AI: Asymmetric Information
Imagine that you are a used car dealer. You have 100 cars on your lot, each of various quality. Some will explode within the first mile while others will function better than a new one. We can assign each car a “quality point” score. The worst car has one quality point, the second has two quality points, and the pattern continues until the top car has 100 quality points. As a dealer, you value each quality point at $100. That means you’d sell the worst car for $100 and the best for $10,000. If someone offers you $9,999 for the best car, you’ll reject the offer. If someone offers $10,000 or more, they get the vehicle.
Naturally, the buyers value the quality points more than the sellers do. If they didn’t, no sales could ever be made! Buyers are willing to pay $150 for each quality point. That means buyers would pay $150 for the worst car, $300 for the second worst, and $15,000 for the best.
Simple, right? You can sell the worst car for between $100 and $150, the second worst for between $200 and $300, etc. Here’s the catch, the buyers don’t know which car is which. They can’t tell the difference between the 1-quality-point lemon and the 100-quality-point peach.
Of course, you could be honest and direct the buyer to the right car. That’s lame, though, so you’re going to have to imagine yourself as a dishonest jerk. What sales can you make?
Start from the perspective of an individual buyer. Since he doesn’t know which car he’ll get, the purchase amounts to a gamble. On average, the cars on the lot are worth $7,575 (50.5 quality points) to him. That’s the most he’s willing to pay.
Great news! You, the seller, value each quality point at $100. That means you can sell him any car with 75 quality points or less. You pick one of them (probably the worst one, might as well get that trash off the lot) and offer it to your buyer for just under $7,575. You get the paper and tell him to sign at the dotted line. He refuses.
What happened? Here’s the problem: you know too much. You know the quality of each vehicle, and the buyer knows that you know. Thus, he reasons that you could give him one of the crappy ones. Remember, the $7,575 was the expected value of all the vehicles on the lot. But you’re not picking a random car from a hat. You can choose which vehicle to give him, and it’s in your best interest to give him one of the stinkers. Even though 75% of the vehicles on the lot would be a good deal to the buyer, you can’t sell a single one.
That’s the gist of George Akerlof’s 1970 paper The Market for Lemons. Asymmetric information (one side of the transaction knowing more than the other) can prevent otherwise mutually beneficial transactions. If neither the buyer nor the seller knew the value of the cars, both would just agree to gamble. If both knew the value, they’d negotiate on an individual vehicle. However, since one side knows and the other doesn’t, the market collapses.
Cool theory, but, you know, we do have used car markets. Why is that?
There are a few different reasons. For one, we have government regulations. Although I’m not sure how valuable these are for used cars, we can see their importance in drug and food safety. Another solution involves seller signaling. This is why companies offer warranties. When Toyota attaches a three-year warranty to their new vehicles, they’re telling you “Hey, if we’re wrong about the quality of our cars, we’ll pay the tab.” Word of mouth can also solve the problem of asymmetric information if the business relies on repeat sales or referrals. One common piece of travel advice is to avoid restaurants near major landmarks. People may eat at these establishments once and never see them again, so the restaurant can offer trash at a high price. However, a restaurant that serves locals needs to give patrons a reason to return. A good example of a referral-reliant industry is Hollywood. Industry journalists often break down a film’s gross into its “opening weekend” and “multiplier.” A movie that earns $10M in its initial Friday-to-Sunday release and $70M over its total run is said to have a multiplier of 7. A substandard film can make a ton of cash in its opening weekend, in the subsequent weeks, but those early moviegoers will tell their friends “That movie sucked, stay home and read the Simplify backlog instead.” … that is what you tell your friends, right?
Finally, there’s the squishy truth that people tend to trust each other. This level of trust can vary from place to place. I recently read a history of American immigration called Streets of Gold. In it, the authors discussed Korean immigrant communities where businesses manage loans without any written contracts. That’s a level of trust you won’t see in many other places. I don’t even trust myself that much. On the other end of the spectrum, there are areas where everyone knows that everyone is full of crap. I’m talking, of course, about LinkedIn.
The Online Job Market
Prospective employees find white-collar jobs in one of three ways:
Referrals
Recruitment
Cold Applying
The first one, referrals, remains a bit overrated. I’m sure there are some areas where they matter a lot, and it’s helpful for a first job. In my personal experience, however, few people obtained their current role via referral. The best office jobs come from a recruiter sending you a message via LinkedIn InMail or an industry-specific equivalent. Contrary to whine-fests you see on LinkedIn feed, many people also find gainful employment by throwing their resumes into those giant piles.
I was, of course, being a bit tongue-in-cheek when I said that everyone is full of crap on LinkedIn. Everyone may exaggerate, but recruiters do find useful info on profile pages and resumes (cover letters, though, seem to have gone extinct.) Yes, people can hire professional resume writers, tailor their resumes to a specific job listing, or use today’s relatively primitive robots. I did so in 2021 to land my current role. Still, these activities test for a level of conscientiousness, drive, and cultural fit. This might not be the case going forward.
I’m not going to pretend that I can predict the future of artificial intelligence. Instead, I’ll present of future that, to me, seems relatively conservative compared to most predictions. Right now, anyone can subscribe to LinkedIn Premium for a boost in visibility. In the future, the site could offer LinkedIn Super-Duper Premium. With this service, any resume you submit will be automatically tailored to the job description. LinkedIn can train its model on previous applications, so the model can re-write your resume to look more like the ones that have landed jobs in the past. The AI-infused subscription could also re-write your LinkedIn profile to match those of successful applicants. Maybe it could even show different profiles to different recruiters. For example, the AI could show a more coding-oriented profile to data science recruiters and a more business-savvy one to recruiters looking for operation skills.
In our current world, recruiters can filter candidates by scrolling through their LinkedIn Pages while hiring managers can filter candidates by sorting through resumes. If my predictions come to pass, however, these systems lose reliability. Everyone, even the worst candidates, will boast perfect profiles and resumes.
The New Lemons
This leads to the familiar problem of asymmetric information. Candidates know their value. Even though we all indulge in some delusion and self-deception, most of us roughly understand our relative strengths and weaknesses. The employer, meanwhile, won’t know anything. We therefore find ourselves back in the market for lemons, with the “lemons” being low-quality candidates. As with our real-life used car markets, I don’t expect AI to end the labor market and every position to stay open indefinitely. Instead, businesses will need to solve the asymmetric information problem. From our discussion above, we saw that asymmetric information can be overcome via government regulation, repeat sales, signaling, word of mouth, and trust. None of these apply to every market, but they share one common trait: they involve a level of human connection. To sift through the labor market lemons, firms will need more recruiters.
Anyone who’s hired knows that most resumes are trash. That’s why you should apply for that opening with 200 applicants. If your resume fits the job description, it’s probably ahead of 195 of them. Although I’ve never been on the recruiter’s side, I imagine the ratios are similar for LinkedIn profiles in the sense that most can be easily slotted into the “no” pile.
In our robot-heavy future, however, this may not apply. All 200 applications might be perfectly tailored to the job. In addition, if AI makes it easier to apply for jobs, the figure might sit well above 200. The employer’s only option is to speak with a larger number of candidates. I’ve conducted interviews, and I can usually spot someone who doesn’t have the chops for an analytics position within a few minutes. Most hiring managers tell similar stories. It doesn’t take that much time to spot a faker, but it does take a human. Future firms will need a combination of people with strong social skills and expertise in the relevant fields to filter out large quantities of bad candidates with perfect resumes.
A Test of Time
After the pandemic shut down in-person testing centers, several universities dropped SAT and ACT requirements. Many colleges continued this policy in the name of social justice, though sanity seems to be prevailing. Yale, MIT, Dartmouth, and other schools have reinstated testing requirements, and I predict that other colleges will soon follow. The logic is simple: it’s better to compare students via a homogenous exam than through high school classes or extracurriculars that can vary widely from school to school. The same logic applies to jobs. I understand that some legal barriers prevent companies from hiring via general aptitude tests. I will not pretend I know what laws or court decisions could reverse this. I’m just going to pretend, for the sake of this article, that these prohibitions disappear in the future.
Though estimates widely, there is some correlation between IQ and job performance. That’s a good start, though employers will want tests where the correlation is stronger and more consistent. Future companies may run trials to craft exams that predict the highest performers. We might therefore end up with a sort of “workplace SAT,” and hiring managers could filter resumes via this test score. Of course, firms would need to ensure that the test itself can’t be taken by AI. I remember reading about people swapping their smartphones for less-addictive flip phones. AI might create a similar market for “dumb” computers that lack the technical capacity for AI. Tests will be taken on these computers (if not by old-school pencil-and-paper).
Alternatively, we may see a rise in certifications. These already exist for some professions (think accounting and engineering), and AI might turn that “some” into an “all.” Today, you can land a data analyst job with a mildly technical degree and a good resume. Tomorrow, you may need a general analytics certification or a bundle of certifications in smaller areas.
Where will these certifications come from? Education scams are already easy in our low-fi world, and AI will only increase our capacity for fakery. Thus, I imagine that most of these certifications will come from established institutions. These may include prestigious companies like Google and Amazon alongside name-brand universities like Harvard and Stanford.
Three Implications
I’ll organize my conclusions into three major buckets. The two AIs (artificial intelligence and asymmetric information) will impact the job market in the following ways:
The value of social acumen will increase. First, firms will need an army of recruiters to differentiate good candidates from liars. Second, socials skill themselves will matter more for the candidates. Since people can no longer differentiate themselves on paper (or, more accurately, on PDF) the best jobs will go to those who can market themselves in virtual and in-person interviews.
Firms will invest in aptitude testing. Universities are re-discovering the value of exams. Companies will do the same.
Established institutions will gain influence. In a world where anyone can produce infinite amounts of nonsense, firms will value certifications and experience with tried-and-true institutions.
A few notes:
There's two things you didn't touch upon with used cars (an area I'm very familiar with although, admittedly, from a while ago, due to working for my cousin's dealership). The first is the Carfax report. Used cars are almost never sold to even remotely savvy buyers without a Carfax report. Leaving aside how people put too much faith in them, I'm curious what the analogous version of that would be here.
The second is the test drive. Again, too much faith is put in a test drive. But the ability to use the product - even if only for 5-10 minutes - is an important factor. I've often thought something similar will be important to hiring in the future. Your (excellent amd thought provoking) article raises my belief in that.
Really, aptitude testing is going to need to be done. Interviewing gives an idea for "will they be a culture fit" but very little for performing the job. In my industry, we go very down the line. The idea being that because the LSAT is akin to an IQ test, employers look for signs someone has a high LSAT - ie, attending a prestigious school or working for a prestigious firm - even if those rarely correlate with improved skills beyond aptitude. Unfortunately, the much more direct route is illegal.
One final thing that seemingly does not apply to the tech sector but certainly applies to my (and I suspect many other) industries: the job market is often doubly harmed by lack of information. Not only do applicants hide much information, but so do employers. Often the idea is get the employee in the door for a shit job because they won't quit (due to the disastrous effect on their resume) so you mislead them about the job they're applying for. It's not particularly relevant to your main points but it's an interesting (to me) part of the analogy for many industries.