Kategoriarkiv: Tech Policy

Vi är alla adjektivens fångar eller den vulgärempiriska wittgensteinianismens framväxt

Jag har absolut för mig att Jeff Goldblum i Jurassic Park vid ett tillfälle säger nedlåtande att ”adjectives are the crutches of weak mind” – men jag kan inte hitta citatet, så jag får väl nöja mig med att tänka att det vore bra om han sagt det. Adjektiv beskriver oss, och det betyder förstås också att de begränsar oss — om man kunde se till hela världslitteraturen och spåra de olika spänningsfält som adjketiven låser upp oss i så skulle den landskapsskissen säga mycket om oss som människor och om våra inre liv. Det skulle möjliggöra en sorts vulgärempirisk wittgensteinianism som skissar upp språkspelen och ger oss möjlighet att se hur tänkandets avlagringar definierar oss.

Det visar sig att det är fullt möjligt att göra just detta. Ett gäng danska forskare publicerade just en intressant artikel om vilka adjektiv som används för män respektive kvinnor i mer än 3.5 miljoner böcker (ur Google’s samling digitaliserad litteratur). Resultatet är inte överraskande men talande:

Screenshot 2019-08-28 at 11.42.03

Det är lätt att se enkla saker som att de positiva adjektiv som används om kvinnor oftare relaterar till utseende, menar forskarna och det är också uppenbart hur mansidealet ser ut när man ser ord som ”honorable” och ”brave” och ”just” och ”righteous”.

Inte överraskande – alltså. Men intressant – särskilt i en tid när vi diskuterar hur vi skall kunna komma tillrätta med enkla schablonbilder och representera världen på ett bättre sätt. Den tröghet som detta projekt möter är inte främst teknisk – det handlar inte om algoritmer – det handlar om oss, om språket och om vårt tänkandes utveckling de senaste tusentals åren.

Artikeln i sin helhet finns här.

”Valuable speech” – a note on legal mechanism design

I am reading a series of essays on free expression on the Internet. One of the authors repeatedly uses the ideas ”low value speech” and ”valuable speech”. I feel great unease. I wonder why, but think that it is because such a dichotomy assumes that we can say that this piece of speech is valuable and this other piece lacks value. Am I more comfortable with thinking about this problem in terms of ”speech” vs ”criminal threats / defamation”? Oddly I think so. I would like my speech with as few qualifiers as possible, and then I would like to define that which should not be protected as something else. As criminal defamation or illegal threats, or something else.

I think the reason is fairly straightforward: it imposes an intellectual discipline on the legislator, and associates limiting speech with a threshold test. It is a question about design.

In general there seems to be at least three legislative design strategies here: one is to try to categorize and qualify the speech as such according to its inherent value, one is to concentrate on the medium in which it is expressed (Mill actually seems to have been leaning towards this strategy, favoring deliberative debate in the newspapers of his time over talk in the street) and one is to simply define everything as speech (and thus protected) that is not criminalized, making a point out of differentiating between what is speech and what is not. All three systems can rule that something should not be protected, but in different ways – and it seems to me that the method, the algorithm, the mechanism design here really matters.

Perhaps we spend to little time thinking about legal mechanism design.

Metrics, Models and Methods – Wired for Innovation

But what is it?

Tech policy poses a number of different challenges. The perhaps most interesting one is that we do not agree on what the subject is and how to measure it. What is technology? How does it impact things like our economy? In Wired for Innovation Erik Brynjolfsson and Adam Saunders try to attack that issue in a number of different ways, essentially saying that our old measure of, for example, GDP, fail to capture the majority of the values created by technology brings to the table. The perhaps most powerful example in the book is E-bay. Brynjolfsson and Saunders note that research shows that Ebay creates 4 dollars of consumer surplus value (notoriously hard to measure) per transaction, totalling, back in 2003, 7 billion USD. The actual captured value in GDP for Ebay the same year? 1 billion USD. New services and products may well create something like 7 to 10 times the value that is captured in our old measures. This will necessarily lead us to make decisions that are at best suboptimal, and at worst really hurtful. When the likes of Nicholas Carr argue that technology does not matter, or, in fact, that it is crippling us, he does so from these old measures (and not only of economics, but also from an old conception of what culture and thinking is).

Getting the numbers right thus needs to be a priority. And getting them right means new models for value creation and capture.

*

In Wired for Innovation the authors note that there are now new methods being developed for capturing these new values. Measuring consumer surplus by looking at time is one. Currently, in our GDP-model, the value of Internet Access is equal to the fees that consumers pay for internet access, they note, and nothing more. That is roughly 100 dollars a year and consumer, which then becomes the value economists will say that the Internet has to that consumer, too. Added up that makes 0.2 percent of GDP. That is the value our models have of the Internet.

The average time people spend on the Internet of their leisure time is 10 percent. That means that we spend 10 percent of our (free) time on something that is worth 0.2 percent of GDP. Not necessarily insane, right, but still thought provoking. Measuring time gives another estimate of worth however, and with this method the average value of internet access to a consumer is 3000 USD a year. 30 times the GDP-estimate. Now, all of these methods are in their infancy, but they are usable, and should be – as the authors argue – developed in much greater detail.

*

It is well worth while to follow the work in MIT on this issue, here.