BAS and GPT: Toward Intelligent Buildings that Talk

Millions tried the conversational chat A.I. language model of GPT (Generative Pre-trained Transformer) on the OpenAI interface. It is a remarkable thing. However, testing it on Building automation and operation use cases is yet to be discovered, at least with the OpenAI GPT model once they release its API so that programmers and innovators in that space get into the applications' mud.

I thought about how GPT can be utilized for Intelligent Buildings. Referring to the Controls Brain concept I wrote about, GPT could be the engine running the show at the heart of the building brain, giving it a perceived cognitive ability, even though it is only a deep neural network ML model. However, there are many things to consider and solve, like what we should do with the current status of the Building Automation Systems. It takes work.

It is good to agree first on what defines an intelligent building beyond the marketing hype where some offer a sensor tagging the Building as "intelligent" by implanting that sensor in it (I am being sarcastic here, and it is needed sometimes to remind ourselves about unlocking the actual collective value beyond quick financial gains built on misleading Jargons).

The way how I see it, which is arguable, an intelligent building is a building that uses innovative cognitive technologies that allow:

• Automated and advanced controls mechanism
• Incorporation of occupant preferences
• The ability to learn from the collected data and have situational awareness
• Provides integrated capabilities to control and operate the building systems intelligently towards reducing energy consumption and operational costs

What I mean by cognitive technology is one that continuously learns from its interaction with data, people, and situations and thus eventually improves its learning and reasoning capabilities.

The majority would agree that Building Automation System (BAS) plays a vital role in supporting the functionality of intelligent buildings, especially for advanced control of the electrical and mechanical systems, the reach to field sensors and monitoring devices, and the ability to harvest data from the field and provide them to emerging technologies that contribute to the smartness of buildings.

BAS has undergone different development stages, from a standalone control system to a connected and integrated one leveraging A.I. and machine learning technologies to provide insights and analytics and shift from reactive to pre-emptive operation, leading to more efficient operation. While many are already working on A.I. solutions to optimize the building controls, with GPT, BAS would take an extra step towards a robust cognitive system, increasing the building intelligence degree.

Why is that?

With GPT integrated into BAS, we allow the Building to express itself, mimicking a human being, an operator of its own. We talk to the Building, understand its status and requirements, and discuss actions it took or corrections it needs to be done or addressed. The Building is virtually alive, fine-tuned, and trained with the collective knowledge of laws of physics, standards, regulations, operations and control intents, and owners' requirements. A building that interacts with its occupants and takes care of its systems' health status.

While it sounds cool, there are many underlying challenges, especially how to integrate it. Not to mention existing BAS challenges like openness and interoperability, unified or at least an agreed upon ontology, security, privacy, and dealing with big data, let alone A.I. standards for use in Buildings (well, who needs them, let us innovate, and they will follow).

The Integration Challenge

One of the main integrating challenges that puzzled my thoughts is incorporating a natural conversational language with actual commands, status pulling, and further situation analysis. Consider, for example, a scenario like that: I speak with the Building and ask what alarms we have now and what we can do to address them. The Building speaks back, listing critical ones and stating it has already informed the X team to handle physical rectification. Then you tell the Building to log them for future predictions and consider some actions to mitigate any risks (according to whatever standard – Okay, I do love standards).

The challenge here is how the Building would act based on a conversation. How can it analyze and convert a speech into a system-understandable set of commands? How to get the alarms from the BAS and conversationally present them?

For the luck of my thoughts, I came across an interesting A.I. research study (Schick et al., 2023 - arXiv:2302.04761) titled "Toolformer: Language Models Can Teach Themselves to Use Tools," yes to use tools, and it was published three days ago. It could be an initial potential approach to solve this challenge. An advanced specific version can use Building Automation as an external tool enabling real-time access to building data and external information by embedding the API calls as part of the natural language processing. If we think about it, it can also call GPT (depending on how the APIs are structured), which will be fine-tuned to Building Automation knowledge (i.e., standards, laws of physics), facilitating bi-directional command, feedback, and learning.

Just wow!

(We still need a way to structure and design API calls specific to BAS integration. Think about a middleware API server to facilitate the model interactions. Yeah, you might be sounding like... "Ugh, not another middleware. Can we blend them all together in an open solution?")

The Future

The potential of having a cognitive model that you can fine-tune to specific building design that would know the inside out of the building systems, components, and endpoints, and then you make a conversation with (an actual voice conversation) is a huge game changer for how we could interact with buildings. (It sounds like Poe of the Raven Hotel from Altered Carbon movie)

Being the first to bring to the market such a solution would force significant players within the "Intelligent Building" space to redefine their products to catch up – or potentially acquire you! Soon enough, evaluating BAS systems will consider a sub-criterion for A.I., which is natural language enabling.

Putting aside the limitations and biases in A.I. models (we will figure out get-around solutions for them), what matters now is that we live in the most exciting times of disrupt-everything with A.I. towards advancing our civilization and enhancing our lives. Let's go, innovators.

Thank you for being here and reading this piece.