Whether you’re an appraiser, analyst, reviewer, or assessor, you’ve likely considered the impact of the blazing-fast expansion of AI on the valuation industry. Did you notice when Open AI first released their ChatGPT-3 large language model and think that this could be an interesting evolution of the technology we use? And as the hype built throughout the summer after the release of ChatGPT-4, did you start to think in the back of your mind “is this AI chatbot thing going to replace me as a valuation or assessment professional?”
It’s time to address the elephant in the room – are appraisers and assessors going to be replaced by cheap, fast, trained AI bots? My short answer is “no”. Real people contribute critical and irreplaceable skills to the industry such as personal interactions with industry participants. We possess macro and micro market knowledge along with nuanced judgment that is important to complicated valuations. And we understand how to apply valuation methodology in ways that must be taught to AI before it can be deployed effectively. But AI also has exciting skills that we can begin to use to enhance our jobs and our profession.
AI, and specifically large language models (LLMs) like ChatGPT, are incredibly powerful at consuming vast amounts of data. Consider ChatGPT to be the most knowledgeable appraiser that’s ever existed – it has been trained using a vast dataset of valuation publications and articles, classes, case studies, and even appraisal reports (but note training sets contain public, not private data). Beyond that, it can consider and analyze every piece of relevant data within its training set, like parcel records and maps, flood zones, easements, sale records, leases, rent surveys, etc., again, given restrictions that the data is public, and contained in the training dataset. Couple the vast training dataset with the fact that ChatGPT can be a brilliant writer, as after all, it has ‘read’ nearly all great authors in history and can mimic any writing style you desire, and ChatGPT sounds like the perfect appraiser hire. If ChatGPT showed up as an appraiser resume on your desk tomorrow, who wouldn’t hire this valuation professional that can consume data in the blink of an eye, write perfectly, and not run up your coffee expenses?
But here’s where the human touch comes in, and where I believe that human touch will be necessary for a good long time. All that data that ChatGPT can read and analyze? It’s not all good data. And all those public records and transactional events it can analyze? They’re not all relevant. ChatGPT and other AI frameworks, just like any appraiser just getting into the biz, need training. The AI needs to be taught when and how to apply specific data, and it needs to recognize when the various valuation approaches are relevant. Beyond that, AI models generally have a knowledge cut-off (ChatGPT-4 is currently knowledgeable through April 2023, while ChatGPT-3 cuts off somewhere in late 2021), making the application of up-to-date data unreliable.
I know many of you have experimented with AI by asking it to gather comps for an assignment, or even to write an appraisal. So far I’ve seen mixed results to those attempts. Comp searches within Open AI systems have generally returned only a small percentage of accurate, usable comps, likely due to the knowledge cut off and lack of access to private or proprietary data. I asked ChatGPT-3 to appraise my house earlier this summer, but because the AI had not been trained or prompted to specifically analyze my home using recent data and established methods, the result was irrelevant and unreliable. On the flip side, Zillow is powered by machine learning, which is a subset of AI, and Zillow’s accuracy reportedly improves with each model update. Zillow currently claims an error rate of only 2.4% for on-market homes, and 7.49% for off-market homes. And LLMs can be excellent writing partners given the proper prompts when you need assistance writing up a specific section of a report.
Obviously, given the tremendous pace with which AI development is progressing and the financial implications at stake, development of AI models specifically geared toward real estate and valuation will continue to accelerate. We can all benefit from development of AI tools – there are many use cases in which augmentation of existing workflows with AI widgets could greatly streamline our work. Think about pulling basic property research without manually visiting multiple different web pages, or exposing your draft report to an editor that checks spelling, grammar and consistency across text, images, and charts in the blink of an eye. With planning, patience, and appropriate guardrails, we can certainly welcome AI into the valuation profession as an extension of our skills and tools to enjoy greater productivity and lasting careers.
But if you’re still skeptical, don’t take my word for it – head on over to chat.openai.com and ask ChatGPT what it has to say about AI and the appraisal profession.