In a conversation with Tyler, Reid Hoffman said,
I almost never meet with an entrepreneur that doesn’t come from an introduction from someone I trust. . .
the way that I do investing is entirely through my network
He also says,
most jobs are described as “must have a BA or a BS,” a bachelor’s degree, whereas a lot of jobs don’t actually need that. But it’s like, “Okay, what’s the simplest credential that everyone’s aware of that I could throw on the table that says you have some capability of learning, and you’ve been trained in some learning institution?”
And one of the things that I actually wrote is an essay that’s, I think, on both LinkedIn and reidhoffman.org, is thinking about creating a diploma that’s not this old sheepskin, but actually is a modern set of attributes and set of characteristics. And we could start looking at these certificates as something that has a much richer language that can apply to different things.
Can entrepreneurs, especially using artificial intelligence, create a more efficient solution to trust problems than Reid Hoffman’s reliance on his personal network, or hiring organizations’ reliance on college diplomas? There is a lot of money to be made if you can solve trust problems more efficiently. That is the way that financial technology firms make their money.
Or how about the trust problems involved in evaluating contentious books? Consider Nancy MacLean’s attack on James Buchanan, or Diana West’s books on Communist influence in the U.S. government.* Could an AI program walk through the webs of sources of such books and give a measure of the reliability of the narrative, which presumably would be less costly than having humans try to settle the issue?
*West argues, for example, that Lend-Lease was passed under Soviet influence. Even though she knows that it was passed in March of 1941, before Hitler surprised Stalin by invading Russia. In March, the Soviets were adhering to the Non-Aggression pact with Germany, and only Great Britain was eligible for Lend-Lease.
Not currently. This type of functionality requires a breakthrough in Natural Language Understanding (NLU). The best we have now is Sentiment Analysis which is crude and requires a large data set from which a rough probability is produced.
The lack of even rudimentary NLU also places a hard constraint on the likelihood of automating credential analysis. Without NLU, you need to create some kind of structured document or test to represent credentials and take this system mainstream.
A good rule of thumb for any entrepreneurial idea is that it should produce a 10x improvement along some measurable dimension. I don’t see the opportunity but that might be due to my lack of imagination.
What I find interesting about Hoffman’s trust network is that it clearly went from zero-to-one and the PayPal Mafia and it’s many connections have been written about extensively. It rode the wave of The Web (1995). Perhaps one of more webMafias is the natural outcome of innovation revolutions and Hoffman’s skill set was simply good enough for the time and place he resided in.
I’ve long applied Hoffman’s approach to hiring educators for my teams. I hire teachers/other only if they come from a school where I can get a reference (peer or boss) from someone I trust, even if it’s “second degree” (a friend of my friend). It restricts me to <1% of the possible pool but in absolute numbers it's still a decent size pool.
Hoffman misses 99%+ of investing opportunities but what's left is still a decent size pool.
So in this example, AI needs to meet a really high bar, right? Needs to outperform the personal network of someone with a good personal network.
In typical example, let's say a school district that doesn't even try to use its combined personal networks, and instead uses a terrible proxy like "interviews" and "cover letters" and "fake letters of recommendation," AI need only beat a low bar.
I kept thinking about this sentiment and Hoffman’s network as I started reading “A Time to Build” today. Levin says that one of the defining features of an institution is that they are durable. I think of an institution as a collection of people, processes, and purpose. The people, however, are defined by roles rather than named individuals; it is the roles that are durable.
Hoffman’s network is a product of his unique experience, and perhaps his personal narrative. Greylock Partners is an institution that benefits from Hoffman’s network but, if he leaves, I think the institutions retains nothing of value.
Maybe “personal networking” is the most important benefit of attending or belonging to specific institutions and that value doesn’t seem to be transferrable.
I hope we don’t rely on AI for any of this.
Rather, I think the solution is to be found in reorganizing relationships to allow for better trust. Employment trust, or financial trust, or even internet trust has the same problem. Directly establishing trust between parties is intrinsically difficult because the relationship starts out cold, individual transactions justify only a limited effort, and there is a mostly negative set of incentives for third parties to provide a useful reference.
It would be more efficient for a worker, online user, or someone seeking a credit transaction to always work through and maintain a long term relationship with a trust institution that they pay for and has a fiduciary responsibility to them. The trust institution would know about the entire record of the individual’s relevant transactions and could maintain a reliable model of that person’s long term behavior. This could allow them, in turn, to be trusted to attest to it in a predictable way that markets could accept. Incentives on both sides would be better aligned.
Trust requires experience. That cannot be innovated away. We have a complex economy, and peer to peer is efficient, but there is no way to make it trustworthy.
Can entrepreneurs create trust using AI? I’d argue that is what Bezos did.
I’d argue the college diploma problem is a policy problem resulting from Griggs v. Duke Power. Dr-incent the plaintiffs bar from suing every alternative to diplomas and you would go a long way towards eliminating the current degree related deadweight losses and distortions in the labor market.
An AI book review would certainly be useful in affirming confirmation biases and ignored when they do not. But contentious books typically impute motives. We have no way of knowing motives, at best we imply them based on a totality of circumstantial evidence. AI might be useful in coralling that evidence but weighing and identifying motivations is largely masturbation and the centrality of such considerations to contemporary scholarship goes a long way in explaining its irrelevance and worthlessness.
This is a smart comment — any alternative to a B.A. or B.S. will run into the problem of disparate impact, which is already a problem for secondary and higher education.
Ironically, a diploma requirement also have a large disparate impact. A much higher percentage of whites have them than blacks or hispanics. But no one cares enough to contest the diploma requirement, because education is just about sacred in respectable opinion (and education equals going to school and passing).
I used to think “the college diploma problem is a policy problem resulting from Griggs v. Duke Power.” But now I’m not at all sure. How are things in Europe? In Japan? In China? I get the impression there is a similar “hiring organizations’ reliance on college diplomas”.
Reliance on diplomas existed well before Griggs. Charles Peters of the old Washington Monthly railed against it. For the hiring organization diplomas ARE an efficient screen. After all, they aren’t paying anything for them. And as Bryan Caplan points out, diplomas aren’t just signalling intelligence or a certain amount of knowledge. They are also signalling conscientiousness along with the ability and willingness to “play the game”, to figure out how to get the diploma and then to do it.
I disagree that diploma-ism is a universal phenomenon. In Europe, where the middle class is doing better than in the USA (https://dispatcheseurope.com/pew-research-european-middle-class-stable-even-thriving-compared-u-s/ ) but without all the diplomas. 27% of Germans, for example, have diplomas compared to 44% in the USA. And the percentage of the population with diplomas in the USA has increased rapidly since Griggs. Canadians have more diplomas per capita than the USA and not surprisingly the OECD reports that the middle class is shrinking faster there than any where else.
Is the German middle class better because they avoid the college degree credential or is it that Germany apprentice/unions are great at creating good working/middle class. (It was Megan McArdle that pointed out the German apprentice system might work because the union/labor laws are more pro-worker.)
The reality was in the Post WW2 boom years there was not as much of need for college degrees although it was growing. High School Vocational training was respected in 1950s but it is not today or when I went to HS in the 1980s. What aspects of that society could be used to today to counteract the private business focus on degrees?
It is also wise to remember in 2000 most free market economist were arguing against Germany (which economy looked stagnant since 1989) and exceptionally highly praising the US economy.
Some argue Bezos is the singularity. It appears the end is nigh, at least for plausible arguments. In other AI generated headlines from the (Bezos owned) Washington Post: the world breathes a sigh of relief as it learns that credentialism is restricted to American jurisdictions.
I don’t have an example that comes to mind, but I would think that looking at actual examples of jobs that went from not having a degree requirement to having one would be helpful.