Why we need to understand machine behavior

Machine behavior refers to the actions, reactions, and responses of machines in various contexts. It can be as simple as the regular hum of the spin-drying cycle on a washing machine, or as dramatic as when the same machine randomly breaks and floods the whole house.  

Machine behavior is attracting attention because artificial intelligence (AI) determining how these machines interact with the world and with human users. It is apparent that their intelligent behavior may differ from their human counterparts.  

Beyond the washing machine example, there are various ways in which machines can exhibit behavior: machines can now behave reactively, deliberatively, and show social behavior. There are many diverse cases in which machines show these types of behavior: 

Reactive behavior 

This is the simplest type and happens when machines react to the current state of their environment. For example, smart thermostats like the one incorporated in Google Nest show reactive behavior by adjusting the temperature in response to changes in the environment. They also exhibit deliberative behavior by learning a user's schedule and preferences and adjusting the temperature accordingly.  

Deliberative behavior 

This involves machines using past experiences and knowledge to make decisions about future actions. Zipline, in Rwanda, offers a good example. It is a service that parachutes 150 single small medical boxes a day to remote areas by drones that can also show reactive behavior by adjusting their flight paths in real-time to avoid obstacles, maintain a desired altitude for delivering the box, and returning safely to base.  

More examples of deliberative behavior relate to self-driving cars that use sensors such as radars or cameras to navigate a route and react in real-time to traffic lights and the sometimes-unpredictable actions of pedestrians. 

Social behavior 

Machines can also show social behavior by communicating and interacting with humans and other machines. For example, virtual assistants like Siri, Alexa, and Google Assistant show social behavior by responding to voice commands and answering questions. They also exhibit deliberative behavior by using machine learning algorithms to learn a user's preferences and tailor their responses over time. Overall, every chatbot that uses natural language processing to answer questions and provide information is showing social behavior.  

The human-machine relationship 

These examples show how machine behavior has become an important aspect of AI and machine learning that decides how machines interact with the world and human users.

Machine behavior has created new human jobs, such as the ‘sustainers’ who analyze mistakes made by intelligent machines (such as the 11 people killed in autonomous cars accidents).  

The next step could be to learn how to take maximum advantage of the behavior of these intelligent machines. In their book Only Humans Need Apply, Thomas H. Davenport and Julia Kirby explain that HR departments that seek to optimize interaction between people and machines need to understand three points:  

• First, what tasks to assign to humans 
• Second, what tasks to assign to machines 
• Third, when to dismiss old machines or, in the case of humans, when to retrain them. 

This means that machine ‘behaviorists’ can maximize human interaction by observing humans when studying machine behavior. It may seem a vicious circle, but it is instead a virtuous one.  

In fact, in this dichotomy of human-machine, there are at least two ways to study machine behavior: by observing machines or by observing humans. Both provide insights into how machines should behave to interact effectively with people. To achieve this end, we can follow a three-step process: 

1. Setting the context 

As usual, experts recommend starting by identifying the context. Companies should choose a meaningful context where people interact with machines. Instrumental provides a good example in the approach it used to substantially cut costs by monitoring factory workers assembling iPhones.  

We could also set different contexts by observing a self-checkout machine at a supermarket, by interacting with a virtual assistant, or by using a delivery app. In any case, it’s important to choose a context where the machine's behavior is relevant to the user's task or goal.  

2. Observing user behavior 

Once the context is identified, it is time to observe user behavior, this is looking at how people interact with the machine, paying close attention to their actions, reactions, and responses, while taking note of any challenges or frustrations that users experience, as well as successful interactions.  

Once insights have been collected, they can be used to analyze user behavior to identify patterns and trends in how people interact with a machine. We can look for areas where users may be confused or frustrated, as well as areas where the machine inspires users by effectively meeting their needs.  

3. Finding improvement opportunities 

Finally, the most interesting part is finding opportunities for improving the machine’s behavior. For example, you may need a machine to provide clearer instructions, offer more personalized recommendations, or adjust its behavior based on user feedback. We also need to test and refine while adjusting the algorithms, interface, or other features to align them with user needs and preferences. This involves a continuous monitoring and improvements. 

Behavior as a tool 

It is now possible to improve an employee’s performance by observing their behavior with machines. This approach is not new and is borrowed from anthropology and ethology (the branch of biology that studies animal behavior) and it can help ensure that machines are designed to effectively meet the needs and preferences of their human users.  

Iyad Rahwan, an expert on machine behavior and managing director of the Max Planck Institute for Human Development in Berlin, wrote in Quanta Magazine:  
 
“Behavior doesn’t mean that it has agency [in the sense of free will]. We can study the behavior of single-celled organisms or ants. Behavior doesn’t necessarily imply that a thing is super intelligent. It just means that our object of study isn’t static — it’s the dynamics of how this thing operates in the world and the factors that determine these dynamics. So, does it have incentives? Does it get signals from the environment? Is the behavior something that is learned over time, or learned through some kind of copying mechanism?” 

To start observing machine behavior, we must identify a relevant context, observe user behavior and analyze the data, and then identify opportunities for improvement. These could be the steps to improve human life.