Design Thinking in AI

March 11th, 0202

AI Organization

An AI organization intends to employ machine learning and data science tools and techniques to automate and/or enhance its manual or rule based businesses processes. This would help that organization to improve efficiency, reduce cost and scale the business. These days most organizations are working to transform themselves into AI organizations, mainly due to recent advances in machine learning methods, cheaper compute and storage and digitization of businesses processes. The entire ecosystem has arrived at a point where this transformation is not a choice but has become a necessity to stay both relevant and competitive in years to come.

AI Transformation

AI transformation requires an organization to work on many fronts such as infrastructure development, upgrade existing resources, adding new hires, migrate and uplift legacy systems, develop new machine learning models and AI applications, defining security and privacy protocols, reference architecture etc. Unfortunately, most of the organizations are either not ready for this or doing it improperly.

Design Thinking

It is very important for an organization to adapt design thinking during the process of AI transformation. Design thinking is about taking enough time to understand the organization's existing state, future business trends and potential areas of AI application. Design thinking is opposite of jumping directly to the solution without considering future use cases, integration options, talent and deployment infrastructure.

For example, imagine you are a telecom company and going through AI transformation and there is an immediate need for anomaly detection solution. For now the requirement is to detect any atypical patterns in company's website traffic. Jumping to the solution approach would be hire a data scientist who may have worked on anomaly detection and let him develop method given historically data of website traffic.
You left the following questions unanswered before starting the method development;

  • Did you think about deployment and existing production environment? Would it be deployed on premises or in the cloud?
  • Did you think about the development tools? Python, R, SAS etc.
  • Did you envisioned any other anomaly detection use case in your organization? Video stream events, cellphone call streams, Wi-Fi service events, etc.
  • Have you checked for any off the shelve method? There may be solutions out there and you only need to bring your data to it. You may not need to reinvent the wheel over next few months.

These are very important strategic questions to be answered before solution development starts. If we don't do that now, there may be cost to be paid in future in form of new solutions for similar problems but for different data sources and production environment, integration problems, solution transferability and extension.

There is a very good book on Software Design by John Ousterhout. Though the book's focus is software development and programming but it equally applies to data science and machine learning solutions development in AI organization. In the book, John has rightly talked about strategic and tactical programming. In context of an AI organization, we can thinking about both terms as strategic AI thinking and tactical AI thinking.

Strategic AI Thinking

It doesn't just focus on a working solution, which may bring unnecessary complexities in order to get the job done quickly, but also comes with a great end to end solution design. In nutshell, the primary goal is to come up with an effective solution which works for the organization in long term. This thinking would require an organization to invest 15-20% extra time in planning, brainstorming, coordination, and research.

Tactical AI Thinking

Tactical thinking is short-sighted and focused on solving the problem as quickly as possible without considering long term implications as this thinking doesn't plan for the future. To get the task done, it would introduces a lots of unnecessary complexities and dependencies, which would become roadblock for future upgrades, extensions of solution, knowledge transferability tec.

Author: Nasir Mahmood, PhD