Lars Ericson
Sept. 14, 2018
12:29 p.m. PDT

What is Karl talking about?

Here's a way to stay busy until the last question is resolved: Let's try to reverse-engineer the "AI" thing in Almanis that Karl Mattingly is boasting about here:

He is making two claims:

1. "Through collaboration with the Brain Mind and Markets Lab at the University of Melbourne, Dysrupt Labs has applied Artificial Intelligence (AI) to the dataset to increase accuracy. The capability has been built to utilise AI to identify when a poll or a survey or market is possibly wrong. In essence, AI strips out the emotion of the crowd to provide a more measured view of opinion." Hmmm...stripping the emotion out of crowds. What kind of weight function does that give you? Are the coldest people upweighted? Psychopaths especially? Are cold people more truthful or perceptive? Is this anything different from experience weighting Carbon?

2. The magic of.....Combinatorics! "A particular area of interest for the company has been applying combinatorics, a complex theory of maths which enables you to look at the likelihood of a combination of events happening. The more questions you add, the more computing power you need. The research has involved testing the capability and building the back and front-end design to facilitate it." So, they're doing what exactly? They've got a crowd, and an IFP, and people vote, and then they....add more computers. OK, what? They're doing some correlation between different IFPs? Joint probability distribution? Help! What are they talking about?
3 Replies

Mike Jones
Sept. 15, 2018
4:52 p.m. PDT

Lars Ericson
Sept. 16, 2018
6:06 a.m. PDT
OK here's a clue on their use of the term "combinatorics", for a pollution market called "ACE" (no relation to "IARPA ACE") in this talk by CalTech (so he probably knows his generating functions) professor John Ledyard:

So it's kind of an economics/business school thing at first glance. From his biography

"Professor Ledyard's primary research is on the theoretical foundations and the applications of mechanism design. He has contributed greatly to our understanding of the roles of incentives and information in organizations. His theoretical work has provided insights into what is possible and what is not in the design of incentive compatible organizations and voting systems. His more applied work has included the design and development of computer-assisted markets for trading pollution rights, acquiring logistics contracts, swapping portfolios of thinly traded securities, prediction markets, and advertising time. His current research includes the design of market-based approaches for managing spacecraft and instrument design (approaches designed to reduce cost-over runs and improve the science recovered), and the design of cap and trade systems for the control over-fishing and creating sustainable fisheries."

So...HFC. Incentives and voting systems. Not clear yet why this needs an extra roomful of computers. Voting systems are tricky but not compute bound:

Mike Jones
Sept. 27, 2018
7:59 p.m. PDT
In 1950 Edward Nelson, then a student at the University of Chicago, asked the kind of deceptively simple question that can give mathematicians fits for decades. Imagine, he said, a graph — a collection of points connected by lines. Ensure that all of the lines are exactly the same length, and that everything lies on the plane. Now color all the points, ensuring that no two connected points have the same color. Nelson asked: What is the smallest number of colors that you’d need to color any such graph, even one formed by linking an infinite number of vertices?