The Appraisal Blog

Appraisals and AI Part II
September 19th, 2023 4:18 PM
As the article states, AI (Artificial Intelligence) is all the buzz.  From creating writing reports to passing the bar exam, from realistic photos and deepfake videos.  It can write music and solve math problems ... is there anything it can't do?

I recently gave ChatGPT a shot. 

I started by asking it if it could help analyze the value of a specific piece of real estate:
A: (summary) It gave a very generic response on look at comps, consider location, evaluate condition, look at unique features... and of course ... consult a real estate professional.
So I told it I was already an appraiser and wanted more specifics:
A: Consider market conditions, zoning and land use environmental factors, historic significance, income potential, etc.
I wasn't getting what I wanted to I asked it if it could help me do my job better:
A: stay up to date on industry, develop skills, network, use tech, be objective, etc etc
I asked if I provided an address could it give me market data:
A: It said it did not have any real time data and I should consult with an agent, or on-line news sources,etc
So I decided to switch gears ... maybe it could describe a house to me.  So I sent it a picture of a house I recently appraised ...


It provided a mostly inaccurate response:
A: Based on the picture you shared, the house appears to have a modern design with clean lines and a minimalist aesthetic. The exterior of the house is predominantly white, with some darker accents on the roof and around the windows. The roof is flat, and there appears to be a small balcony or terrace on the upper level.

The house is two stories tall, with large windows on both levels that allow for plenty of natural light. The windows are rectangular and appear to be floor-to-ceiling, with black frames that contrast with the white exterior.

There is a small front yard with a neatly trimmed lawn, and a driveway leading up to the garage, which is located on the left side of the house. The garage door appears to be made of wood and has a simple, modern design that complements the rest of the house.

Overall, the house has a sleek and modern look that emphasizes simplicity and functionality.

I asked it to estimate the size:
A: Based on the picture, it's difficult to estimate the exact size of the house, as there is no point of reference to compare it to. However, I can make a rough estimate based on the size of the garage door and the number of windows.

Assuming that the garage door is a standard size, which is typically around 7 to 8 feet high and 8 to 9 feet wide, and assuming that the windows are roughly standard size as well, which is typically around 3 feet by 5 feet, I would estimate that the house is somewhere between 1,500 and 2,500 square feet.

Again, this is a very rough estimate based on limited information, and the actual size of the house may be different.
It is a ~10,000 sf. house
I got a similar result with a picture of a kitchen.
I went back to asking about the market ... this time for residential property in a specific zip code ...
A: As an AI language model, my knowledge is based on previously published data and my training data cutoff is September 2021. However, I can provide some general information about the residential real estate market in the 92128 zip code based on available data.
I didn't verify the median numbers it produced... that data was worthless to me occurring over 18 months ago.
I asked it if it could analyze sales data that I provided.  And sent is a csv of the Spark export I use for the 1004MC.  I even told it which was the Sales column for it to look at.  It came back with a number of $1.526M.  I told it was wrong and its numbers didn't make any sense based on the range of values ...
A: I apologize for the error in my previous response. Upon re-evaluating the data, it appears that I made a mistake in my calculations.

Based on the data in the CSV file you provided, the average closing price for the 170 sales recorded is approximately $884,329. The average price in 2021 was approximately $1,020,714, while the average price in January 2022 was approximately $811,111. It's worth noting that these averages are based on the data provided in the file, and may not be representative of the entire market in the neighborhood.

At least it admitted it was wrong!!
So ... based on my first impression.. this stuff is not going to affect the appraisal industry any time soon.  It can help solve math problems, summarize a piece of literature ... even help with coding I'm told.  But it can't describe pictures very well and has no current market data.
Anyone else play around with this?  What were your results?




Posted in:Technology and tagged: #AppraisalModernization
Posted by Robert Abbott on September 19th, 2023 4:18 PMPost a Comment

Subscribe to this blog

Recent Posts:

Archives:

Categories:

Appraisal (12)
Appraisal, Lending (3)
Architecture (1)
Bias (10)
Development (15)
Education (3)
FNMA (2)
FSBO (1)
Home Buying (4)
Houses (6)
Investing (1)
Laws & Regs (2)
Lending (5)
Market (6)
Mortgage (1)
Technology (3)
Transit (2)
VA (4)
Value (1)

My Favorite Blogs:

Sites That Link to This Blog: