Post Pandemic, Major Assessment Jurisdictions should consider Hiring AI Engineers (rather than Traditional Modelers)

"AI engineers don't write code to build scalable data pipelines like a data engineer...instead, they understand how to extract data efficiently from a variety of sources, build and test their own machine learning models, and deploy those models using either embedded code or API calls to create AI-infused applications."

Conversely, the existing mass appraisal (CAMA) regression models are not intelligent enough as they are highly modeler dependent (subjective). Thus, given the same sales dataset, five modelers may come up with five different models with very different results. Of course, the biggest failure is the Sales GIS -- generally developed off the current market attributes -- so it's representativeness relative to the population (which is more or less static) is difficult to establish and justify.

[FYI -- That is why, in my AVM books, I propose the use of fixed neighborhoods as they are not sales-dependent and are population-derived, rather than the Sales GIS, which is totally sales-dependent and is not necessarily representative of the population the model is applied to.]

AI engineers do not use any data-variable modeling. Their data extraction process is ingenious, leading to brilliant machine learning models. In a mass appraisal environment, they will precisely identify and demonstrate where the sales datasets and populations are at variance. Whereas, the mass appraisal modelers will remove them from the model as outliers, creating unexplainable gaps and significant fault lines they won't even know.

Alternatively, in a traditional CAMA environment, it's all sample-based, so the results are, at best, bell-curved with the customary 68% efficiency. The error-based CAMA regression models fail to test the solution; for example, is a model Coefficient of Dispersion (COD) of 8 better than a COD of 10? The COD of 8 could represent a post-optimal solution, whereas the COD of 10 could correctly represent an optimal solution. But in a CAMA environment, the COD of 8 would be universally preferred (In fact, several years ago, I presented a paper along these lines at a national conference, raising some serious questions). 

The mass appraisal industry is too antiquated, using the 30-year old regression modeling and mostly Sales GIS. That is why it is high time that the major mass appraisal jurisdictions start hiring some AI engineers, proving that the industry needs to look ahead. 

Granted, given the paucity of AI engineers, it will not be easy to hire AI engineers, but the agencies should widen the search and try. Given how the union-heavy civil service system works, they should also remember what Steve Jobs said, "It does not make sense to hire smart people and then tell them what to do. We hire smart people to tell us what to do." In other words, they must be given the necessary flexibility and autonomy to develop forward-looking solutions, without being bogged down to backward-bending maintenance. 

Of course, to make the modeling environment more efficient and solution-oriented, these agencies should also hire STEM graduates instead of traditional business and humanities graduates who lack advanced quantitative training and knowledge and make very poor modelers. Since a sizable percentage of municipal hires are non-civil servants, these folks could easily qualify in that segment, citing an urgent need for high-level quantitative talent – just the way the major US companies hire skilled foreign nationals under the annual H1-B visa quotas. 

The combination of STEM and AI could be the nirvana for these major jurisdictions.  

Similarly, in the futuristic consumer environment (e.g., free home valuation apps and online sites), the AI-based solutions would, optionally, ask the first-time users to take a short tutorial. As the user interacts with the tutorial, the machine learning models will extract and store the data (by reading the user's responses) and fine-tune the model for each user. When the user returns to value a subject, the stored model will populate the comps as soon as the issue is defined so that the ten-minute exercise would be reduced to fifteen seconds – and customized.

-Sid Som
homequant@gmail.com

https://www.datarobot.com/wiki/ai-engineer/