Media | Articles
Can AI Appraise Your Car? Maybe, but a Human Touch Helps
Oh, great. Another AI article.
We have to talk about it, though, because it’s everywhere. Take a look in almost any corporate stock prospectus, investor outreach, or PR blurb. It’s a corporate must-have, and at this point it seems like everyone is touting their AI chops in ads and on their socials. I’m not sure when my local oil change shop will jump on the AI bandwagon, but I’ve convinced myself that at some point they will. It’s all over the news, too, and not always for flattering reasons—lawyers submitting briefs with made up cases, a newspaper that posted books for summer reading that included a few that do not exist, a government report that cited AI-fabricated studies. And don’t fret, there will be more screw-ups, lots more. But this is a car website, so how does any of this affect our hobby?
Let’s get one thing straight—I’m not an AI hater who’s stubbornly against new technology, nor am I a total AI fanboy. I see promising possibilities as well as some of the downsides, and it is likely going to affect my field of automotive valuation and appraisal.
Like many of you, I don’t really need to see an AI-generated image of how an old Stanley steam car would look if they built it in 2025, or any of the other “art” you can generate in under five minutes (four minutes, technically, for the Stanley). But on the positive side, there’s potential for AI-assisted medical breakthroughs that will save lives. And like almost any major innovation, AI will likely be one of those game-changers that will, like it or not, make life better for some and worse for others. It will both take and create jobs, and many of the jobs taken will be so-called white-collar ones, those skilled professions where staff works in an office and not a factory. My job is one of those white-collar jobs.
All day, I focus on the valuation of mostly specialty, collector, special interest and antique cars, the whole gamut. I’m an automotive appraiser, and the Publisher of the Hagerty Price Guide. There’s a lot of data involved in valuation as well as the car market in general, and AI thrives on data. In my opinion, though, today’s AI often proves the old adage that “even though you can do it doesn’t mean you should.”
To be fair, I have seen big data-backed AI models used successfully for the valuation of late model cars and trucks, the sort of “every day” vehicles most of us use to get to and from work, run errands, take the kids to practice, run to the doctor or vet, etc. I see these AI models as worthwhile for putting a value on a something like a 2017 Chevy Suburban or a 2022 Kia Soul. Naturally, one might think that AI could be useful in valuing old cars, too. But I also feel that they should not—and cannot—be used to place a value on a 1964 Lotus or a 1967 Ford Mustang. Why? Well, for a lot of reasons, a few of which I’ll list here.
The chances are that the 2022 Kia Soul mentioned above has a pretty simple history. 39,000 miles. One, two, maybe three owners. A small (and now repaired) accident in 2023. Owned by a non-smoker. Three oil changes, regular service. You get the idea. All of those details, (excepting the non-smoker) are one hopes show up in the CARFAX report, which something we can deem, for purposes of this article, a reliable third-party source. And they’re straightforward, recorded data points that an AI model could instantly analyze and weigh against other Kia Souls to come up with a reasonably accurate market value
If we are talking about valuing older, classic and collectible cars and trucks, that’s a completely different story and I have concerns, lots of them.
Back to our 1967 Mustang. The odometer shows 22,500 miles, but is that even helpful? Without an independent, third party source (such as regular, dated and often quite aged repair records or inspection receipts), how can you actually tell? Has the odometer been around once? Or twice? Or three times? Keep in mind that there were no existing federal laws against rolling back an odometer in the first few years of this Mustang’s life. How about a history of accidents? Is there rust that was repaired or is it still present? Again, there might be tell-tale signs, but in the end it’s anybody’s guess.
Was the car repaired or restored? By whom? When? How extensive was the work? How good was the work? Were original parts used? How many owners? In most cases, no one can really know absent original paperwork, which, over five decades on, very rarely exists.
There is more history to a car as it ages. After a largely undefined number of years, a vehicle can go from grocery-getter to an object of desire. Examples here include numerous formerly entry-level BMW and Mercedes-Benz products, domestic pickups, hot hatches from multiple sources, and more. Those previously “undefined” number of years from simply “used” to “collectible” often vary from make to make and model to model. In some cases it might take years, in other cases a car might now be considered collectible while still in production, or very shortly after production ceases. A car that costs $5000 new in 1960 or 1970 can be worth many multiples of that years later. Or, if it doesn’t have the right options or it has a sketchy history, it might not be worth much at all. Ownership can make a difference in value, too. Exceptional records, glory on a race track, celebrity ownership, or a dozen other hard-to-define factors may or may not give it a value bump, and the size of that bump can vary widely.
I’m not just trying to justify my own job here. What I am saying is there can be many variables that continue to come down to intuition and even the human touch. AI in 2025 is complex and complicated, and will continue to be more so in the future, but the collector car hobby and the market surrounding it operate on human factors like emotion, nostalgia, and taste.
Will AI continue to grow, and replace jobs conducted by actual humans in the near future? Bet on it. Is that a good thing? Not necessarily, or maybe not at all.
Remember, the future always arrives even if we prefer it would wait, and it appears we have entered the fourth industrial revolution. You might want to hone your skills and become really outstanding at something, be it gardening, building, or fixing broken things. AI can’t and won’t be able to do everything. There will still be room for problem solvers, big thinkers and people not afraid to do things too complex, or too simple, or too human for AI to make a difference.
Ai would not understand the emotional aspect of collector cars. AI could read car fax reports analyze condition pictures and maybe give a reasonably accurate idea of worth for commodity cars but that’s about it right now.
A bit off subject, however, curiosity and the thought of AI driving cars got the better of me today. I have never liked the idea of self driving cars. Sometimes a choice has to be made as to what is the course of least destruction or physical harm. For example, there is a wrong way car coming at me head on, at night, and the closing rate is 140 mph. Both of us are doing 70 mph. Do I snap to the left, hoping I can jump the guardrail and there are no big rocks. trees or holes on the other side or do I snap to the right and take out the car that is passing me not knowing what their and other traffic’s reaction will be? My lifespan is down to two seconds. Is AI far enough along that we can empower it to make the right decision? I snapped to the right and the driver had correctly analyzed what was going on and hit the brakes. We all passed clear. Interesting that the probable life saving decision was made by a third party. Thank you!
A1 cannot appraise a collector car because of one fundamental characteristic of AI that makes it entirely different from a human brain. AI cannot relate or interact with the world. It cannot touch it, feel it, smell it. Without this feedback, invariably as AI learns more and more, it will probably become less and less accurate. AI just collects data. That’s what it bases any predictions upon–what has already occurred. As for what *IS* occurring at the moment, it has no idea. So AI will be very good at some things, and really bad at others.
Before anything that is collected and put in readable form by AI you first have to realize what AI really is …..
Its a Data trolling engine, that takes its results and puts it into readable form.
As such it tends to pull in results in that are from invalid or inapplicable sources.
Two cases at point ….
If you value a car never sold in the US, such as a Ford Escort Twin-cam it sources prices from Australia, England, and the handful of the cars sold in the US over the last 5 years.
The values in Australia are around 20 to 30% lower than those in England with less than 50 % sold in Australia than in the UK, and total sales less than 20 of this model sold in 5 years – hardly a representative sample.
The ones sold in the US command ridiculous values because of their rarity, often listing modified clones as originals or even actual cars that are not Twin-cams but rather RS2000 or Mexico, or even cloned 1300 GTs fitted with crossflow 1600 engines or worse.
For very rare cars in the US, dependent on the source, the US values vary enormously – for example Hemmings, Hagerty, or club / private sales, versus other locations such as BAT or any of the Arizona listing auction, are often fueled by emotions and silly money.
Bottom line, do your research yourself and look at places where the cars are likely to get good averages prices, and write it down and do a clinical assessment yourself. Then find a reliable 3rd party, suggested by an enthusiast or register club and get them, with the necessary fee to asses the car for you.
Unless you are an expert in the car model and you can see from the photos that it is authentic, well restored, and well maintained.
And for goodness sake stay away from AI – it will lead you up the garden path, its a mugs game !!
AI bases much, if not everything, on available data. If an exclusive car is offered at an auction and there are only two people with enough money who absolutely want that car, what we call auction madness or auction blindness arises. AI will include the, usually much too high, purchase price in its assessment. However, if a similar type of car is offered elsewhere, there is only one person with enough money and therefore no one who will seriously bid against him. It is easy to guess that the sales price will not be as high as in the first example.
As mentioned in the article, an average value of a recent second-hand car or a classic car of which there are still many in circulation can be quite accurate. In those cases, it will always be the market of supply and demand. The value of real gems will then not only be the plus and minus of such a car, but will immediately play a role in the fact that there will be little supply of cars in such a good condition. Whether that is also the case in America, I do not know, but here in the Netherlands, a restoration of an average car into such a gem will always be more expensive than the gem in original perfect condition and yet we estimate the value of the original car to be higher than the version restored to perfection.
The same applies here for the mainly American cars that were later equipped with, for example, a hemi, modified transmission, brakes and suspension. However, in those cases the value on the valuation report will often be higher than an original car of that make, type and year of manufacture. However, my valuation report will clearly state that the estimated value concerns the rebuilding value, so of a value of an identical vehicle INCLUDING all modifications made. After all, an owner of such a vehicle will only be compensated in the event of a total loss if he can purchase an identical vehicle for this amount. And since his damaged car was unique in composition, it will most likely not be for sale anywhere and a purchased vehicle will have to be fully modified again to the specifications of his total loss vehicle. This will, as more time has passed between the modification and the moment of total loss, probably be considerably more than he will be paid by the insurance due to increased rates and parts prices.
On the really very rare cars, the cars of a collector that only come on the market when the owner has passed away or is in financial trouble, it is almost impossible to determine a “real” value since there will be no or very limited transactions to be found except for the aforementioned auctions where the highest bid often only says something about how much someone wanted to buy something and money actually plays no role. The purchased car will be worth to that person what he was willing to pay for it, but is that the “real” value. Will there be someone somewhere in the world with a lot of money who is willing to pay just as much for it?
With which I only want to indicate in the above that the value of something rare will often not be object-related but person-related.
That is at least my honest opinion, whereby I realize that there will probably be many other opinions. In the Netherlands we also sometimes say “better expensive, than not for sale”
I am curious about the reactions to the above and would like to mention that the translation from Dutch to English was done by Google 🙂
AI is flawed, as already mentioned. Yesterday I was looking for the thread size & pitch for a Holley accelerator pump arm, and AI stated unequivocally that it’s 1/4-20 – that’s ridiculously oversized, just from common sense.
(It’s 10/32, by the way.)