Collective action+altruism, ethnic recognition regimes+violence, and the nature of agency

• Altruism, Ethnicity, Nationalism – and collective action
  • How does doing collective action effect group identities and foster altruistic collective action (cyclically)?
• When does recognizing ethnicity decrease or inflame violence in plural societies? Diversity Violence and Recognition – Elizabeth King and Samii, (King and Samii 2020))
  • King and Samii are present a theory and study on recognition of ethnic identities within formal institutions. The incorporation of identity is termed the ‘recognition regime’, versus non-recognition regime. The type of recognition within these can be thick (accompanied by diverse and rich benefits) or thin (merely symbolic).
  • Non-recognition include laïcité in France and/or banning of explicitly ethnic parties, such as in Turkey. Explicit ethnic power-sharing, such as in Burundi or Switzerland.
    • Note: The Design Space (Nadel and Pritchett 2016) for these policy designs is high dimensional. This presents a challenge for theories and empirics.
  • These regimes not only incorporate ethnic identities, but also shape them through categorization and institutions implemented, made (more/less) legible, and potentially reinforced by the state.
  • Liberal and modernization theory supposes identities should become less salient for conflict with development. However, there is a puzzle in which ethnic identities, and often intertwined nationalistic identities (see Benedict Anderson’s Imagined Communities), have become only drivers of conflict (citing (Horowitz 1985)).
  • Political entrepreneurs often use ethnic cleavages to drive conflict. In these cases, the use of ethnic conflict hardens ethnic cleavages (“it is often conflict itself that further shapes and hardens group identities”).
  • King and Samii point out a few starting premises causal mechanics with opposite intuitions for how recognition effects violence:
    1. mobilization effects of recognition mobilize people around ethnicity in political competition (which can become violent conflict). In this case, this hardens the alignment of political competition on ethnic lines.
    2. assurance effects of recognition show the acceptance of plural ethnicity in a nation. Especially adopting recognition institutions meant to promote equity, can ameliorate ethnic violence by decreasing grievances.
  • King and Samii’s solution is that the ‘leadership regime’ is positioned differently across contexts vis-a-vis the recognition regime.
    • If leaders are from an ethnic plurality, then it is both politically safer to adopt a recognition regime and more effective at reducing conflict (both mechanisms work in their favor). This is because the mobilization effect or recognition benefits them due to a larger population to mobilize
    • But, if leaders are from an ethnic minority, these two mechanisms class (mobilization works against them, while assurance works for them) it is politically riskier and potentially less effective. The mobilization effect draws on the smaller percentage of their co-ethnics, thus benefiting other aspiring political leaders (of the larger ethnicity(ies))
      • The reason recognition doesn’t work for minority leaders is second-order. Due to the negative mobilizatio effect, they are likely to take counter-measures to avoid losing power (relative to before the recognition regime).
  • Their cross-national regressions support this (60% of post-conflict constitutions for majority-led governments vs. 24 for minority-led include recognition regimes).
• Benedict Anderson’s Imagined Communities – to read next
• Self Efficacy extends but does not contradict rational choice theory:
  • Self-efficacy is an individual’s belief in his or her capacity to execute behaviors necessary to produce specific performance attainments (Bandura 1997)
  • Self-efficacy does not necessary dispute rationalist theories, as it can be seen as a coefficient on the expected value. If we believe our actions are successful with a lower probability, we are less likely to initiate them.
  • This is a mediating step in rationalist calculations. Not only do we need to calculate the expected payoffs of different actions, we also need to incorporate our likelihood of successfully achieving those actions conditional on deciding to attempt them. Our efficiacy is a function of our capacity in a certain domain and under certain conditions. ( \(\text{efficacy} = P(\text{success | attempt})\)). This \(\text{efficacy} = f(\text{percieved capacity to complete X task})\).
• Agency and democracy: What does it mean to have agency? My intuition is that agency is core to democracy. Both as an outcome (democracies should safeguard agency), but also as an input: agency (the ability to take political action as Arendt would say) at a group level is a core characteristic of democracy.
• This is a variation of the argument that democracy is the capacity to collective problem solve, as in Warren’s Problem-based conception of democracy’ (Warren 2017). There are a few
• Feedback is all you need… for agency to emerge: Gordon from squishy computer discusses agency as a spectrum, specifically the breadth of action one is able to take in response to feedback (this is in reference to AI systems). He argues that a thermostat is a (very limited) agent. The ability to update based on goal orientation.
• • Agency as a strategic game: Cate Hall discusses it as a strategy and approach to life (e.g. increasing surface area for luck, and seeking feedback), but does not delve into the term directly.
• Agency as Freedom Amartya Sen discusses agency as Freedom. I believe he means freedom to both make and pursue goals. However, I am unaware of whether he discusses where people get a sense of agency from. Arguably, there is an innate motivating engine towards creating. Action, as Arendt may call it in Human Condition_Arendt|The Human Condition. This force is political in her telling, but it is also the notion that impulse to take action outside of oneself.
• Goal vs. Action Agency, and understanding: The literature on agency seems to fail to differentiate between Goal Agency and Action Agency. Deterministic systems, by definition, have no goal agency. My first impulse is that this is at least heuristically true even if you introduce stochasticity into a deterministic system. LLMs are excellent examples of entities which seem to have agency due to stochasticity.
  • Agency and AI: However, there is a (largely philosophical) debate about whether LLMs actually understand concepts, and therefore can meaningfully generate goals, or if they are ‘stochastic parrots’ which mimic (but do not have) human goal setting. These may not be distinguishable at the limit (as Altman famously quipped, we may be stochastic parrots “i am [note: maybe] a stochastic parrot, and so [note: maybe] r u”’). The distinction likely comes down to experiences of consciousness, and whether you need to experience understanding (computers couldn’t provably do) or simply show symptoms of it (computers have provably done, espcially with LLms). Of course, there are many kinds of understanding, such as tacit, explicit, feeling of, etc.. We cannot prove that computers experience any of these without consciousness.
• Goals require an implicit questions, which LLMs cannot ask: All goals are responses to the question of ‘what should be done to solve… X’. This is often implicit, but this ability to ask and select between ‘answers’ to a question is a fundamental aspect of agency. In this, many agents which has action agency, do not have goal agency, because they cannot select their own questions. For instance, a thermostat has responsiveness (action agency). But it cannot ask what a good temperature for a house would be. Because this requires answering positing the question “how should the house feel?”. Both asking and answering this question are core to agency. Asking the question supposes that the answer is meaningful to something. That something is the agent (the proposer). But, critically, selecting between multiple answers to the question is the second half of what agency is. Because, there is no ‘correct’ answer without reference to an agent. To go back to the example of the thermostat, a human agent decides they want the temperature either extremely cold to keep meat frozen, or even colder to carry out a plasma experiment. Alternately, for tropical plants, a thermostat may be tasked to keep a room at a balmy > 25°C. This goes back to determinism. If an agent is entirely deterministic, then it has no agency, even if it seems to have agency due to a large and ‘black box’ process for generating and selecting between options. At its limit, we cannot prove that humans have agency. However, we perceive them as having agency because we could never predict all human action mechanistically.
• Other:
  • SSH Keys
    • These are cryptographic keys but functionally passwords meant to stop ‘man in the middle’ attacks.
    • Establishing an SSH key for my github locally means I do not have to enter my credentials as often from the CLI.
    • These are stored in an .ssh directory, typically in the ‘home’ folder.
    • These have associated passphrases (which are shorter, user generated passwords) which further verify the use.
    • The private ‘identity key’ is a very long key stored in ./.ssh/ssh_id_rsa. RSA stands for “Rivest-Shamir-Adleman cryptosystem”. RSA is based on the computing difficulty of factoring very large integers.
    • The public key is stored in .ssh/id_rsa.pub. This is used to authenticate the private key.
    • Finally, ‘known hosts’ can be put in the ‘known_hosts’ file under ./.ssh/known_hosts. Known hosts stores the keys of servers you’ve connected to in the past. For instance, I can add github’s to my known_host file and it will not prompt me to verify github’s RSA ‘fingerprint’. The fingerprint is a hash of github’s full public RSA (i.e. much shorter). e.g. github’s fingerprint is: SHA256:+DiY3wvvV6TuJJhbpZisF/zLDA0zPMSvHdkr4UvCOqU.
      • NB: Hashes are algorithmically generated versions of data. These are irreversible and small changes in the original will change the hash. They, therefore, can represent a unique version of data which is unperturbed. The same algorithm run agains the data, if tampered with, would produce a verifiably different hash