AI Agents and Appraiser Oversupply

[George Hatch]

Yeah, the majority of it is reduced demand for mortgages. I don't see any trend of retirements in the 2017 to 2021 span that weren't offset by replacements/new entrants. The reduction happened in 2022 as volumes fell off a cliff. If we go through another 2-year licensing cycle with low mortgage demand, plus "new forms," I don't see any recovery in these numbers. And I don't see a compelling need for these appraiser numbers to recover.

Imagine going full steam ahead with PAREA under these conditions. They should really cut their losses ASAP.

I think adding a PAREA component to the QE would be of benefit but I oppose its use to satisfy 100% of the experience requirements. IMO PAREA shouldn't be counted for more than perhaps 40%. Just enough to practice some of the mechanics.

 

[J Grant]

100% agree.

I've always poo-pooed the doom and gloomers on this site. Not that they were wrong about the industry not needing appraisers anymore, but that the danger was imminent. The past two years watching AI evolved has changed my mind and the past few weeks only confirmed it.

I don't think the industry will be gone tomorrow, but it has been declining for the past couple years. That decline will rapidly increase as AI changes the entire lending pipeline and risk evaluation process.



Those books were outdated by the time they were published. That's how this is different. The speed of this tech is not something we're not going to be able to adapt to

events are actually a bit behind the books, which were written several years ago.

Intersing, that all the ebooks and developers of AI, including Elon Musk, warn that nations and businesses and people must develop ethical guidelines and regulations for its use and prepare for major losses of jobs, such as re-educating those whose jobs disappear. This seems like a naive solution since if jobs are disappearing across industries, including blue-collar jobs such as truck driving or retail service, how in the world are all those people supposed to retrain, and for what?

If AI is only as good as , and no better than, an appraiser, why replace a human appraiser with AI ? That is more a philosophical question than a business one, but business is not just $ in and out; it involves moral decisions that affect lives. And creates backlash - the shooting of the health care exec is the tip of an iceberg. Do you think millions of people will sit quietly around while their jobs disappear and they lose their house, and their kids face a bleak future ?

When AI replaces jobs, it reduces the customer base for those same products or services it is making them more efficient. Seems they have not solved that end of the equation yet

 

[J Grant]

I think adding a PAREA component to the QE would be of benefit but I oppose its use to satisfy 100% of the experience requirements. IMO PAREA shouldn't be counted for more than perhaps 40%. Just enough to practice some of the mechanics.

Who in the world wants to take PAREA to enter a dying profession? The numbers, I understand, are a trickle. Came years too late.

 

[J Grant]

just out of curiosity - what books are you reading on the topic? What drove your decision to read books on AI instead of deciding to learn some of the tools? Or are you doing both?

I will list them later - so far only two

What tools for AI do you suggest I learn?

I am reading the books out of general interest and perhaps to apply with buying some AI stocks or developing a use for a nonappraisal venture. Mostly out of curiosity, though- after a few books, it becomes repetitive.

 

[gregb]

https://futurism.com/the-byte/AI-investor-goal-crash-human-wages

 

[Vermonter]

If AI is only as good as , and no better than, an appraiser, why replace a human appraiser with AI ?

Because it's cheaper and faster. By A LOT

I'm curious as to why you think AI cannot be better than a human appraiser?

 

[alebrewer]

What tools for AI do you suggest I learn?

Take George Dell's class on data analysis - it's a bit disorganized, but the class centers around an introduction to R and R Studio. I'm in the middle of the class now, and it's started me down the 'R' rabbit hole. I've used Excel and MiniTab in the past, but those are fairly static offerings. R Studio has a literally unending amount of already coded 'applets' that are basically plug and play once you've learned the environment. WRT literal 'AI' applications, I've used Claude, Copilot and ChatGPT. What's really cool is that I can tell ChatGPT to write a script in R to do whatever. For instance: "Write a script in R that plots (X,Y) observations and adds both first and third degree polynomial trendlines"

Here is the script:

# Generate example data (x and y observations)
set.seed(42) # Set seed for reproducibility
x <- seq(1, 10, by = 0.5)
y <- 2 * x + 3 + rnorm(length(x), 0, 2) # Linear relation with some noise

# Fit first-degree (linear) and third-degree polynomials
linear_model <- lm(y ~ x)cubic_model <- lm(y ~ poly(x, 3))

# Create the plot of the (x, y) observations
plot(x, y, main = "Scatter Plot with Trend Lines", xlab = "X", ylab = "Y", pch = 19, col = "blue")

# Add the first-degree (linear) trend lineabline(linear_model, col = "red", lwd = 2)

# Add the third-degree polynomial trend line (predict values over a smooth range)
x_smooth <- seq(min(x), max(x), length.out = 100)
y_smooth_cubic <- predict(cubic_model, newdata = data.frame(x = x_smooth))

lines(x_smooth, y_smooth_cubic, col = "green", lwd = 2)

# Add a legendlegend("topleft", legend = c("Data Points", "Linear Trend Line", "Cubic Trend Line"),
col = c("blue", "red", "green"), pch = c(19, NA, NA), lwd = c(1, 2, 2))

 

[J Grant]

Because it's cheaper and faster. By A LOT

I'm curious as to why you think AI cannot be better than a human appraiser?

Because it does not simultaneously synthesize information and has no moral guideline or end goal. The books I am reading say they do not know if AI will ever be able to do that - and if it can it is a decade or more out -

What do you mean by Better? And If AI turns out to be "better" at most things than humans are, what will happen to all the humans whose jobs and income disappear?

What will happen to all the companies taking advantage of AI effects when fewer people are implied and able to buy or rent the services or products?

Most info is that AI is generative and will be in the background like electricity is now, rather than replacing each and every job and function. Though it will replace a number of jobs. The strength of AI seems more to be in medicine and science - some will use it as a shortcut to creativity, and I run into copyright problems with AI plagiarizes that exist.

 

[J Grant]

Take George Dell's class on data analysis - it's a bit disorganized, but the class centers around an introduction to R and R Studio. I'm in the middle of the class now, and it's started me down the 'R' rabbit hole. I've used Excel and MiniTab in the past, but those are fairly static offerings. R Studio has a literally unending amount of already coded 'applets' that are basically plug and play once you've learned the environment. WRT literal 'AI' applications, I've used Claude, Copilot and ChatGPT. What's really cool is that I can tell ChatGPT to write a script in R to do whatever. For instance: "Write a script in R that plots (X,Y) observations and adds both first and third degree polynomial trendlines"

Here is the script:

# Generate example data (x and y observations)
set.seed(42) # Set seed for reproducibility
x <- seq(1, 10, by = 0.5)
y <- 2 * x + 3 + rnorm(length(x), 0, 2) # Linear relation with some noise

# Fit first-degree (linear) and third-degree polynomials
linear_model <- lm(y ~ x)cubic_model <- lm(y ~ poly(x, 3))

# Create the plot of the (x, y) observations
plot(x, y, main = "Scatter Plot with Trend Lines", xlab = "X", ylab = "Y", pch = 19, col = "blue")

# Add the first-degree (linear) trend lineabline(linear_model, col = "red", lwd = 2)

# Add the third-degree polynomial trend line (predict values over a smooth range)
x_smooth <- seq(min(x), max(x), length.out = 100)
y_smooth_cubic <- predict(cubic_model, newdata = data.frame(x = x_smooth))

lines(x_smooth, y_smooth_cubic, col = "green", lwd = 2)

# Add a legendlegend("topleft", legend = c("Data Points", "Linear Trend Line", "Cubic Trend Line"),
col = c("blue", "red", "green"), pch = c(19, NA, NA), lwd = c(1, 2, 2))

Thanks, but I do not see a benefit to myself to learn it, I will never catch up to those who are talented and have an interest in a deeper dive AI-generated data analysis. I am reading the books for an overview, a perspective on investing and how to possibly apply it in a creative project - it is already used in data analysis by the GSEs when they review our appraisals and in teh sttaiical tools we use.

 

[Surf Cat]

Because it's cheaper and faster. By A LOT

I'm curious as to why you think AI cannot be better than a human appraiser?

Plus..... a lot less friction. The lender doesn't have to hear about scheduling delays, waiting for revisions, deal with "subject to" reports which include 1004d's, listen to the borrower complain that their place is worth "at least" as much as the sale across the street.....

As to your second part, AI can't differentiate quality of construction, slipshod DIY remodeling, end of useful life Roofing, deferred maintenance, Etc....yet. This is where the human appraiser shines.

As JGrant points out, when AI starts taking away all this employment for humans, what are humans going to do? What purpose will they have? How will they make money and circulate it in the economy? I can easily see pissed off humans revolting and destroying the servers running AI. Just like that whack job that took out that health insurance CEO.