, being that which achieves a specific goal. Unfortunately, both of those terms somewhat diminish the strength of a marketing message, and it's little wonder we don't see the latest software described as "Weak-AI-Enabled" or, better, "Powered by Weak-AI".
Machine learning is a sub-discipline of AI.
One of the seven sub-disciplines of AI, and the one we're here to talk about today, is machine learning, which can be described as a computer algorithm's ability to automatically improve its accuracy and predict outcomes through experience.
Machine learning is already widely used today, with many of the world's biggest companies employing it to shape our experiences online. Notable and relatable examples include things like Google's ability to predict what you're searching for before you finish typing it. Far from being simple text prediction, Google's machine learning algorithm is being fed countless information about both your online and offline activity to enable it to present you with what it thinks are the most useful predictions. Similarly, Netflix's algorithm will collect countless pieces of information about your viewing behaviour to serve you with tailored recommendations of programmes you may wish to watch. These "recommendation engines" are just one example of machine learning; others include predictive threat detection, spam filtering and perhaps most applicable to SMEs, business process automation.
How is machine learning used in business process automation?
Business process automation (BPA), sometimes called robotic process automation, uses technology to execute recurring tasks in an organisation, reducing the need for manual effort . With BPA, you speed things up, reduce human input, and improve efficiency simultaneously.
A great example of a process perfect for BPA is employee onboarding, as there are numerous, reasonably straightforward, but ultimately essential tasks that need to be completed each time someone joins a company. You don't, for example, need Francis in HR to remember all of the things they should do when someone starts and robotically do it every time; a computer can do it for them.
Of all of the things our bespoke business systems do for companies, the most impactful is how they automate repetitive business processes. From holiday bookings and employee onboarding, to quotes and invoicing, the systems we've created all have elements of BPA designed to speed things up, minimise errors and ultimately improve profitability.
All systems designed to automate business processes require a level of learning, be that human or machine. Invariably the people building bespoke systems aren't the end-user, and therefore time is spent investigating, learning and documenting underlying business logic (the role of a solutions architect). However, certain learning tasks are perhaps better done by machines - an example being extracting information from documents. With machine learning, you could set up a learning model specifically to extract text from documents, and then by providing the computer examples, it could be trained to identify and extract text in a significant number of settings. Over time, and as the computer receives more samples (and human feedback), its accuracy would improve. While a human could easily extract text from documents and do so with higher accuracy than a computer (at least initially), the amount of time spent would be excessive and would never significantly decrease.
The downside of machine learning
As you can see from the example above, machine learning can have its uses, but it is highly prescriptive, meaning that you need to create a learning model for every process you want it to learn. Not only that, but there's still a requirement for a human to learn about the problem too, so at least in the short term, you're increasing the required resource and investment, not decreasing it. Finally, it would be fair to say that many of the businesses we encounter haven't even started to embrace business process automation, so jumping into the realms of AI and machine learning is somewhat premature.
"The best route for an operations manager or director to take the first steps into automation is to speak to a knowledgeable solutions architect, who'll be able to help identify which processes are ripe for automation."
The future of AI and business process automation
Looking into the future, and as AI improves and the tools to implement it become more widespread, we may well see a digital workforce emerging. These machines would utilise various types of AI, including computer vision, machine learning and more, to genuinely act "intelligently", but we're some way off that yet. The takeaway, though, is that to get to that point, a human needs to understand the lifecycle of your processes first. It's doubtful that AI will become a "point and shoot" solution, it'll need configuring and training, and the only way that'll be done is by a human understanding your business logic.