Will predictive AI and generative AI like ChatGPT, Bard and LlaMA change jobs in leasing, investment, lending, research and property management? Yes: Artificial intelligence can do your reading for you and then answer your questions. It can draft reports, make recommendations, monitor your buildings. But it also needs humans to fact-check, filter and fill in the lay of the land.

“We’re in the early stages of all this, and so if you ask 30 people what ChatGPT is, you’ll probably get 20 or 30 different answers,” said RetailStat head of product innovation and marketing Gregg Katz. “Part of what we need to do here is cut through the hype and get to the truth.”

He does see potential for AI models and monitoring systems to change day-to-day responsibilities. He pointed to a scenario in which property managers rely even more heavily on digital tools. “The AI could be telling you right away what needs to happen on the property,” Katz said. “Maybe certain facilities or equipment need analysis or replacement, or it’s: ‘There’s a danger over here’ or ‘There’s a parking inefficiency over there.’ The AI then gets that information to those who can act on it.” A broker could use AI to generate a detailed report on the likeliest tenants for a vacant space, he said. However, AI doesn’t represent a huge threat to brokers in his mind. “Remember when Zillow was supposed to replace human brokers?” he asked. “Brokers know the market and lay of the land, things like ingress, egress and what’s going on with that local consumer. AI is not there today.”

CREModels managing director Mike Harris, a veteran coder with a focus on real estate tech, said AI is starting to make certain tasks easier in real estate, eliminating the tedium of reading through a 30-page retail lease, for example. “We created an AI-powered tool that allows you to plug in a lease and ask different questions not covered in the typical lease abstract,” he said. “The AI can have a real-time conversation with you about it.”

Landlords can ask AI tools like those created by CREModels for insights into data on revenue, net income or tenant lease expirations, Harris said, and lenders are using specialized underwriting and loan-sizing tools that tap into the power of large language models like ChatGPT, Bard and LLaMA. “Their loan-generation salespeople can ask it for the 10 largest loans the company did last quarter and then drill down for information on which rates or other loan metrics applied.”

To be sure, tech-savvy analysts already provide these types of insights by building spreadsheets and business intelligence dashboards, Harris noted, but AI makes data wrangling faster and easier for less technical colleagues. “With AI, you can interact with the data in plain language,” Harris said. One could upload, for example, local crime statistics, sales comps from a property appraiser and a spreadsheet of Census Bureau demographics into an AI tool. Then, one can quiz the AI for insights about these disparate datasets. “How many of these properties are strip malls that have sold more than once in the past five years? Can you give me the 10 most recent sales in each of the four highest-income areas? Oh, and let me know the crime stats on those, as well.” He explained: “The inquisitive, back-and-forth nature of it is interesting.”

But Harris has experienced a tendency among generalized LLMs like OpenAI’s GPT-4 to give answers that can be slightly off, flatly incorrect or even nonsensical. IBM describes these “hallucinations” occurring when a large language model “perceives patterns or objects that are nonexistent or imperceptible to human observers, creating outputs that are nonsensical or altogether inaccurate.” Until this problem is solved, Harris said, inspecting and questioning the output of such tools will play a bigger role in people’s jobs. “Right now, LLMs will give you an answer that can sound very convincing — ‘This is exactly how it’s going to work’ — but if you dig a little deeper, you might find that it’s inaccurate,” Harris explained. “So at least for now, you have to very carefully fact-check anything a general LLM produces.”

As he sees it, the industry stands to benefit the most from finely tuned AI models built specifically for real estate needs and populated with vetted and/or proprietary data. “That finely tuned model is going to be far less likely to invent believable-but-false scenarios,” he said. “You’ve put guardrails in place to make sure the workflow is functionally helpful.”

Katz predicted that specialized human expertise will be needed to make sure the inputs and outputs of AI models will be both relevant and accurate. “AI can give you actionable insights, but you need the right context and questions,” Katz said. “Individual experience is going to continue to be super important.”

There is incentive for C-suite executives and others to spend more time studying AI and tracking and discussing AI trends, he said. “It is going to have an impact. That much is clear. If CEOs and COOs want to create a vision and road map, they’re going to need to keep their finger on the pulse of AI and understand its power and possibilities.”

Ultimately, most in the industry will see AI as an integral part of doing their jobs, like a calculator or spreadsheet software, not as a fundamental threat to their livelihoods. “A person with useful tools can simply do more, and AI is no different,” Harris said. “Those who take advantage of it are going to outperform their unaided competitors.”