Skip to main content

The dark side of automation: Machine learning

We are starting to build machines that can learn and adapt to their environments. While this offers many possibilities that are exciting and beneficial, it also has some severe drawbacks. A look at both, and a warning about where this learning can go, are featured in part 2 of this series.

Part 1 of this post examined a bit of the back-story of displacement of labor by automation, highlighting how automation has interacted with human roles throughout history.

The current challenge arises from advances in artificial intelligence (AI), and more specifically, advances in statistically adaptive data-processing methods collectively called machine learning. This set of evolving techniques for interpreting data and generating useful responses has been steadily improving, benefiting from advances in computer speed as well as insights from cognitive psychology and neuroscience.

Our brains serve as control systems for our bodies, and for the environment through our behavior. Our neural hardware allows us to manipulate physical media with great dexterity, make judgments about causal factors in our environments, and exploit those causes to arrange the world in ways that we prefer. We can incorporate vast amounts of new evidence and improve our performance in real time. Advances in machine learning are filling out this space of functionality as it becomes clearer to science how such things work.

An important thing to understand about this progress is that there is no 'sound barrier' for artificial intelligence based on human performance. Once computers can learn, artificial intelligence will surge ahead on its own schedule, regardless of how well people perform in any role.

Another important thing to understand is that human performance has never been uniform. Some people are far better than others at various cognitive tasks. Standardized tests, like IQ, SAT, and ACT, are designed to highlight the differences among humans, which are staggering. As our environments have become more cognitively enriched, successive generations are also performing better on such tests. There is no single 'human level' of performance with which to compare machine performance. Where livelihood and purchasing power are at stake, the coming machinery is a power tool that can cut both ways, just like any other power tool.

Part 3 of this post will seek to understand where these power tools may be taking us, stay tuned.

The author's affiliation with The MITRE Corporation is provided for identification purposes only, and is not intended to convey or imply MITRE's concurrence with, or support for, the positions, opinions, or viewpoints expressed by the author.