In the era of digital innovations, when the customer is king, organizations are leveraging AI to enable and empower their products and services with the goal of enhancing user experience (UX).
Simply defined, UX refers to people’s interactions with products or services and includes their emotions/attitudes about using them. AI technologies incorporated to enhance UX include, but are not limited to creating virtual assistant programs/chat bots to extend customer service after hours, increasing security for client online accounts, and enhancing website log in for improved user accessibility, which we will use as an example.
There are critical elements to enhancing UX that combine multiple disciplines, from engineering to graphical design, as well as cognitive modeling, which is the highlight of this blog.
What are cognitive models, and why are they critical?
Cognitive models are detailed accounts of human cognition (reasoning/thinking) interacting with user interfaces, i.e. logging into your social media website. Essentially, a cognitive model is a model that acts like a human by replicating cognitive processes that we as people go through to complete a task. They are critical for generating the cognitive, visual, and motor actions that produce behavior at the level of milliseconds. These models run on top of cognitive architectures, which are theories, implemented in software, of how the mind works as a whole (see Carnegie Mellon University).
Now, let us use the simple task of logging into your social media website as an example. The mere task of logging in will be our goal for the model. A cognitive model works with two kinds of knowledge: declarative and procedural. Declarative knowledge involves knowing definitively that something is the case -that J is the tenth letter of the alphabet, that Paris is the capital of France. In our example, declarative knowledge relates to your user name and password, which is required as part of the task of logging in. Procedural knowledge involves knowing how to do something, like ride a bike. In our example, procedural knowledge relates to how to interact with a login interface, which is also a required task. To enhance the UX of the login page of the website, we are going to connect a cognitive model to this interface and later we are going to extract the set of detailed actions required to interact with it.
Cognitive models are critical for generating the cognitive, visual, and motor actions that produce behavior at the level of milliseconds.
Remember that we are using Artificial Intelligence (AI) to replicate intelligent behavior, i.e. logging into your social media website. Therefore, the logical starting point is to perceive the elements of the user interface, i.e. the user identifies what they have to do to interact with the interface. Examples of this include moving your mouse to the login section and typing in your user name into the textbox. To do this, a semantic segmentation process is run on the user interface, which detects the nature and location of visual elements. These elements are later sent to the cognitive model via abridge developed in python. The cognitive model receives these elements and starts triggering a set of procedural rules that represents the knowledge required to interact with the interface. From this point a series of messages are transmitted to and from the model.
To understand this model’s level of detail, let us consider the action of introducing your user name. You need approximately 50 milliseconds to retrieve your user name from Long Term Memory (LTM), and 200 milliseconds to visually locate the appropriate textbox for input. Another 500 milliseconds are needed to move the mouse pointer and later click on the textbox. Depending on the length of your user name, it would take around one to two seconds. The great advantage of a cognitive model is that all this data is recorded, making it possible to detect opportunities for an enhanced UX. Now imagine that you find that your model takes an extra 500 milliseconds between capturing your user name and capturing your password. By looking at the output of the model, you determine that the manual actions are taking longer than expected. The next step is to formulate a hypothesis, such as making the password textbox more conspicuous to reduce the time needed, to complete the login task. You can make the changes to the interface and run the model to confirm or reject your thesis, but you also need to test it with actual people.
In order to move your change to production properly, it is advised to use A/B Testing or Split Testing, a method for comparing two versions of something to determine which is more successful. The process for comparison is to send 50% of your traffic to the new UX and keep 50% of the traffic for the current experience. The time needed to determine if your new UX of logging into your social media website outperforms the current one depends upon the number of visitors of your site. There are well-known formulas to determine the duration that best fits your needs. After testing is complete, you can evaluate the results. If the new experience reduces the time needed to log into your social media website, it is clear which login site provides the enhanced UX and 100% of your traffic should be accessing the new site.
Cognitive modeling is a fine-grained approach and sure way to achieving an enhanced user experience.