40 Matching Annotations
  1. May 2022
    1. we assume that the agent has N primitive cause–effect relationships that each assign probabilities to state, action and next-state transitions, 𝜙𝑖:𝒮×𝒜×𝒮→[0,1]ϕi:S×A×S→[0,1]{\phi }_{i}:{\mathscr{S}}\times {\mathscr{A}}\times {\mathscr{S}}\to [0,1], 𝑖=1,...,𝑁i=1,...,Ni=1,...,N. Each 𝜙𝑖(𝑠′|𝑠,𝑎)ϕi(s′|s,a){\phi }_{i}(s{\prime} |s,a) is a potential function representing, for example, the local effect of colliding with the couch or pulling on the coffee table. Then a construal is a subset of these primitive cause–effect relationships, 𝑐⊆{𝜙1,...,𝜙𝑁}c⊆{ϕ1,...,ϕN}c\subseteq \{{\phi }_{1}\mathrm{,...,}\,{\phi }_{N}\}, that produces a task construal, 𝒯𝑐Tc{{\mathscr{T}}}_{c}, with the following construed transition function:𝑃𝑐(𝑠′|𝑠,𝑎)∝∏𝜙𝑖∈𝑐𝜙𝑖(𝑠′|𝑠,𝑎).Pc(s′|s,a)∝∏ϕi∈cϕi(s′|s,a).{P}_{c}(s{\prime} |s,a)\propto \prod _{{\phi }_{i}\in c}{\phi }_{i}(s{\prime} |s,a). (1)

      formal definition of a task construal in terms of a (sub)set of state-action-next-state transitions with assoc. est. probabilities.

    1. we should be more explicit about what exactly they can help, how they can help, and whom they can help

      Now we're talking :)

    2. The ability to create singular written works is mostly impervious to education and technical supplementation; it is overwhelmingly what we used to call gifted or God-given and today call either genetic or inspired.

      CLM: The majority of the variance between individuals in terms of being able to produce novel and nontrivial insights is determined by genetic factors that are out of their control.

      Hmmm the genetic argument is now extended to cover not just "being organized", but "being able to produce novel and nontrivial insights". Here I start to get more skeptical. What about skill acquisition / learning for example? Or collaboration with transformative peers? Or... mathematical notation? Or abstractions and algorithms? Why learn computational modeling, or engage in practices of field notes and memo-ing and iterative passes over data (for qualitative theory-building)?

    3. The most important thing about writing is discovering novel and non-trivial truths, and determining which of your truths is most important—then imposing order, hierarchy, and linearity—through judgment, decisiveness, and will.

      More on defining the target.

      Before, it was brute, linear, willfulness.

      Now it adds: - novel and nontrivial truths (I like this! very much in the spirit of the consensual definition of creativity: new and useful/impactful) - order, hierarchy, linearity (there is it again: the collapsing of the distribution) - judgment - decisiveness - will (again)

    4. The concept of the Knowledge Graph deserves the classification of bullshit because its allure derives primarily from the false impression that it can mechanistically deliver—or substitute for—the brute, linear willfulness that defines all non-trivial writing

      I think this is where the Knowledge Graph ideology is expounded upon?

      But now there seems to be a contrast with three other ideas that I'm uncertain of: - Brute (? not sure what this means. manual? embodied? not automated? slow?) - Linear (linearized? I could see this. a collapse of the distribution over structures/models into a single structure for consideration) - Willfullness (? not sure what this means. intentional?)

    5. you are unlikely to author anything very profound or forceful, so long as you are possessed by the Knowledge Graph ideology

      CLM: If you believe that merely linking documents without active attempts to add structure is sufficient to generate insight, you are unlikely to generate profound or forceful insights.

    6. most of the variance between individuals is genetic and relatively invulnerable to intervention

      CLM: The majority of the variance between individuals in terms of attaining a state of order in their information/data is determined by genetic factors that are out of their control.

  2. Feb 2022
    1. The researchers’ high-speed videos showed that the creatures, who jump at speeds as high as 8.7 miles per hour, cocked their back legs in a jumping position, then pushed forward, with each moving within 30 microseconds (that’s 30 millionths of a second) of the other. The finely toothed gears in their legs allow this to happen. “In Issus, the skeleton is used to solve a complex problem that the brain and nervous system can’t,” Burrows said in a press statement.

      [[PTN]] - Design systems, not tools

    1. A whole body of work already exists that explores related ideas under the label of “situated cognition” (e.g., Clancey, 1997, 2009; Kirsh, 1991, 2009; Kirshner & Whitson, 1997; Lave, 1988; Robbins & Aydede, 2009a; Suchman, 1987). What exactly is meant by “situated cognition,” however, is a matter of debate (Robbins & Aydede, 2009b; Roth & Jornet, 2013; Wilson, 2002). At least on some formulations, situated cognition and embodied cognitive science are not simply similar and compatible but they actually overlap conceptually and historically (see, e.g., Gallagher, 2009). On other formulations, “situated cognition” is understood broadly enough as to encompass even work in science and technology studies in the vicinity of SSK (see brief discussion in Solomon, 2007). While these convergences and overlaps are suggestive, there are other formulations, still, where “situated cognition” is interpreted in a more traditional cognitivist vein as being roughly equivalent to the first sense of “embodied cognition” reviewed in Sect. 1 (e.g., Wilson, 2002).

      entry points into situated cognition from POV of embodied cognition; situated cognition is still relatively broad tend, though the threads associated with it often tend to be intertwined with embodied cognition

  3. Nov 2021
    1. The alternative is not between standardization and chaos, but a potential third way is designing infrastructures that allow the diversity of approaches, tools, and techniques to be expressed in a common framework or protocol along with the community infrastructure to allow the continual negotiation of their relationship.

      marking this here while i'm thinking about it: this is the philosophy of project cambria: https://www.inkandswitch.com/cambria/

    2. Calls for standardization without careful consideration and incorporation of existing practice have a familiar cycle: devise a standard that will solve everything, implement it, wonder why people aren’t using it, funding and energy dissipiates, rinse, repeat

      obligatory xkcd reference!! https://xkcd.com/927/

    3. Infrastructuring

      first use as a verb!

    4. structured by

      really appreciate this up front! one fun way for folks to contribute is to say something like "yes, this is what it's like for me too in [x domain]"

    5. the promise of digital technology for science

      again, what is this promise? faster science? more innovative science? less error prone science? more reliable use for public? more open?

      eager to see if this is unpacked crisply later on. otherwise, worried that the goal is too diffuse to then assess if "decentralization" will in fact help

    6. There is no guarantee that it “optimally” satisfies some set of needs for the benefit of the greatest number of people

      key insight from infrastructure studies: infrastructure always serves some, but not others

    7. The yawning mismatch between our ambitions of what digital technology should allow us to do and the state of digital infrastructure

      so far, more hinted at, and told, rather than shown explicitly. need a few concrete examples of what we should be able to do but can't.

      some examples:

      • notice a claim in a paper that you can in principle check the stats of; should be able to easily get the data and just... do it!
      • read an interesting hypothesis: should be able to check "has anyone followed up on this hypothesis, and if so what is the current consensus on it?"
      • get a list (not necessarily comprehensive!!) of open problems in some domain
    8. deinfrastructured

      curious about this word choice: seems active (undoing infrastructure), and more heavily implies a subject: who is doing the deinfrastructuring?

    9. a robust archive is not only a matter of building a database with some frontend, but by building a community

      This section really really resonates. Not to add to an already amazingly expansively cited piece, but this exact issue of designing sociotechnical systems, not just technical ones, and understanding that infrastructure is as much social as it is technical, is a key insight that is so easy to overlook when you're caught in the weeds of building this new infrastructure (technically)!

      if you don't build the community with it, it just dies. but obv you can't just build the community: you also need th new tech! the two need to go hand in hand. this is very hard!

      an entry point for this: http://pne.people.si.umich.edu/PDF/Edwards_etal_2013_Knowledge_Infrastructures.pdf

  4. May 2021
    1. If she goes offline, we may still be able to get the same photo from Grace or another peer.

      this is interesting: so we actually do want some amount of duplication and redundancy. i wonder if it is enough to assume that some of this is going to happen, or whether we want to also incentivize or make it happen in some way.

    2. hashes can prevent malicious actors from deceiving us about the content of files

      a very specific sense of trust. i wonder how this might be generalized, if at all. this is a form of proof.

    3. trust the information that's shared

      in a specific sense, i think, namely that the "location" actually contains the thing it says it has.

    4. Cryptographic hashes can be derived from the content of the data itself, meaning that anyone using the same algorithm on the same data will arrive at the same hash. If Ada and Grace are both using the same decentralized web protocol, such as IPFS, to share the exact same photo of a kitten, both images will have exactly the same hash. By comparing those hashes and confirming that they're the same, we can guarantee that every single pixel of those two photos is identical.

      I wonder whether this is all or nothing, or whether there is any room for fuzzy / approximate matching. Could hashes derived from "similar" data (e.g., picture of the same kitten, but from a slightly different angle) be "similar" in some sense?

  5. Apr 2021
    1. Last year the futurist Ayesha Khanna even described smart contact lenses that could make homeless people disappear from view

      Worth noting that this description was in the context of a warning against the temptation of "deletive reality": https://www.ayeshakhanna.com/blog/welcome-to-the-hybrid-age

  6. Jan 2021
    1. building theories of capacities is a daunting task. The space of possible theories is, prima facie, at least as large as the space of effects: For any finite set of (naturalistic or controlled) observations about capacities, there exist (in principle) infinitely many theories consistent with those observations. However, we argue that theories may be built by following a constructive strategy and meeting key plausibility constraints to rule out from the start theories that are least likely to be true: We refer to this as the theoretical cycle

      effective synthesis is hard.

      here they offer strategies for dealing with this difficulty. note here that plausibility constraints are empirical!

    1. PredicTER is a tool for estimating how long a review will take to complete. The tool calculates the time requirements for various tasks involved in reviewing evidence, from planning and coordination to quantitative synthesis and reporting.  The tool contains default values provided by an assessment of 5 years of systematic reviews and maps published by the Collaboration for Environmental Evidence, but users can tailor the tool based on what they know about their own subject.

      estimates based on 5 years of experience with systematic reviews/maps at Collaboration for Environmental Evidence (but details of the estimation approach aren't known here)

  7. Dec 2020
    1. Clusters of infections within families living in Bnei Brak were identified and investigated. The parents were asked regarding the first case of the infection in the family and regarding the pre-sumed source of the infection.In addition, household members underwent polymerase chain reaction (PCR) testing whether they were symptomatic or not

      testing regime: exhaustive (whether symptomatic or not)

    2. Thirteen family clusters were investigated; all families reside in the city of Bnei Brak.

      N=13 families

    3. Mayenei Hayeshuah Medical center is located in the city of Bnei Brak, Central Israel.Bnei Brak is a “young” city. Children of 0–19 years of age comprise almost 50% of its 200,000 population, and the average number of children in a family is 4.57

      setting: central Israel

    4. SARS-CoV-2 positive PCR was documented in the different age groups as follows:1. In 21 of 36 adults (>18 years) (58.3%).2. In 13 of 40 children, 5–17 years (32.5%), (P = 0.037 for the difference between group 1 and group 2, risk ratio: 0.61, 95% confidence interval [CI]: 0.39–0.96).3. In 2 of 18 children, 0–4 (<5) years of age (11.8%), (P < 0.002 for the difference between group 1 and group 3, risk ratio: 0.47, 95% CI: 0.30–0.71)

      main result: children ~half as likely to get infected given equivalent exposure in same household

  8. Nov 2020
    1. young children in particular are approximately half as likely as adults to acquire the infection given equivalent exposure.

      Current belief on children's susceptibility to COVID

  9. Sep 2020
    1. While so much about the virus and how it operates remains unclear, sub-Saharan Africa so far has dodged a deadly wave of coronavirus cases. Many factors have contributed to this. A number of West African nations already had a pandemic response infrastructure in place from the Ebola outbreak of late 2013 to 2016. Just six years ago, Liberia lost nearly 5,000 people to Ebola. At the beginning of this year, Liberia began screening for covid-19 at airports. Travelers coming in from countries with more than 200 cases were quarantined. To date, Liberia, a country of some 5 million, has 1,335 cases and around 82 deaths.

      example-of the importance of [[infrastructure]]

  10. Jul 2020
    1. in the unexposed patients, 40 to 60% had CD4+ cells that already respond to the new coronavirus

      cross-reactivity of cd4+ t-cell responses from previous coronaviruses?

    2. 100% of the exposed patients had CD4+ cells that responded to all three of these. There were also significant CD4+ responses to other viral proteins: nsp3, nsp4, ORF3s, ORF7a, nsp12 and ORF8

      good cd4+ t-cell responses in COVID patients

    1. The difference between detecting COVID antibodies and detecting T cells with an effective anti-COVID response is kind of like the difference between detecting guns and detecting assassins. One of these is relatively straightforward; the other one requires extensive investigation.

      Good analogy to explain why it's hard to detect (and therefore study) T cells

    1. just because antibodies in plasma wane over time doesn’t mean that individual’s aren’t protected right, Derek? What’s really important is Memory B-Cells, which can rapidly reproduce antibodies if the antigen is seen again? Have peoeple tested individuals with SARS-CoV-2 for Memory B-Cell formations?

      memory B cells are important, even if antibodies are gone, can produce antibodies again quite quickly if needed

    2. BUT there is a hidden assumption, an artefact of the initial model’s simplified reality assumptions, the Susceptible and Recovered must be randomly spread out. This randomisation is the mechanism that extinguishes the virus (in the model !). In observable reality, these two populations are not perfectly intermixed.

      herd immunity works differently for vaccination scenarios vs. population infection, due to distribution of susceptible vs. recovered (near random / equitable with vaccination, systematically nonrandom with population infection)

    3. antigen specific memory B cells themselves–by definition fairly low abundance and difficult to distinguish from standard B cells using assays robust enough to handle large numbers of samples.

      memory B cells are important to immune responses, but can be hard to detect reliably at scale

    4. There are hundreds of viruses that can cause common cold symptons. Endemic coronaviruses, enteroviruses, rhinoviruses, coxsackie viruses, parainfluenza – probably a bunch that are unknown at this point.

      common cold is caused by many other kinds of viruses, not just coronaviruses

    5. Highly conserved sections of the virus are also often highly protected. Look at the effort that’s been put into the universal influenza vaccine. The parts that are stable enough to dependably target are also usually hidden, and only revealed briefly when the virus activates.

      reasons to doubt that targeting aspects of viruses that are "conserved" (i.e., remain stable across mutations / strains) is a viable strategy for a vaccine

  11. Feb 2020
    1. “I can imagine a scenario where this becomes a fifth endemic human coronavirus,” said Stephen Morse of Columbia University’s Mailman School of Public Health, an epidemiologist and expert on emerging infectious diseases. “We don’t pay much attention to them because they’re so mundane,” especially compared to seasonal flu.

      Can imagine != likely == possible