Jan 2, 2019 · 26m · a16z
a16z Podcast | The Taxonomy of Collective Knowledge
Jay Komarneni · 13m spoken Luis von Ahn · 5m spoken Malinka Walaliyadde · 2m spoken Vijay Pande · 1m spoken Sonal Chokshi · 51s spoken
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this a16z podcast panel, experts Luis von Ahn, Jay Komarneni, and Vijay Pandey discuss collective intelligence, data ontologies, and human computation. They examine how combining human intuition with artificial intelligence drives innovation across healthcare, language translation, and civic governance.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The host holds 3.5% of the talking time here. How this is scored →
The host as informed peer 3.5 Guest teaching 2.8 Guest disagreement 1.3 The host pushing back 1.3
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 22:39 Luis defends human visual recognition superiority
When Vijay challenges whether humans are still better at recognizing cats in pictures than AI models, Luis gently insists that humans maintain the edge.
Hardest push from the host ▶ 22:36 Vijay questions guest claim on AI visual benchmarkVijay directly interrupts Luis to question whether it remains true that humans outperform computer vision algorithms at basic image classification.
Biggest teaching moment ▶ 10:59 Jay explains Clinical Quotient and weighting truthIn response to Vijay's query about minority opinions, Jay details how reference cases establish domain-specific Clinical Quotients to weigh expert consensus over simple majority rule.
The host holds their own ▶ 8:09 Malinka articulates domain-specific healthcare frameworkMalinka sets up the conversation with deep industry expertise, outlining how machine augmentation can elevate lower-tier healthcare workers to handle complex tasks.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Defining Data Ontologies and Their Purpose | 3 | 2 | 1 | 0 | Malinka introduces the core question regarding data ontologies, allowing guests Luis and Jay to define the philosophical and practical foundations of structured knowledge without friction. | |
| Human Involvement and Ground Truth in AI | 3 | 3 | 1 | 1 | The host asks targeted questions about human involvement in training machine models, prompting Luis to explain the ground truth mechanics behind reCAPTCHA and Duolingo. | |
| Scalable Knowledge in Healthcare with HumanDX | 5 | 3 | 1 | 2 | Malinka presents an informed hypothesis about augmenting healthcare workforce layers (N-1) using machine ontologies. Jay elaborate on HumanDX clinical quotients and collective accuracy. | |
| Expanding Ontologies to Governance and Epistocracies | 3 | 3 | 1 | 1 | The conversation broadens to governance, epistocracies, and legal systems. Malinka asks clarifying questions about user weighting mechanisms in Duolingo and HumanDX. | |
| Distinguishing Crowdsourcing, Human Computation, and Collective Intelligence | 3 | 3 | 1 | 1 | Jay distinguishes between key terms while Malinka inquires about user incentive structures, drawing collaborative responses from Luis and Vijay regarding gamification. | |
| Evolving Ontologies and Blockchain Incentives | 4 | 3 | 3 | 3 | Discussion covers evolving ontologies and blockchain tokens. Vijay pushes back slightly when Luis asserts humans are still superior at cat image recognition, prompting a polite exchange. |
Statements from this episode (14)
Luis von Ahn: Many AI systems are just fancy ontologies
“A lot of the things that are quote unquote called AI or artificial intelligence, a lot of times are just fancy ontologies.”
Jay Komarneni: Human input is required for human-interpretable AI ontologies
“The only way you're going to get the representations that are most valuable to humans is from human beings themselves, right?”
Luis von Ahn: Deep learning requires vast human-entered ground truth data
“The other way in which humans are needed here is to create the ground truth. All of these deep learning or deep AI algorithms need A ton of ground truth in order to get very accurate. It has to be entered by humans.”
reCAPTCHA required 10 matching user responses to establish ground truth text
“And the way we got the ground truth here is, is simply by having, you know, 10 different people agree with each other. So if you gave the same word to 10 different people and they all agreed that we consider that ground truth that to a very large extent that w…”
HumanDX: Physician collectives outperform 90% of individual doctors in diagnostic accuracy
“We're now seeing that a collective of multiple physicians can outperform 90 plus percent of individual physicians.”
Luis von Ahn: Most of the legal system is extremely inconsistent
“Most of our legal system is extremely inconsistent.”
von Ahn: Profiling crowdsourced contributor accuracy enables precision with only 3-4 human checks
“When you're doing things that are very large scale and you don't have very many humans, you may have a thousand humans or 10,000 humans. And if you need to label millions of things. You can't afford to start giving, you know, the same thing to all 10,000 human…”
von Ahn: reCAPTCHA digitized 2 to 3 million books annually
“There are a hundred million books that needed to be digitized. That was the total number of books that has ever been written, you know, before the digital era was one hundred million. At the pace that we were going, we were able to digitize about two to three …”
Luis von Ahn: Paying crowdsourced workers is ineffective due to spam
“I have found that paying people is not so good. Then you really have to spend a lot of effort trying to stop people who are just there to, you know, get your money.”
Jay Komarneni: HumanDX rewards physicians based on medical data demand
“In human DX, we actually like to use what we call impact in the system, and it allows us to differentially provide contributors who contribute more valuable contributions to the system in terms of what contributions the system most needs. So perhaps there's a …”
Komarneni: Duolingo inspired HumanDX's gamified clinical case interface
“Louise had actually his work on Duolingo and really kind of this idea of creating these micro interactions that almost have this gamified structure was really a major inspiration also for the way that we built HumanDX to be these brief interactions where you c…”
Komarneni: Token rewards incentivize decentralized collective intelligence and knowledge creation
“The ability to ultimately compensate people with application specific tokens is a really interesting incentive to use ontologies and distributed knowledge creation, collective intelligence to come to better answers around given issues or given problems.”
von Ahn: In 2017, humans still beat AI at basic image recognition
“Computers are better at playing Go than humans are, but humans are still better at recognizing whether a picture has a cat or not.”
Jay Komarneni: Human intelligence excels at multi-scale data synthesis
“One place where that you see that natural place that humans are really good is when there's a high number of scales of different types of information or data, right? So, for example, in healthcare, everything from how your mitochondria or the electron transpor…”