Chris von Csefalvay is a computational epidemiologist/data scientist working at the intersection of AI, epidemiology and public health.

Chris von Csefalvay

The year is 1959. Eisenhower is on his second term, Castro just kicked Batista out of the country and Ray Charles’s Let the Good Times Roll is topping the charts.

The 95% myth

Over at the work blog, I’m discussing what knowledge means for large language models (LLMs), and the ways in which we can leverage this knowledge dividend for better inference.  Read the full post here.

The knowledge dividend of large language models

The awesome thing about language is that, well, we all mostly speak it, to some extent or another. This gives us an immensely powerful tool to manipulate transformational tasks. For the purposes of this post, I consider a transformational task to be essentially anything that takes an input and is largel intended to return some version of the same thing. This is not a very precise definition, but it will have to do for now.

LAIR - Language As Intermediate Representation

As the new Omicron variant of concern of COVID-19 emerged over Thanksgiving week (of all times!), attention has been turning to Paxlovid (PF-07321332). If, like me, you torment yourself by perusing conspiracy theories, you might have heard it referred to as Pfizermectin.

Peace, love, Paxlovid and ‘Pfizermectin’

I used to be on Quora. I’m not anymore. But I used to be. And I wrote some answers. Here are some of my favourites, categorised by field… sort of.

 AI

If I wrote an AI and trained it to score 200 on a standardized IQ test, what could a human of 100 IQ still do better? Read the full answer here. Will true AI come before the ability to completely simulate a human brain? Is there a difference? Read the full answer here.

My favourite Quora answers

Tip  If you’re here for the DOI matching script, the link is here. Quarto is a great tool for reproducible research. It is also a great tool for building websites. After about a decade of Wordpress-based websites, I’ve moved to Quarto primarily for two reasons. First, most of my content is largely static.

Quarto project scripts are awesomeness

There’s a notion in artificial intelligence known as Moravec’s paradox: it’s relatively easy to teach a computer to play chess or checkers at a pretty decent level, but near impossible to teach it something as trivial as bipedal motion. (Hassabis 2017) The sensorimotor tasks that our truly wonderful brains have mastered by our second birthday are much harder to teach a computer than something arguably as ‘complex’ as beating a chess grandmaster.

The hardest AI problem you’ve never heard of.

It appears that in what is clearly a wonderful little PR stunt, a Polish rum company managed to do a Sophia and appoint an ‘AI-driven’ ‘robot’ as its ‘CEO’.  The other guilty party to this pile of steaming bovine excrement is Hanson Robotics, famous for giving us Sophia, the “world’s first robot citizen”. Most of what I’m saying here goes just as well for Sophia.

The skeuomorphic fallacy

Yesterday, I completed a marathon on the SkiErg, with a pretty respectable time (see results here). I know I can do better, and I know I

 will

do better. On the other hand, this was a marathon where I pretty much did everything wrong. And that was sort of an instructive experience in its own right. The title is somewhat misleading. Not much went wrong – the marathon itself was smooth sailing.

How to do a SkiErg marathon entirely the wrong way (but still finish)

There’s something special about language. It is ‘our own’, it is ‘us’, in a profound way, and quite surprisingly, more so than art. I was deeply struck by this when I first saw reactions to large generative language models that created realistic, human-ish prose.

Beyond Broca

In my latest post for the Starschema blog, I discuss the end of dashboards, and what comes next:  Read the full post here.

As we may see: the world after dashboards

There might be valid concerns about the mRNA vaccines. Then there are the clearly insane ones, such as the claims that it involves the injection of a microchip. The argument that mRNA vaccines are a form of ‘gene therapy’ are somewhere in the middle, and that makes responding to it somewhat difficult.

Why the COVID-19 vaccines are not gene therapy

The temple of Asclepios, the Greek god of healing arts and medicine, at Epidaurus was pretty much the ancient Greek world’s equivalent of the Mayo Clinic. It then tells us a lot about the Greek worldview of healing that one of the things the temple complex prominently featured was a theatre. The Greeks believed in drama therapy, but in perhaps a slightly different way of how we think of it today.

The Moral Pulse of the Machine

Every year, the Commandant of the Marine Corps publishes a reading list of books that often only bear on warfighting tangentially at best. The idea behind this is that those entrusted with the lives of servicemembers should have an understanding of the world that goes beyond the profession of arms. In much the same way, I have been advising data scientists to go beyond professional literature.

Five non-data-science books for data scientists

In the middle of the green and pleasant countryside of England lies a small piece of commons –– land, that is, over which a number of people share certain rights. In this hypothetical scenario, it’s six farmers,  to . Each of them has a hypothetical cow, which differs from ordinary cows in that it grazes a fixed unit, and produces a fixed amount of milk (unlike real cows, which vary wildly in their input and output parameters, or so I’m told).

Ransomware attacks as n-person prisoner’s dilemmas

Say you’re busing tables and you’re trying to pass someone in a wheelchair. What do you do? Do you say “excuse me” and wait for them to move? Do you say “excuse me” and then try to pass them? Do you just try to pass them? Do you say nothing and just try to pass them? All of these are, actually, pretty legitimate answers. Now, say you’re a robot.

What I learned from getting bodied by a robot.

Asemantic Induction of Hallucinations in LLMs

There’s a style of visual design I’m inordinately fond of called Raygun Gothic. It’s hard to describe what the hell exactly one needs to be on to enjoy it, but think of it like the aesthetic from the latter Fallout games with a more optimistic outlook on the future. Gibson described it as “the future that never was”, and I think that’s a pretty apt description. We get these competing futures in technology all the time.

Prompt Engineering: The Art of Yesterday

It’s not every day that you find out you have climbed the exalted heights of another discipline. My work is pretty interdisciplinary, but it shocked me, too, that I’m apparently holding forth on neoliberalism and the epistemic question in African universities (archive link):  This, of course, came at some surprise to me, as I have never written anything on the topic.

Stochastic parrots, cap and gown edition

In the first four entries (1 2 3 4) of this sequence, I have focused primarily on what LLMs aren’t, can’t, won’t, wouldn’t and shouldn’t. It’s probably time to conclude this series by that much awaited moment in all stories, where the darkest night finally turns into a glorious dawn, where we finally arrive at the promised land, where we finally get to talk about what LLMs

 could

be. What I see as the most successful potential model of

Teams of Rivals

My recent post on the Starschema blog discusses the need for better data products to tackle future pandemic challenges:  Read the full post here.

Data for the next pandemic

I have probably spent more time looking at Poussin’s

 Dance to the Music of Time

than any other work of art. Sneaking off to the Wallace Collection in London and just looking at the

 Dance

was my comfort activity while living in London – a time that was not exactly devoid of its trials. It’s not, by any measure,

 great

art, insofar as such judgments can be made with any objectivity.

The Lyre of Hephaestus

In a recent paper that has attracted the interest of popular media as well, Fabio Urbina and colleagues examined the use (or rather, the abuse) of computational chemistry models of toxicity for generating toxic compounds and potential chemical agent candidates.(Urbina et al. 2022) Urbina and colleagues conclude that Urbina, Fabio, Filippa Lentzos, Cédric Invernizzi, and Sean Ekins. 2022.

A different shade of grey

I love posts that allow me to merge some of my addictions.

Auto-DOI for Quarto posts via Rogue Scholar

::: {.page-columns .page-full}
In the middle of the green and pleasant countryside of England lies a
small piece of commons ---- land, that is, over which a number of people
share certain rights. In this hypothetical scenario, it's six farmers,
![](https://latex.codecogs.com/png.latex?F_1) to
![](https://latex.codecogs.com/png.latex?F_6). Each of them has a
hypothetical cow, which differs from ordinary cows in that it grazes a
fixed unit, and produces a fixed amount of milk (unlike real cows, which
vary wildly in their input and output parameters, or so I'm told). It is
important to note that the land is a form of common property, meaning it
is held by all of them jointly----or as ye olde Law Latin would put it,
totum tenet et nihil tenet: they all hold the whole together, but no-one
holds any specific part of it.^1^

::: {.no-row-height .column-margin .column-container}
^1^ As opposed to forms of co-ownership where each person owns a defined
fraction of the co-owned property.
:::
:::

If you are one of the farmers, you have a very clear dominant strategy:
get another cow! Even better, get a handful. More cows equal more milk,
and the land can sustain at least six cows.

The problem is, if all participants
![](https://latex.codecogs.com/png.latex?F_%7B1...6%7D) pursue this
individually dominant strategy, the sum of their outcomes will be
inferior. Every player choosing the dominant strategy results in a worse
outcome than it would if they had not chosen the dominant strategy. This
phenomenon is sometimes referred to as the 'tragedy of the commons', but
is in fact a simple result of an n-player prisoner's dilemma.

::: {#raising-the-stakes .section .level1 .page-columns .page-full}
# Raising the stakes

::: {.page-columns .page-full}
Let's raise the stakes a little. Kidnappings happen because a
significant fraction of people pay ransoms. The law on this is a little
weird, and this reflects the general unease with which we regard the
issue.^2^ We might agree that paying ransoms is a bad idea as a matter
of principle, while also feeling rather disinclined to sacrifice a loved
one on the altar of that principle. Personally, I know very few people
who genuinely would not pay a ransom they had the means to pay for the
release of a loved one -- even if as an abstract principle, they agree
that paying ransoms is a bad idea. A lifetime ago, when I studied (and
later taught) moral and political philosophy at Oxford, these questions
were popular dilemmas to make students think about consequentialism.
Now, they are questions that affect every-day lives. Just earlier this
week, much of the Eastern seaboard experienced gas shortages due to a
ransomware attack on the Colonial Pipeline -- over here in Northern
Virginia, most gas stations were out of supply yesterday and expect to
have fuel no sooner than early next week.

::: {.no-row-height .column-margin .column-container}
^2^ The law is outright weird, in that in most countries, paying ransoms
is not illegal per se, but giving money to certain entities is illegal
regardless of the purpose of the money. In that sense the law does not
distinguish between paying a ransom to, say, Boko Haram, who happen to
be on the Treasury's shitlist, or giving them the same amount of money
due to an overwhelming conviction that boko is very haram indeed -- it's
illegal either way. This makes paying a ransom needlessly complicated,
especially as in my modest experience with kidnappers, they rarely
discuss their organisational affiliations at length. With international
jihad in particular being run on the franchise model in this day and
age, it may well turn out that Joe Kidnapper, who held himself out as a
common criminal, was in fact a full franchisee of al-Qa'ida, and
suddenly the FBI is going through your underwear drawer.
:::
:::

::: {.page-columns .page-full}
And that leads us to the ethical problem of ransomware, in particular
paying the ransom (as the maintainers of the Colonial Pipeline did). If
the common response to ransomware would be 'go f------ yourself', making
ransomware would not be worth it. This scenario slightly differs from
kidnapping in that the investment in each ransomware operation is very
significantly smaller than it is for a kidnapping.^3^ Consequently,
adopting the collectively dominant strategy of telling kidnappers what
to stick where on which family member of theirs would only work if
virtually all potential victims would do so (i.e. cooperate). This makes
ransomware a trickier problem than plain, old kidnapping: if only a few
ransom demands are not met, kidnappers are faced not only with the issue
of cleaning up the bodies but also a significant financing shortfall.
Ransomware cybercriminals, on the other hand, can afford a significant
proportion of their victims to 'cooperate' (i.e. not pay), as long as a
small number 'defect'. Thence comes our first important point to
consider: the cost of extortion governs where the deficient equilibrium
point is located.

::: {.no-row-height .column-margin .column-container}
^3^ Kidnapping, I am told, is expensive, and has to be funded in
advance, with no guarantee that it will yield a payout at the end. The
more complicated the operation, the greater the risk. The advent of the
paseo milionario or 'express kidnapping' indicates that there is clearly
a much reduced appetite for the risk involved with kidnapping. It is
then no wonder that ransomware, which lacks many of the expensive and/or
gory parts of kidnapping, is so much more favoured.
:::
:::
:::

::: {#possible-solutions .section .level1}
# Possible solutions

The consequence of low-cost extortion (express kidnapping, ransomware
&c.) is to move the deficient equilibrium point to the far right,
meaning that an unrealistically large percentage of the population must
commit to cooperate before kidnappers start to see their business model
break down. This is, ultimately, the problem known to economists as
free-riding. There are, overall, two major solutions to it, none of them
good.

Forced cooperation. In forced cooperation, a third party (typically, the
state) intervenes and shifts the payoff matrix so as to make cooperation
relatively more profitable. This happens typically by making defection
less profitable, such as would happen if paying ransoms were to be
criminalised. Voluntary cooperation. In voluntary cooperation, the
payoff matrix is unaffected, but players set aside their individually
dominant strategy voluntarily. It's important to note at this stage that
it is perfectly consistent to argue against forced cooperation and in
favour of voluntary cooperation----that is, holding the opinion that
nobody should pay ransoms but the government should not criminalise
paying ransoms is not an inconsistent public policy position.
:::

::: {#the-problem-with-forced-cooperation .section .level1}
# The problem with forced cooperation

Let's face it -- I'm an optimist, but even I'm not optimistic enough to
think people will voluntarily cooperate to the extent required by the
extremely low cost of ransomware attacks. The example of kidnappings,
where ransoms are almost always paid (even where this is strictly
speaking illegal), shows that such cooperation is unlikely to arise
voluntarily.

The answer to that, many would argue, lies in enforcing cooperation by
altering the payoff matrix. [Robert Knake makes this point in an article
from
2016](https://www.cfr.org/blog/paying-ransom-ransomware-should-be-illegal):

> What I will argue is that when looking at a public policy problem, the
> best place to create liability is where it will have the desired
> impact. If the goal is to stop ransomware attacks, raising the costs
> of paying ransoms beyond what the criminals are demanding is the best
> way to do that. Those costs could come in the form of civil fines or
> misdemeanor charges. For most American companies and most individuals,
> simply knowing that paying a ransom would violate the law might be
> enough to dissuade them. If enough victims are persuaded to forgo
> payment and accept the consequences, there will be fewer future
> victims.

Regrettably, this argument is several kinds of wrong and at least two
kinds of immoral. It is wrong because the sad ubiquity of ransomware
attacks and their relatively private-synallagmatic nature (unlike
kidnappings, which are widely publicised) means that enforcing such
legislation would be quite difficult in practice. It is also wrong
because the impact of "civil fines or misdemeanor charges" is far from
clear. Short of exorbitant fines (which would face legal challenges for
a range of reasons), these costs would just be factored into the 'cost
of doing business' the way many companies already accept various
unsavoury practices that are part and parcel of economic activity in
some parts of the world. And it is fundamentally wrong in failing to
understand the logic behind ransomware. The reason why ransomware makes
sense is because data may be unique and irreplaceable (it shouldn't be,
and having well-designed IT policies with TTPs that encourage data
replication and encryption are the most effective tools against
ransomware attacks, but that's a different story). Compared to the
consequences of losing years of research, experiencing months of service
interruption or leaking the medical records of thousands of patients,
any fine the government can levy pales in comparison. You cannot fine a
complex problem out of existence -- and fines are in any case
economically flawed deterrents, reducing the available funds for
compliance. Fines work to reduce the economic profitability or viability
of a course of action -- they do not work in preventing the
circumstances from arising that open up the choices including that
course of action.
:::

::: {#a-better-model-of-state-action .section .level1}
# A better model of state action

If, thus, the coercive model of state action in response to ransomware
is a ubiquitous failure, the question is what role the state can play in
this matter. The alternative model that I support is one of cooperation
rather than coercion. In the cooperative model, state actors assist
enterprises within their territory in changing the payoff matrix by
increasing the costs of successful ransomware attacks. Ransomware works
because it's so cost-efficient: even if only one in a thousand victims
pay up, the monies paid out are sufficient for a generous profit. Up
until here, I'm in agreement with the coercive model----the only way to
eliminate ransomware is to make it economically less viable. The
difference is how that is accomplished.

In the cooperative model, the state invests in hardening IT systems
against ransomware attacks. It stands to reason that if a large number
of such attacks do not succeed, the entire enterprise becomes
unprofitable. From the blackmailer's perspective, the results are pretty
much the same, i.e. vastly reduced income. However, from what H.L.A.
Hart would have called the 'internal point of view', i.e. the
perspective of the community and the individual, there is a crucial
difference: cooperation forgoes the moral horror of punishing the victim
in favour of preventing individuals from becoming victims. In the end,
the "best place to create liability" is not where it will have some
fictitious "impact", but where it is most appropriate. Re-victimization
is not only morally abhorrent, it is also ineffective: a company that
sustained a loss from a massive ransomware attack may be less capable of
upgrading security to withstand such attacks in the future, whereas a
company that received support to shore up their defences will have a
strong (indeed, contractual) incentive to do so.

In the end, how we allocate losses says a lot about what we as a society
are like. The choice, then, as far as ransomware is concerned, must lie
in collaboration over coercion. Government has an opportunity to be a
force for good in this field, if only it can forego the empty moralisms
that focus on the victim's blameworthiness. Certainly ill-prepared
companies are more likely to be struck by ransomware. Blaming it all on
the victim, however, and treating ransoms as illegal payments merely
allows governments that have for too long dropped the ball on
cybersecurity preparedness to further postpone their duties and deflect
blame.
:::

::: {#quarto-appendix .default}
::: {.section .quarto-appendix-contents}
## Citation {#citation .anchored .quarto-appendix-heading}

<div>

::: {.quarto-appendix-secondary-label}
BibTeX citation:
:::

``` {.sourceCode .code-with-copy .quarto-appendix-bibtex}
@misc{csefalvay2021,
 author = {Chris von Csefalvay},
 title = {Ransomware Attacks as n-Person Prisoner’s Dilemmas},
 date = {2021-05-14},
 url = {https://chrisvoncsefalvay.com/posts/ransomware-prisoners-dilemma},
 doi = {10.59350/r74zv-gsd67},
 langid = {en-GB}
}
```

::: {.quarto-appendix-secondary-label}
For attribution, please cite this work as:
:::

::: {#ref-csefalvay2021 .csl-entry .quarto-appendix-citeas}
Chris von Csefalvay. 2021. "Ransomware Attacks as n-Person Prisoner's
Dilemmas." <https://doi.org/10.59350/r74zv-gsd67>.
:::

</div>
:::
:::