What is the difference between Black box solutions and Switch?

Striking a balance between data interpretability and accuracy of black box models and the Switch solution

With increasing reliance on artificial intelligence to support businesses, data science is key in turning data into actionable insights to optimize operations and impact risk management and decision-making. Artificial intelligence (AI) techniques and solutions often operate like ‘black boxes’, leaving organizations open to new AI risks.

To maximize the value in data science to influence business outcomes requires transparency and accountability. How do we determine the interpretability and accuracy stemming from Black box model and the Switch solution?

In this article, we evaluate two major approaches to how users interact with the software solution:

  • Black box solutions: We refer to solutions whose goal is to reduce energy consumption and/or improve tenant comfort through advanced AI/ML controls algorithms with minimal transparency into the actual functionality.
  • Switch solution: With solutions across multiple verticals and use cases, the data granted to users is visible and democratized—empowering users to actively make decisions and influence the software solution.

How Intelligent is Artificial Intelligence for End-Users?

Black box providers utilize AI to produce insights based on a data set. Machine learning programs develop deep learning algorithms and reach conclusions from the data inputted. However, there is a lack of clarity around its inner workings for the end-user and can be complex even for developers.

“Opacity is the heart of the Black box problem – a problem with significant practical, legal, and theoretical consequences. Practically, end-users are less likely to trust and cede control to machines whose workings they do not understand.”

Burrell, 2016; Ribeiro, Singh & Guestrin, 2016

Nonetheless, Black box machine learning projects tend to offer high degrees of accuracy despite generating minimal actionable insights and a lack of accountability in the data-driven decision-making process. This accuracy comes from the algorithms’ complexity, resulting in a lack of transparency. Trusting a Black box model means that you trust not only the model’s equations but also the entire database that it was built from.

In contrast, Switch grants visibility and insights into its behavior and the factors to influence user’s decision-making. This solution is more practical for companies or portfolios to understand and act on the predictions—enabling businesses to find tangible ways to improve workflows, clearly explain how they behave, produce forecasts, and the influencing variables.

While Black box users can only observe the input-output relationship, we assess through a comparative analysis how the Switch solution differs as a comprehensive data provider and to scope Black box limitations and opportunities.

Switch Automation

  • Switch openly displays data and analytics for a collaborative site approach
  • Switch gives a holistic overview of all the equipment in the building
  • Switch leaves existing controls in place and provides suggestions for improvement
  • Switch continually monitors sensors and alerts when data is out of calibration or offline
  • Developers can continually optimize and sustain performance given iterative nature of the process
  • Switch can ingest all data points, irrespective of equipment type or system i.e, energy efficiency or FDD
  • Switch cleans ingested data into good format that is normalized and harmonized for internal and external interpretability
  • Switch analytics have quicker deployment and testing timelines through ingesting historical data

Black box Solutions

  • Black box solutions are siloed and hidden with algorithms
  • Black box solutions typically only measure and monitor data points specific to their scope/outcome
  • Black box solutions overwrite building controls
  • Black box solutions are only as good as the data they are fed
  • Black box models can lead to technical debt to be frequently reassessed and retrained – ultimately driving OPEX costs
  • Black box solution providers may be exclusively focused a select subset of points i,e, limited connection only to the central plant to improve energy efficiency
  • The challenge for Black box solution providers is getting fed clean data for collaboration
  • Black box analytics require a big data set for machine learning systems

Understanding the Process 

In the built environment, enormous amounts of data from disparate sources and configured data points are generated from buildings—posing a severe problem for facility management teams attempting to analyze and optimize their operational efficiency with smart building technology.

Data ingestion is the first step towards digital transformation of smart buildings. For Black box providers, the speed of integration derives from only connecting to a small subset of points, i.e., the central plant. Switch is software and hardware agnostic and can ingest all forms of data and then harmonize and normalize them for consistency. The analytics are fully transparent to provide higher reliability and trust for the end-user.

Can Switch and Black box solutions work together?

As another layer to Switch, existing Black box solutions can integrate into our systems. While Switch has an appliance that runs Switch Dx3, it is transmitting data for transparency on what the ‘black box’ of the current controls and operations of the buildings are doing. There is an option to balance the two solutions to achieve greater interpretability and accuracy. 

Collaboration:  Between Black box’s advanced AI optimization and control generally with energy efficiency as the outcome, and Switch’s data-driven maintenance, Switch is an inclusive and collaborative platform democratizing access to data and enabling enterprises to build their own decisions. Our solution brings the right people, processes, data and technologies together transparently for strategic decision-making throughout the lifecycle.

Data preparation and analysis: Switch takes on an additional step of cleaning and normalizing the data. Switch offers easy-to-use data transformation and cleaning (to ensure data quality); data analysis (to identify outliers and key information); and many other features. This will help deliver a cleaned data set to Black Box solutions so they can learn and train their models faster and with greater accuracy.

Thinking outside the box with Switch Automation.

Switch Automation democratizes the data science process and reduces the skills barrier to enable broader participation—empowering companies to execute with optimal data interpretability and accuracy across portfolios and buildings.

Talk to a smart building expert to learn more about how Switch helps portfolio managers reach their sustainability goals.

New call-to-action