One of the examples that comes up in the Robin Hanson podcast concerns dating. If the point is to signal to the person that you are healthy, wealthy, and intelligent, why not just bring your health records, your bank statement, and your SAT scores?
One hypothesis, which may be implicit in what Robin is saying, is that you would not want to date someone who went for those sorts of signals. Somebody who will hook up with you based on such simple information could just as easily cheat on you or dump you when someone else comes along and provides similar information. We want to go out with people who are "discerning," which in this context means that they signal a habit of processing other people's expensive signals. That in turn makes it more likely that they will be loyal, which is a quality that Robin thinks we often need to signal.
So you have what I might call co-operative signaling. You signal to me that you are looking for "deep" signs of my health, intelligence, or whatever. I take that as a signal that you are into long-term relationships. So I then go through the rituals of dating and signaling.
The issue is whether one can tell this story in the job market. If you are an employer who gives an intelligence test rather than looking for a Harvard degree, you may very well hire more efficiently. But to me as a potential employee, that means that you might replace me quickly and easily with someone else who does well on the intelligence test. Instead, if you require an expensive signal, that might make me confident that you are more loyal. So the co-operative signaling arrangement is that you signal that you make long-term investments in employees by using a Harvard degree as a hiring criterion rather than intelligence test, and I in turn signal by getting a Harvard degree.
This model explains why an expensive signal (the Harvard degree) persists in a world where cheaper signals are available. Employers who use cheap signals are (perceived to be) less likely to be loyal to employees, so they lose the opportunity to hire employees who value loyalty.
By the way, I am not such a fan of the signaling story, so I may not necessarily be putting it in the best light.
UPDATE: In a comment, Robin Hanson points out that this story will not work unless the "good" firms have some comparative advantage in using the signal. To make it work, I could posit the following:
1. One firm, call it McKinsey, is a desirable employer because it adds a lot to the worker's human capital.
2. The Harvard MBA is a noisy signal, and McKinsey, with its Harvard MBA's on staff, is best able to tell the good Harvard MBA's from the bad. (this is the crucial step for addressing Robin's comment.)
3. While an intelligence test is more cost-effective a signal than a Harvard MBA, the combination of McKinsey's inside knowledge and the Harvard MBA is more cost-effective than the test.
4. Workers are not sure whether they are top quality or not by McKinsey's standards. They have a probability p of being top quality, but they have to graduate Harvard (or come close to graduating) before learning whether they measure up. (I think I need this step. Otherwise, inferior students would never try for a Harvard MBA, and the signal would not be noisy.)
In equilibrium, the top quality workers go to Harvard and on to McKinsey. Lesser-quality workers try Harvard and fail to measure up, or else they go straight to inferior firms. The inferior firms never get to see any of the top quality workers.