How ChatGPT Can Aid the HERS Industry
ChatGPT, a tool developed by OpenAI, is a powerful artificial intelligence software tool that can generate believable, well-written text responses to any prompt imaginable in almost any language. It is generally awe-inspiring, unsettling, and inspirational all at the same time.
I enjoyed its intuitive responses and dexterity of knowledge since it was released but have developed skepticism about its reliability and practicality for the HERS Rating market. Having explored and experimented with ChatGPT I want to share my experience, and thoughts on how it could be used in the HERS Rater workflow, and how it may disrupt some aspects of our business. I’d also like to hear how you’ve experimented with it or the ideas you’ve developed.
What is ChatGPT?
At a very basic level, which is all I am able to offer, ChatGPT is a machine-learning model that has read vast amounts of text from the internet. Each time it reads text it records a little bit of information about the context of the words and the relationship between words. Using this data it created a machine-learning text response chatbot that can respond to questions in a conversational manner.
It does this by observing that this text includes words like “what”, “HERS”, and “Rating”. Then, based on past observations, it develops a response that matches patterns it has previously observed in cases that have the word or words “what”, “HERS”, and “Rating.” It is mimicking other patterns to answer this question. It understands what a response to a question should look like and spits out that response. However, it does not understand the correctness of the answer because it could have read an incorrect description and did not know that was the case.
By doing that it can do some amazing things like respond to emails, summarize notes, write code, translate, review and comment on code, and write whimsical rational about why to get a HERS Rating. In addition to the chatbot that I’ve been discussing in this blog, OpenAI, and other AI tools can be used to create fine-tuned tools that utilize Artificial Intelligence and Machine Learning to provide better contextually accurate responses.
How can this fit into the HERS Rater or Field Inspector tool belt?
Help build integrations between your software tools.
I’ve had success providing ChatGPT with documentation for software APIs and asking it to write a script to connect these tools. For example, I have an inspection app that I use to demonstrate Inspection Sync and our API. The inspection app needs to have a form ‘dispatched’ to an inspector with data in the form from the Ekotrope energy model. I do this dispatching from a Google Sheet calendar. I pasted the Ekotrope and the inspection app API documentation into ChatGPT and asked it to write me a google script to dispatch a form to my inspector. With some finessing and minimal debugging, it wrote me a functional integration between these tools! One of Ekotrope’s engineers could have done it just as quickly but ChatGPT supercharged my coding ability.
Answer building science questions.
ChatGPT could provide guidance on how to fix installation issues, diagnose odd building behavior, or suggest ways to meet energy code if the duct leakage test is worse than expected for inexperienced inspectors. Often home inspectors run into new situations and need support to understand how to fix a problem. This usually requires contacting a more experienced energy modeler or inspector. ChatGPT could provide those answers instantaneously without distracting other co-workers. I put this to the test and asked ChatGPT some building science questions. The ChatGPT chatbot does not do very well. It understands the concept but does not get the details right. However, if a Machine Learning model were trained on better data this could work very well. You can see that the Google response does that and finds a trusted source.
Answer scheduling emails?
Confirming your inspection schedule each day requires a lot of back-and-forth emails/texts/phone calls with each site’s superintendent. ChatGPT is relatively good at composing emails and contextualizing responses. This could be a big opportunity to leverage ChatGPT. However, you would likely need to use the fine-tuning that ChatGPT offers to build in guardrails to avoid any unexpected results. I think a similar outcome could be achieved using simple scheduling tools similar to how you may have experienced with dentists, barbers, and hair stylists.
Inspire people to solve problems.
In my opinion, the best thing ChatGPT has accomplished so far is inspiring people to take the large and small frustrating, annoying, and brain damage-inducing problems they face and create solutions. It has such endless use cases and a friendly approachable UI that it reduces the barrier to innovation and encourages us all to immediately start innovating. Here are just a few ideas that we’ve had:
Can it quickly tell users if a design change is likely to improve the performance of a home and guide them toward better home designs when using Scenario Modeling?
Could it be used to answer User questions about how to model certain homes, mechanical systems, etc more quickly?
Can it enhance QA Track’s automatic QA checks to improve the accuracy of ratings?
In summary, ChatGPT is a powerful tool that can create credible text responses that are not always accurate. Especially in the complex, knowledge-based industry of energy-efficient home construction. However, ChatGPT is causing us (and our partners) to dream and create new innovative solutions to age-old problems. I love that outcome! And I can’t wait to implement new tools to improve the industry (whether ML based or not) and hear the ideas that ChatGPT has fueled.