The Why of AI and ML

In recent years, Artificial Intelligence (AI) and Machine Learning (ML) have been in the spotlight. I think by now folks understand that neither are some strange form of technological magic, but rather a science and some working knowledge of the domain is now well understood. However, up to this point, the question has been “what are AI and ML?”  But what I want to take some time to ask, is why? Why do we need them? Why will they still be a part of products and services; not just in security but throughout our digital lives?

The short answer is because we can no longer operate at human-scale to be competitive. It is necessary that we operate to some degree at machine-scale and this is where advanced computer science techniques become valuable. AI and ML are just a few of these techniques and to maximize their benefits, we must know how to use them safely and effectively. Like any advanced technology, they can be used for good or for bad or maybe I should say that it could be used for your benefit or your demise.

There is a pattern that joins humans to machines, machines to machines, machines back to humans, and humans to humans. We have hundreds of years of social science that we can use when examining human to human patterns. Machine to machine includes a multitude of well-known patterns in computer science discussed on a daily basis. So for now, let’s concentrate on the patterns that integrate humans with machines and vice versa.

Human-to-machine communication is largely based on the human’s ability to communicate their “intent” to the machine. This is done via a model that the machine can process and that the human can express and understand. The precision of this model is critical to the overall success. A model that is too coarse limits the machine’s accuracy in its automation; while a model that is too precise may lead to humans making errors in their expression or just be too tedious to maintain. A great example of a model done well is cloud-native orchestration like Kubernetes. The admin can specify his/her intent for production in a model and Kubernetes orchestrates these microservices in an adaptive manner depending on future demand of the environment – scaling up and down depending on criteria.

One last thing to add about these models that sit in-between human-to-machine is that in the example above, the initial model may have been instantiated by the human, but over time, machines via their observations, can create their own models at scales well beyond human perception. You could say that these machine derived models are “machine-learned.”

The machine-to-human pattern is largely constrained by human cognition and human understanding. No matter how fancy your machine learning system may be, if the human cannot understand how the machine arrived at an answer, it cannot be trusted. Machines must “explain” their findings and analytical outcomes in a way that humans can understand. Failing to do so means that automation is not being safely managed and should not be coupled with actions that are critical to human life or to the business. To operate at machine-scale effectively and safely, machines must be able to communicate their operational integrity and analytical outcomes in ways that their human steward can comprehend. This is challenging because in some cases, machines are interfacing with experts and in some cases, non-experts. In the end, you must design systems that are observer-centric and accommodate for the different personas that use the system.

Getting this right means that we can leverage machines as tools that help us go beyond human perception and even what is humanly possible to build as a workforce. This would not have been such a useful capability if it were not for the Internet. Because of the Internet, businesses are asked to understand questions that are global in scale, that deal with petabytes of data, quantities of data processing that are just no longer at human-scale. Businesses are also having to operate with dynamic ranges never experienced in our recorded history: On Monday you may have to service 30,000 customers, on Tuesday THE ENTIRE INTERNET SHOWS UP, and on Wednesday 20,000 customers. Without the help of machines, we could not take advantage of these opportunities.

The term Machine Learning has been used synonymously with Artificial Intelligence when in reality, ML is a child of AI. So, if AI is the parent of ML, does ML have any brothers and sisters? The answer is yes and over the years, we will move beyond the data science-biased ML as we meet and get to know these new siblings that will help us humans operate safely and securely at machine-scale.