Increasing Employee Efficiency by Automating 93% of Benefit and Expense Reconciliation Processes
Automated reconciliation reduces manual processes for the operations team by 93% and eliminates human entry errors.
How to Find Opportunities for Machine Learning
Back to insights
It seems like everyone has something to say about machine learning or artificial intelligence nowadays — whether it’s the fear of an imminent robotic overlord or the redundancy of all human labor. Fortunately, we don’t have to worry about either of these for quite a while. In the meantime, machine learning has not come close to being able to simulate the general intelligence of humans, but it still provides a useful tool that can be harnessed by businesses in all different industries to automate tedious and mundane cognitive tasks. This frees up human labor to work on tasks requiring more problem solving and creativity.
Current State of Machine Learning
Recent artificial intelligence research can be broken down into two main categories; general intelligence and narrow intelligence. When Elon Musk warns about the dangers of artificial intelligence, he is referring to general intelligence, which is a system capable of learning any task that a human is able to perform. We are not close to achieving this goal. Instead, most research and development in corporate America is focused on narrow artificial intelligence. Narrow AI is predominantly concerned with training a model to achieve high levels of success on one specific task.
Where Narrow AI Excels
This “Narrow AI” is exposed to mass amounts of historical data (labeled or unlabeled) in order to detect patterns that it can harness to make decisions. This process of detecting patterns is called machine learning. For example, a machine learning algorithm can be fed years of historical data on loans. The computer may detect that whenever the customer’s credit score is below 500, they are significantly more likely to default than a similar customer with a credit score over 500. So, the computer can learn that customers under this threshold should not be given a loan. This is just the tip of the iceberg for such algorithms. Most modern machine learning systems contain hundreds if not thousands of similar “checks,” and the best part is that these don’t have to be explicitly programmed by a software developer. The machine learns these rules on its own.
These capabilities do not stop at numerical data. Documents are also a medium through which machine learning shines. Many companies spend countless hours poring through documents or images to extract important information or to make some decision. Computers are now capable of scanning documents and images, extracting important information, and making decisions about the contents all on their own.
What kind of tasks can computers learn?
We can break down tasks into two dimensions: the predictability of a task and whether it involves physical manipulation of the environment or not. Manual tasks, or those tasks involving manipulating the physical environment, still present considerable challenges for the technology of today. Oftentimes, the software side of the equation is ready, but the robotic technology doesn’t exist quite yet or is prohibitively expensive.
Non-routine tasks, on the other hand, require a machine to make decisions and perform actions that it has never seen before. This is outside the reach of the “Narrow AI” of today and will require the general intelligence of the future. Today’s machine learning thrives in the “Cognitive Routine” area of this visualization.
Routine Cognitive Tasks
In today’s economy, there is a surprisingly large amount of labor that falls under this category. Most tasks that are repetitive, monotonous, and digital are ripe for automation. Fortunately, this coincides with some of the pain points of employees. A good rule of thumb when looking for opportunities to harness machine learning is to ask your employees what the most boring part of their job is. Reviewing 1000s of applications to assess the likelihood of success for potential students at your institution is a tedious, boring task that also happens to be a great task for machine learning. Reading 100s of documents to organize them based on subject sounds like a terrible job, but once again machine learning can automate the entire process. So, talk to your employees, find out what part of their job they feel is mundane requiring no creativity, and let’s see if we can get a machine to perform that task allowing your employees to focus on where their skills are better utilized.