Aw shucks, first we have to agree on what “smart” means! It can mean clever, shrewd, correct, intelligent, glitzy, or something that smarts, or “stings.”
And it seems we must also consider what is our context. What world are we in?
I recently tried to do a simple, core analysis that we do in SGDS classes. This is a fundamental recognition that market analysis is what appraisers do – analyze a market.
For market analysis, trend, and time adjustment, our (econometric) method has simple rules. We need:
- The relevant “market” – the one the subject property competes in. And nothing wider. Not the city, region, nation, not the “neighborhood.” Only the CMS (Competitive Market Segment).
- Graphs to see and understand that market segment.
- “Enough” data, and rules to expand, if needed.
- To consider weird stuff, like outliers.
- A trend line formula/adjustor.
In short: 1) the right data; and 2) the indexing recipe.
AI comes to the rescue?
Using Claude AI, the response was both exhilarating and terrifying!
The good part. AI was terrific at the mechanics – dealing with extra fields, fieldnames, data types and conversions, formats, and simply following a good example of a proven price-indexing analysis method.
The program tried to be helpful. The results were definitive. and so sure. It was good, very good at following the dependable, established guide (as above) for time adjustments. Very, very good.
The scary part. The AI could not stop while ahead. It knew what to do: research the past – what others have done before. “Human knowledge came before artificial knowledge.” Simple. So how does AI gather its info? I am told this matters! The program is only as smart as its inputs . . .
AI sources information from: 1) web-scraped content; 2) licensed/proprietary data; 3) human-labeled training data sets; and even 4) consumer, market research results.
The bulk of the “knowledge” comes from history and volumes of writing and opinion – right or wrong, or just simply naive. AI can be gullible. It repeats. AI can make “mistakes.”
The risks of using AI are that results might be biased, misrepresentative, and just plain overwhelmed by the quantity of “amateur” opinions (as well as their quality). Innovation and real research is under-represented. Logic and critical analysis may be overwhelmed by rampant “groupthink.”
Remember the impact of groupthink: “It does not matter if it is true, so long as everyone goes along.”
AI information sources include historical, legacy, groupthink, and even wrong stuff.
Our individual reliance on AI analytics must include human judgment. Else AI joins the groupthink think.
Appraisal analytics and processes effectively stopped evolving in technological and judgment/reasoning some 20 to 30 years ago. (The result of the “Five Frictions” on Appraisal Progress). The results are an overwhelming amount of written regurgitation, which AI regurgitates. Darn.
Individual reliance and society’s reliance, and professional credibility — must lead, not follow.