Video Of The Week: The Currency Of Trust

Trust is rising as a central issue in tech and the Internet. We spend a lot of time at USV thinking about trust and talking about trust. It is part of our current investment thesis.

Rachel Botsman is a writer who focuses on trust and her talk at DLD last month was quite good.

#blockchain#crypto#VC & Technology#Web/Tech

Comments (Archived):

  1. William Mougayar

    Rachel “unpacks” trust really well. I think I watched it live back then. She also has a book on the same topic.

  2. Richard

    Anyone raised in NY knows the first question to ask yourself before reading a book – what are they selling ?Ha! cryptocurrency to date has been the crack cocaine of technology – lead by a cesspool of crack dealers – say anything/ do anything to get main st investors hooked.

  3. LIAD

    Rachel has been writing about trust for years and years. Her focus and tenacity is impressive.

  4. awaldstein

    So good.I’ve been thinking since I listened to where you take these threads and add them to communities and markets more generally.With communities it is more complex.Big thanks.Gonna buy her book and follow her thinking.

  5. sigmaalgebra

    It is now clear that much of business, computing, information technology, applied math, and parts of finance are on the doorstep of a grand hall of history making activity and progress for the US and civilization: The roots go back to R. Bellman and have come forward through work of G. Nemhauser (Georgia Tech), E. Dynkin (Cornell), R. Rockafellar (U. Washington), J.-B. Wets (UC Davis), D. Bertsekas (MIT), E. Cinlar (Princeton), and others and the recent work of Google on Alpha 0 and Alpha Go, and now a new book by Bertsekas as at…and an overview lecture by Bertsekas as at…The subject is big progress in practice in best or more likely much better decision making over time under uncertainty with serious coverage of optimal control and the game theory cases and in iterative schemes of approximation of some forbiddingly complicated functions.I call the progress applied math and not artificial intelligence since IMHO the work has essentially nothing to do with how any known biological intelligence works.(1) There is an old recipe for rabbit stew 6hat starts out, “First catch a rabbit.”. (2) Well, I’ve long translated this into a recipe for applied math: “First find an application.”. (3) At this point I would extend this to, “For progress in pure math, first find some applied math that needs refining and extending.”. I see this work of Bertsekas, etc. as close to both (2)-(3).Disclosure: Also related are a lecture I gave to the founder, COB, CEO and C-level at FedEx, my Ph.D. dissertation, and a project I pursued at IBM’s Watson lab.The field is not closely related to the pure or applied math or anything about my startup. For now I will continue to concentrate on my startup since it is such a good direction in general and a much better direction for me.

    1. Michael Elling

      Problem. Most of these theories focus on how individuals address and attempt to reduce risk. What’s more important is understanding how networks serve to reduce risk and how we can optimize that process. To wit, few understand the important role of settlements that drive incentives and disincentives and balance value and cost between actors and between networks in larger ecosystems.

      1. sigmaalgebra

        >Most of these theories focus on how individuals address and attempt to reduce risk.Yes, but compared with the past, what Bertsekas is talking about is especially general in four respects: (1) We get to attack bigger, more complicated, messier problems; (2) we need and can use a lot of data but, really, can get by with much less data than approaches of the past would have needed for such messy problems; (3) the computational needs are typically large for today’s computing but, still, much smaller than approaches of the past would have needed; and (4) the approach to approximating some forbiddingly complicated functions is at least somewhat new and often surprisingly effective but not well understood or grounded and essentially just heuristic.The generality of (1)-(4) come at a cost, in the past considered unacceptably high: we might get some benefits, e.g., save some money and reduce some risk, but we don’t have really good rational support for that and usually won’t have good information on how much more in benefits we might, in practice or principle, have obtained.E.g., what Bertsekas is explaining covers Alpha Go where it was spectacular to beat all competition to their knees under the table, but it’s not clear how far from a perfect game (which quite broad results in game theory for such games have long shown must exist) Alpha Go is. So, we get the benefit of beating the best humans, but before the actual match it wasn’t clear who would win, that is, we didn’t have solid information on how good Alpha Go was — and we still don’t. And we don’t know how well the current Alpha Go might do against something else in, say, 10 years, against ET, or against actual best possible play.Best possible play? We know it exists. And we know we can find it in finite computer time. But for Go, Chess, etc., with all we know that finite time is absurdly long. But we do have some good, intriguing examples: There is the little game of picking up matchsticks, Nim. There is a really super-cute way to play an optimal game. For a while, knowing the optimal solution, I frustrated the C-suite at FedEx. It turns out, commonly can do the logic just between the ears, and an optimal computer program can be short and nearly instantly fast, e.g., no tree search. I wonder, really doubt, if the Bertsekas methods could discover the optimal way to play??? For how to generalize how to win at Nim to other such two player games of perfect information, I’m not sure anyone knows.In my work covered by what Bertsekas is explaining, in some important respects I did much better than Alpha Go did:(1) Ph.D. Dissertation: I started with a precise statement of the problem. [Right away, some of what Bertsekas is talking about doesn’t have such a statement, and there are pros/cons with that.] With that problem statement, in essentially any real, practical, or engineering sense, I got got the best possible result forever. Even ET 100 years from now would not be able to do better.But there are more complicated cases of the problem, and for those my work would have run into too much computer time. I found/invented some ideas for how to do better but the time, money, energy, effort, risk costs of that research direction were too much for me, There some of the function approximation approaches of Bertsekas might have helped but at the costs I mentioned — i.e., broadly we would no longer have so clearly as in my work just what the heck we really had.(2) Anomaly Detection. Motivated by monitoring server farms and networks, I did some new mathematical statistics quite general for anomaly detection. The work did some tricky things with the usual assumptions and extracted some nice results — e.g., got a real hypothesis test with false alarm rate under exact control and got some good stuff on detection rate. Everything else on this problem, IMHO now including what Bertsekas is talking about, has/would have done much less well. But if the problem got even more messy, my work might have not worked or just fallen back to a promising heuristic.Standard situation: Do a really careful design for, say, a light truck. Now use it to carry 5 tons instead of 1 ton as designed, through ice water or salt water, 5′ deep, or across deep desert sands — now are using the design out of the initial problem statement and are not sure what the result will be. Yes, in some cases, get by with that; in some other cases are being irresponsible.(3) Bounding. In one optimization problem, I got a feasible solution. I was not able to show the solution was optimal, but I was able to show that it was within 0.025% of optimality which meant that in the practical context it was essentially as good as optimality. What Bertsekas is talking about in some broad sense could attack that problem, but my guess is that it would have done very poorly, much less well than my 0.025%. Moreover what he is talking about would not usually report a bound on optimality such as my work gave.But, again, as such problems got even more messy, my work would would have been just another heuristic or failed, and what Bersekas was explaining might have given the best approach.So, I derived some applied math and wrote some software to do the arithmetic. The math had theorems and proofs, said there would be a bound, but didn’t say how close that bound would be. So, the math was less powerful than we would have liked. In particular, before I ran the software on the real input data, I didn’t know how close the bound would be. So, the 0.025% was good news.So, net, my math presented a method. In practice, one problem example at a time, get to try it and find out. WILL get a bound, but in a particular problem the bound might be too large to have the results useful — then have wasted time. The bound might be nicely small — then be happy and take the savings to the bank. Even if the bound is large, the result still might be better than what have already in which case again get a nice trip to the bank and also get some motivation to reduce the bound.The method we would want would be at least as good as showing that P = NP, and no one knows how to do that.For what Bertsekas is talking about, we don’t get just one tool but get several and on a particular problem have to try them one at a time and see which one we “like”. For some really messy problems, that can be progress.For my work, I always wanted to have the result rock solid, with the problem in a Kevlar body bag nailed down at all four corners, essentially no doubt at all about the quality of the results.. I didn’t want to have to guess, hope, or one problem at a time use the TIFO method — try it and find out. For the applied math in my startup (not close at all to what Bertsekas is talking about), again I have some solid results from some just astounding theorems.So, yes, the Bertsekas lecture can open the door to heuristics, guessing, and hoping. For some especially messy problems, that can be big progress.But we are setting aside solid proofs and other good evidence of quality. So, the door is also open to charlatans who would do more of the total sewage they have often done so far with artificial intelligence, machine learning, and “climate science” and the total flim-flam, fraud, scam, wildly overly emotional, hysterical, irrational, destructive, dangerous screaming about human activities hurting the climate.In the past, in medicine the charlatans sold plenty of sewage as snake oil, but eventually we got a lot of good professionalism in medicine. Still, around the edges and close to violating the laws, the charlatans are still selling medical nonsense.The main needed reactions have been clear at least back to The Enlightenment and The Age of Reason: Insist on good evidence. Some of the best cases are mathematical proof and solid physics. Then there is more that is solid in engineering and medicine. In law we have made some progress — e.g., innocent until proven guilty, face accusers, cross examination of witnesses, etc. The best work is astounding — e.g., the pictures from Pluto, first try, as planned, great pictures. Not big surprise — Pluto is a ball of darned cold rock with some evidence of geological activity. No evidence of ET!!! Still, darned good pictures and AMAZING rocket and space flight engineering based heavily on some of the best physics.Generally, buyer beware; measure twice and saw once; etc.Still, what Bertsekas is talking about can have some terrific results, especially on some function approximations. Bertsekas has thought about such approximations before.Yes, it’s applied math that is screaming out for some good pure math to explain, make solid, and generalize.Would I want to do such research? It’d be fun. But in two words, it’s for the researcher financially irresponsible.In strong contrast, my startup lets me sell something and keep the earnings.And, my servers can keep working and making money while I sleep, eat, take a sweet, pretty, darling, adorable, precious, bright … woman really good at music, maybe knows English, German, French, Russian, and Italian, good at cooking, sewing, home decorating, children, and is sweet, did I mention really sweet, maybe she grew up near the North Sea, to an opera or some such.And once a month or so, I’ll look at the deposits in my bank account from the ad networks, and once a quarter have my tax accountant work up the details and make the Governor of NYS really, Really, REALLY, REALLY happy and paying REALLY close attention, calling me back, etc. at any hint I might want to move my business to, say, Tennessee!Each chance I get I’ll tell the Governor that each time I hear green I think again about the tax rates in Tennessee, Texas, Indiana, Kentucky, West Virginia, Montana, etc. and how eager the people there would be to get my information technology training program and in my company make money enough for a good McMansion, with stay at home moms, lots of kids, and really GREAT private schools and neighborhoods for the kids and families!!!! The company should have an annual summer BBQ celebrating the kids! In the server farm, each server rack should be named after a daughter of an employee!!! Right, ONLY daughters!!! The sons? They get summer jobs pulling cables (assuming we still use cables!), building, installing, and maintaining the servers, doing the HVAC BTU and KWh arithmetic, getting started on server and network system management, getting started on coding, algorithms, data structures, and pure and applied math, etc.!!!!While I want to remain anonymous and NOT a public person, if hysterical, irrational, hopeless, helpless, wildly destructive, to both herself and even civilization, ditsy bimbo AOC starts screaming about my startup, e.g., insisting that I put solar panels on the roof and power the air conditioning with gerbils running in wire cages, she’s had much worse ideas, then THIS TIME the Governor will tell her in no uncertain terms to take a swim to Iceland, alone, in the winter or just SHUT UP. Maybe the Governor will call up his really good buddies at ABC, CBS, CNN, MSNBC, NBC, NYT, WaPo, Boston Globe, etc. along with Nancy/Chucky and tell all of them to put AOC in heavy duty “time out” never to be heard from again for at least 6 years, say, once she is mature enough to graduate from pre-school or, to be wildly optimistic, enter middle school. E.g., in pre-school, she needs to learn that she can’t just read some fantasy story and, then, just reach up, grab a sky hook or a moonbeam, and get lifted to the roof. She can spend all afternoon each week reaching up and trying before she gives up and starts to learn. She can learn that getting the groceries home on a tricycle is a real pain, never to be attempted again. With a good pre-school experience, she can move into kindergarten.For what you said about “networks”, I didn’t quite understand. But maybe you are mentioning some problems with more aspects, considerations, or generality than commonly assumed in the past. Okay, fine. It may be, likely is, that the Bertsekas work would apply about as well to your problems as to many others. But, as in my examples above, for some of your problems, there may be approaches, particular to those problems, that would totally blow the doors off what Bertsekas is mostly talking about.

  6. Dan Blechner

    that’s a great one, thanks for sharing

  7. George Whitfield

    Thank you for sharing Rachel Botsman’s video. I enjoyed her book as well, which clearly describes societal problems related to widespread breakdown of trust. I believe that we can improve this situation by giving people control over their attention towards each other. Our choices can form a fabric of shared trust, enabling broader access to collective intelligence.