Surveilled #55 - GPT-3, unpredictable complexity and an animated GIF botnet
Surveilled
Issue 55
A weekly summary of what I’ve found interesting at the intersection of economics, finance and technology.
A shorter edition this week, as we are still establishing new routines in a new environment with new constraints. This is in itself a fascinating exercise, with its implications for productivity, creativity and stress levels. Maybe worthy of a longer rumination at some point. With that, on to this week’s newsletter.
Published Elsewhere
Malaysian weekly The Edge published my op-ed piece on revisiting the role of government in the economy, in the wake of the Covid-19 pandemic. Read (The Edge Markets)
Six Links
-
“AI Research and Deployment Company” OpenAI released a new text-based general artificial intelligence tool called GPT-3. Invitation-only for now, but some developers showcased some of what it can do, and on the surface it looks impressive. As far as I understand, the mechanics are not new, but the training set for this model was orders of magnitude bigger than previous models, hence the far better results. That doesn’t necessarily address issues with bias that we discussed before, and the model is a black box, so fixing any issues will not be easy. The Verge has a very good in-depth look, and there is a tl;dr version as well. Read (The Verge), tl;dr (The Next Web)
-
US pharmacy chain Rite Aid deployed facial recognition systems in hundreds of its stores, usually in lower-income and non-white neighbourhoods. The article plays up the system provider’s links to China, but I think the most worrying aspect is the discriminatory effect, both in the choice of locations where the system was installed, and the built-in bias of the facial recognition algorithm, which regularly misidentified customers, and hence denied them service. Read (Reuters)
-
Matt Stoller, in his BIG newsletter, wrote an interesting look at the unpredictable effects of seemingly insignificant, or at least unrelated, changes in the Big Tech companies’ platforms or policies. For example, the Facebook advertising boycott, launched by some big brands in the wake of the Black Lives Matter protests, has ended up benefiting Chinese counterfeiters. The article has other examples, and points to, first, the difficulty of regulating this, and second, the high unpredictability of living with very complex systems such as these. Read (BIG newsletter on Substack)
-
In the midst of the Covid-19, with their primary business essentially in crisis, companies such as AirBnB and ClassPass started offering virtual classes, through their apps. This attracted the attention of Apple though, who started demanding the standard 30% cut on revenue for the App Store. As mentioned previously, the 30% rule seems less and less defensible, and heavy-handed application like this will surely precipitate trouble for Apple… Read (NYT $)
-
A while ago, the EU put in place a legal framework that enables the transfer of data from the bloc to the US, requiring that the law in the destination country provides “essential equivalence” to EU law for data protection. All Big Tech companies such as Facebook and Google have been relying on this framework for their EU-based users. Now, in response to a lawsuit by privacy activist Max Schrems, the European Court of Justice has determined that US regulation is not “essentially equivalent,” and has in effect struck down the existing arrangement. It’s not clear yet what comes in its place, but one consequence may be that more US firms will store and process data in the EU rather than in the US. In 2015, Schrems already managed to shut down a similar agreement that preceded the current one, impressive achievements whichever way you look at it. Read (TechCrunch)
-
Some context: a botnet is essentially a collection of hacked websites that, when visited by unsuspecting users, install malware, viruses etc. (called the “payload”) on their machines. A group of hackers remotely controls all these websites through special software. It turns out that one of these, called Emotet, was in turn hacked by a vigilante, who replaced the malicious payload with animated GIFs on a significant number of websites, thus scuppering the botnet’s effectiveness. I find stories like these fascinating, and wonder if this will one day become an institutionalised way of combating this kind of threat. Read (ZDNet)
That’s it for this week’s edition. As always, thanks for reading and please forward this to anyone who you think might be interested, it would be much appreciated.