Artificial intelligence (AI) has changed and will continue to change the workplace. Recently, concerns have been raised that advances in the branch of AI called machine learning will put a variety of jobs at risk. Some of these concerns rest on thoughtful analysis while others are based on little more than the-sky-is-falling hysteria. Should you be worried that your job is threatened by machine learning AI?
Alarm bells should ring anytime someone talks about machine learning taking people’s jobs because machine learning doesn’t do jobs – it does tasks. Even the simplest and most basic jobs are made up of a variety of interrelated tasks. The question isn’t whether machine learning can do your job, it’s whether machine learning can do a significant number of the tasks that make up your job more effectively, efficiently or cheaply than you can. To answer this question, you need to understand your job in terms of its task demands and you need to have a good grasp on what machine learning can and can’t do well.
The first step in coming to grips with how your job security might be affected by machine learning is to analyze your job in terms of its task demands. When you’re thinking about task demands it’s a good idea to keep two things in mind. First, nobody knows more about your job and its task demands than you do; second, nobody has a greater stake in believing that your job is essential than you. You have to analyze the tasks that make up your job without overestimating their difficulty to stroke your ego or underestimating their importance to stoke your fears.
Break your job down into the tasks you do regularly. If part of your job involves going to meetings, preparing for a meeting is one of your tasks. Preparing to present the progress that has been made over the past week on Project X at tomorrow’s meeting is a specific instance of the preparing-for-a-meeting task. You want to focus on tasks at the general, not the specific, level.
Identify the task demands for each of your tasks. Task demands are the things that take up your time when you fulfill a task.
When you break a task down into its task demands you may find that some of those demands are actually subtasks. For example, preparing for a meeting usually involves gathering and organizing the information you are going to present at the meeting. Both of these subtasks take up some of your time. The task demands that take time are the ones you want to focus on.
Distinguishing between activities that provide information and use information is a useful way to think about task demands. Gathering information you are going to present at a meeting is a task that provides information. Examples include doing an internet search, running a spreadsheet analysis or asking a colleague for information. Organizing the information you are going to present at the meeting is a task that uses the information you gathered. Organizing information might include deciding what to present and what to leave out, or how to present your information to have the best effect.
The distinction between providing information and doing something with the information you provided is important when considering how your job-related tasks may be affected by machine learning.
Machine learning is the branch of AI that is causing much of the uproar over jobs. If you are concerned with how developments in machine learning may affect the status of your job, the key question you have to answer is which, if any, of the tasks that make up your job are the kind of tasks that machine learning does well.
Work-related tasks almost always involve making use of information in one way or another. The information you use in your tasks has to come from somewhere and one of the places it can come from is a machine learning program. Machine learning excels at providing information which means that, in theory, machine learning can make a contribution to almost any task you do.
Theory is one thing, practice is another. If you want to know how the tasks machine learning is good at map onto the tasks that make up your job, you have to have a basic understanding of machine learning. A general introduction to machine learning that lays out what machine learning is, what it does and how it does it can be found in “What Is Deep Learning And How Is It Useful?” It’s recommended that you read this article if you are interested in how developments in machine learning might impact the workforce in general or your job in particular.
One of the things covered in “What is deep learning?” is that machine learning programs need to be trained before they can perform job-related tasks. The nature of that training puts limits on how easy it is to create a machine learning program that can successfully perform a specific task.
Many machine learning programs need to be trained with structured data sets. (See “What is deep learning?” for an introduction to structured and unstructured data). Structured data can be hard to find, and it is expensive and time-consuming to produce. Deep learning programs excel at learning from unstructured data at the cost of increased training time and computational power. In either case, enormous data sets that often contain millions of individual instances are needed for training.
While a well-designed machine learning program can learn just about anything if the training data are available, once it has been trained it can only provide the kind of information it was trained on. Machine learning programs that are capable of producing actionable results in the workplace are not flexible.
For example, in the field of medical records, a deep learning program could successfully perform the task of identifying the diseases that were diagnosed in a collection of unstructured pathology reports. If you took the program trained on pathology reports and gave it the task of identifying the crimes that were reported in unstructured crime reports, it would fail. In order to provide a deep learning solution to the crime-reports task you would need to take a fresh deep learning program and train it on a massive data set of unstructured crime reports.
This lack of flexibility means that if a machine learning program is going take your place by providing the information that you provide in one of your tasks, it will have to be trained on a massive data set that contains the information of interest. If the training data aren’t available, a machine learning program cannot learn to do your task.
Digital information storage has produced unimaginably large quantities of data that could be used to train machine learning programs. If you want to determine whether one of your job-related tasks might be taken over by a machine learning program, you have to consider several questions. Are there immense data sets available that could be used to train a machine learning program to do your task? If those data sets exist, do they contain structured or unstructured data? Is it worth the time and money that would be needed to train a machine learning program to do one of your tasks?
Threat or benefit?
The doom-and-gloom hysteria over machine learning AI putting people out of work rarely gets down to the nitty gritty of explaining exactly how machine learning programs that are demonstrably successful at carrying out specific job-related tasks are going to do away with someone’s job. I suspect this is because the more extreme paranoia about machine learning turns out to be groundless when you recognize that machine learning does tasks, not jobs, that many tasks are involved in doing a job, and that machine learning programs are not very flexible.
This doesn’t mean that machine learning in general and deep learning in particular will not affect the workplace. It already has and it will continue to do so. But whether that effect is a threat or a benefit to you depends on how you react when you identify which, if any, of your job-related tasks could be completed successfully by a machine learning program.
Your tasks that demand spending time providing information are candidates for being taken over by machine learning. If you think a significant number of your tasks could be handed over to these programs, you might be afraid that you could lose your job.
There’s another way to look at it, however. If machine learning programs did your information-providing tasks for you, it would give you more time to spend on the tasks that involve doing something useful with the information you provide. Completed tasks produce results and a task’s value is strongly correlated with the usefulness of its results. Machine learning has the potential to give you more time to spend on your most valuable tasks.
Machine learning can help you do your job better if you focus your attention and effort on improving the skills that enhance performance on your tasks that involve doing something useful with the information you provide. Machine learning then becomes a benefit rather than a threat, and you may find yourself evangelizing for it rather than being afraid of it.
Update. An in-depth look at how machine learning affects people whose main job is providing information for others to use can be found in “Machine Learning May Look Like It’s A Threat To Information Providers, But Looks Can Be Deceiving“.
Kevin Murnane covers science & tech for Forbes. You can find more of his writing about these and other topics at The Info Monkey and Tuned In To Cycling. Follow on [email protected]