Last year we ran a little thought experiment with some of our technology leadership. We asked them whether they thought artificial intelligence was overrated or underrated in PropTech. Some of the more visionary leaders at our firm were staunch in their views — it was highly underrated and they couldn’t believe why it wasn’t used more in our industry.

I held the contrarian viewpoint. “It’s way overrated,” I said. I had been a part of several beta tests with a few PropTech companies trying to build AI models to a point that they could be useful. We had looked at a couple of use cases, mostly around working with unstructured and semi-structured data. None of them worked well enough for what we wanted them to do. Don’t get me wrong, they were all impressive, but progression in this field is asymptotic, and the technology just hadn’t progressed far enough to bring the accuracy levels to a point where they would be useful.

Until now.

ChatGPT dropped in late 2022 and since then other AIs seem to be launching on the daily. Most of them are better than any AI I’ve run across by several orders of magnitude. OpenAI (the makers of ChatGPT) released GPT-4, which is several times better than their previous language models.

What does that mean for us? I think it means that we’re at an inflection point. A place where we’ve someone has finally cracked the code on building a useful AI platform — one that ordinary developers can use without having to be experts in AI themselves. I expect 2023 to be the year where we start seeing real world usage of these models in PropTech and CRE. Here are 3 ways I think PropTech will benefit from it.

1. AI Will Enable Humans to Add More Value Than Ever Before

Okay, you got me, this is kind of vague. But stay with me for a second and hear me out. The trend I’m starting to see in the way these tools are implemented, are that we’re moving away from “magic black box” implementations where we’re supposed to trust the AI to get it all correct, and into more “Human In The Loop” type workflows.

These HITL workflows rely on people to work with the AI, instead of being replaced by the AI. The end result is that the human is able to spend more of their time on the complex, creative work where they add value, and less on the mundane stuff that’s below their skill level.

As an example, let’s say you have someone who has a data entry-type role. Their job is to flip through a set of contracts and pull out critical details from the documents. AI is well suited for finding well structured values, like a purchase price or an hourly service rate — something that doesn’t really change format a lot. It might also be good at locating information, like where a particular definition is. This leaves the humans able to spend more time making sure the information extracted by the AI makes sense, and using their expertise and reasoning abilities to identify and abstract out more complex data. 

2. Language Model Advancements Will Make Using Unstructured and Semi-Structured Data Much Easier

There’s a LOT of data locked up in unstructured or semi-structured formats in Commercial Real Estate. Most of the important information about your deal is found inside lease documents, loan agreements, rent rolls, etc. You’ve also got an assortment of Excel models scattered all over the place full of analyses and other important data that’s lost the moment you hit “save” and close out of the workbook.

Continued advancements in these types of general-purpose language models will allow the PropTech industry to leverage sources of data not available to them before. For the users of this data, the CRE Operators, it will mean that the companies that generate the most data, in turn can capture the most data, and give themselves deeper competitive advantages through better underwriting and better predictive capabilities.

3. The Foundations Will be Laid for Answering Real-World Questions with Natural Language

If you’ve used ChatGPT, you can already see where this is going. This language model from OpenAI can write about all sorts of subjects for you with very little input from prompts. It will write blog posts (not this one, this is 100% human generated), letters, emails, and just about anything you can think of. I even used ChatGPT and DALL-E to create a story book for my kids about an Irish fairytale my wife often tells our children.

Where this gets cool is when you start looking at the possibilities of using natural language to generate programmatic commands. So instead of using a traditional UI where you use calculator inputs to enter some numbers, you just ask the system to run a calculation for you. Some companies have tried doing this before, with limited success, but we’re finally getting to a point where this might be more mainstream.

I don’t expect this to be the norm by the end of 2023, but behind the scenes the foundations will start to be laid for this type of interaction. Scotty didn’t realize how good he had it in the future. He was definitely ahead of his time…