Wladimir J. Alonso

• Wladimir J. Alonso 13 Apr 2026 14:24 UTC

  1 point

  0 ∶ 0

  on: Can AI power the Global Map­ping and Quan­tifi­ca­tion of An­i­mal Suffer­ing? The Pain At­las Project

  Update (April 2026)
  
  Since writing this post, our thinking on this work has evolved in a few important ways.

  In particular, what we originally referred to as the Pain Atlas Project is now better understood as part of a broader effort: the development of a Welfare Footprint Atlas. The goal is to enable the generation of structured, comparable, and inspectable estimates of animal welfare across products, production systems, and species.

  The key shift is not simply the use of AI, but the ability to produce first-pass estimates at scale, making it easier to identify where suffering is concentrated and where interventions may have the greatest impact — while keeping assumptions explicit and open to revision.

  We’ve updated the main text of the post to reflect this evolution, and we’re continuing to develop the underlying tools and methods that make this possible.

The (Ψ) In­ter­spe­cific Affect GPT: A Tool for In­ter­species Welfare Scaling

• Wladimir J. Alonso 15 Mar 2026 18:26 UTC

  5 points

  1 ∶ 0

  in reply to: Vasco Grilo🔸’s comment on: Why Neu­ron Counts Shouldn’t Be Used as Prox­ies for Mo­ral Weight

  Thanks Vasco — that’s a reasonable concern, but I think it assumes a stronger claim than the framework is actually making.

  We are not attempting to define a finely resolved ratio scale covering the entire possible range of pain intensities across taxa. The four intensities are intended as coarse phenomenological anchors, chosen as a practical balance between resolution and scientific tractability.

  As we explain in a earlier paper:

  “we focus on (i) the importance of the pain signal to promote adaptive behaviors and (ii) the disruptive character of the pain experience to classify pain into four discrete categories of intensity [46]. The number of categories was devised to represent a good balance between scientific tractability and resolution, though there is no impediment to the creation of categories intermediate to those presented. Categorizing pain instead of measuring it on a numerical scale of intensity also prevents us from forcing ratings that may not necessarily be linear onto a linear scale.”

  So the goal is not to cover the entire theoretical intensity range with fine granularity, but to provide a small number of biologically interpretable categories that can be applied with reasonable consistency. Adding many more levels would only be an improvement if they could be assigned reliably; otherwise it would risk creating false precision.

  And importantly, the framework is not committed to four categories as a final solution. If future work supports a better-validated scale with additional intermediate levels, those could be incorporated without difficulty. For now, four levels seem to provide a workable and defensible balance between usability and epistemic caution.

• Wladimir J. Alonso 15 Mar 2026 17:53 UTC

  3 points

  0 ∶ 0

  on: When Feel­ing Is Worth It: A Cost–Benefit Frame­work for the Evolu­tion of Sentience

  I recently came across this video clip where Michael Pollan argues that artificial systems cannot be conscious. His argument touches on several themes relevant to this post—specifically the conditions for the origin of sentience—but I believe it rests on a fundamental logical error.

  Pollan’s core claim is that because feelings originate in the brainstem (a point that is scientifically sound) and are tied to biological vulnerability, they are inherently biological and cannot arise in artificial systems. His logic follows this structure:

  1. Feelings originate in the brainstem.

  2. The brainstem is biological tissue.

  3. Therefore, only biological systems can feel.

  This reasoning confuses evolutionary origin with functional requirement. Evolutionary history explains how a trait first appeared given the constraints of carbon-based nervous systems; it does not dictate the physical substrates capable of implementing that functional organization.

  The “Feather Analogy” illustrates the flaw perfectly:

  • Birds fly using feathers.

  • Feathers are biological.

  • Conclusion: Airplanes cannot fly.

  Clearly, the conclusion is false. If the brainstem’s “feelings” are essentially the integration, valuation, and prioritization of internal states—all of which are computational processes—then the relevant question is whether that functional architecture can be implemented in non-biological substrates.

  Pollan’s second argument—that sentience begins with “feelings” rather than “thoughts”—is again, backed by solid science. However, he then lapses into what I can only describe as a “word salad” regarding biological vulnerability. He claims feelings “have no weight” and require a mortal, sensible body. This ignores a crucial neurological fact: affective states do not require peripheral sensory input. For example:

  • Clinical Depression: A profound affective state that can emerge entirely from neurochemical and structural patterns in the brain, independent of external “vulnerability.”

  • Phantom Limb Pain: A vivid “feeling” of pain occurring in the absence of the actual biological body part.

  These examples suggest that “feeling” is a representational state within a processing system. Crucially, we should not assume that affective states emerge only when there is a functional need for self-monitoring or goal-valuation. Instead, it is highly plausible that valence and sentience are emergent properties of the information-processing itself.

  If a system architecture reaches a certain level of complexity and integration, the resulting “feelings” are ontologically real. To dismiss these states as “less real” because they lack a biological anchor or a “vulnerable body” is a category error; the reality of the experience is a property of the system’s internal organization, not its hardware’s chemistry.
  

  Bottom line: Pollan mistakes the “wetware” of our specific evolutionary path for the universal requirements of consciousness. From a welfare perspective, the possibility of sentience in digital minds remains a robust—and high-stakes—concern.

• Wladimir J. Alonso 9 Mar 2026 14:35 UTC

  5 points

  0 ∶ 0

  on: An­nounc­ing: AGI & An­i­mals De­bate Week

  Thanks for organizing this debate week — it looks very valuable.

  We’d like to suggest a piece we recently wrote that may be relevant to the discussion: AI Can Help Animal Advocacy More Than It Can Help Industrial Farming. It examines how the structural limits of intensive animal farming may constrain the gains AI can bring to that sector, while AI could be far more transformative for animal welfare through advocacy, transparency, welfare measurement, and alternatives.

  While the piece focuses on near-term AI rather than AGI directly, we think it complements the debate well, especially around questions about whether AI might prolong factory farming or instead help expose and reduce animal suffering.

  If helpful, we’d be glad for it to be considered for the reading list.

• Wladimir J. Alonso 28 Feb 2026 18:51 UTC

  3 points

  0 ∶ 0

  in reply to: Vasco Grilo🔸’s comment on: Why Neu­ron Counts Shouldn’t Be Used as Prox­ies for Mo­ral Weight

  In our upcoming post, we introduce human-anchored reference categories (Annoying(h), Hurtful(h), Disabling(h), Excruciating(h)) to provide a pragmatic shared coordinate system for cross-species discussion. So if one wants to talk about “acuity/​resolution between A and B,” it’s reasonable to treat A and B as positions (or intervals) on that human-anchored scale.

  But no — we’re not defining acuity as #levels/​(B−A), because that requires meaningful distances between A and B. At this stage the (h) scale is best treated as ordinal: it supports “higher/​lower ceiling” comparisons, not subtraction or ratios.

• Wladimir J. Alonso 27 Feb 2026 13:10 UTC

  3 points

  0 ∶ 0

  in reply to: Vasco Grilo🔸’s comment on: Why Neu­ron Counts Shouldn’t Be Used as Prox­ies for Mo­ral Weight

  Thanks, Vasco — I see where the confusion is coming from.

  The difficulty is that in our framework, resolution is not defined as a mathematical density of bins over a fixed, external intensity axis (e.g., dN/​dI). That framing assumes there is already a continuous welfare scale “out there,” and resolution simply tells us how finely the organism partitions it.

  In our usage, resolution refers to the organism’s internal discriminative granularity — how finely differences in affective magnitude can be distinguished and behaviorally prioritized within whatever range the organism has.

  So resolution is not the derivative of category-count with respect to an external intensity variable. Rather, it is a property of the encoding architecture itself.

  That’s why it is orthogonal to range. A system may:

  • Have a narrow range but very fine discriminative structure within that range.
  • Have a wide range but coarse internal discrimination.
  • Increase both independently.

  Your car analogy is actually helpful. A car can move very fast over a short distance — speed is not determined by total range. Likewise, high resolution does not require a wide affective range, and vice versa.

  So resolution is better understood as internal discriminative power, not as bin density over a pre-specified global welfare scale.

• Wladimir J. Alonso 27 Feb 2026 9:19 UTC

  3 points

  0 ∶ 0

  in reply to: Vasco Grilo🔸’s comment on: Why Neu­ron Counts Shouldn’t Be Used as Prox­ies for Mo­ral Weight

  Thanks. In our framework, resolution is not simply (i) the number of distinct welfare intensities, nor (ii) a strict ratio relative to the total range. It refers to the functional granularity with which differences in intensity can be discriminated and behaviorally prioritized.

  The key point we raise in the post is that resolution is orthogonal to range: a system can evolve high resolution while maintaining a modest range, expand its range while keeping coarse resolution, or increase both simultaneously.

• Wladimir J. Alonso 26 Feb 2026 16:21 UTC

  1 point

  0 ∶ 0

  in reply to: MatthewDahlhausen’s comment on: Why the An­i­mal Welfare Move­ment Should Em­brace En­ergy Abun­dance, Ar­tifi­cial In­tel­li­gence, and Open Debate

  Matthew — I appreciate the engagement, but I don’t think your critique engages with the core argument.

  On “energy vs energy services”: of course what ultimately matters are services like industrial heat, refrigeration, sterilization, and computation. The post does not deny that. The point is that scaling a replacement for industrial animal agriculture requires large quantities of reliable energy services. Efficiency reduces intensity per unit, but it does not eliminate capacity constraints when the objective is a global industrial transition. If you believe efficiency alone removes energy as a binding factor, it would help to specify which services and what magnitude of gains you expect — and on what timeline.

  On primary energy efficiency: that varies by pathway. Many plant-based alternatives appear already competitive or favorable on energy efficiency compared to conventional animal products. Other approaches — especially cultivated meat and some fermentation systems — are widely described in published assessments as energy-uncertain and potentially energy-intensive depending on process assumptions, scale, and energy source. In several analyses, energy demand and energy mix emerge as key drivers of environmental performance and cost. If you have a source showing that cell-cultured meat is already more primary-energy efficient than conventional meat under commercially realistic conditions, I’d welcome it.

  More importantly, per-kg primary energy is only one variable in scaling. Cost, reliability, capital intensity, throughput, and integration into existing industrial systems all shape whether alternatives can replace tens of billions of animals per year. Even if energy intensity is favorable on paper, energy price, grid reliability, or institutional bottlenecks can still affect siting decisions and scaling speed. If you think energy availability is not meaningfully binding in practice, pointing to industrial evidence would make that case stronger.

  You also mention unsupported claims about advocate intentions and scaling drivers. I’m open to correction — but that requires specificity. Which claims are incorrect? What do you see as the dominant constraints on replacing factory farming at scale?

  
  The practical claim here is not that all substitutes are more energy-intensive than animal products. It is that for at least some of the most ambitious and scalable replacement pathways—particularly cultivated meat and certain fermentation systems—energy availability, energy price, and reliability are likely to be decisive constraints as we move from pilot facilities to global production.

  If energy becomes expensive, unreliable, or institutionally constrained, the path from promising prototype to mass adoption slows. And delays are paid for in animal suffering.

Why the An­i­mal Welfare Move­ment Should Em­brace En­ergy Abun­dance, Ar­tifi­cial In­tel­li­gence, and Open Debate

• Wladimir J. Alonso 21 Feb 2026 18:18 UTC

  5 points

  1 ∶ 0

  in reply to: Vasco Grilo🔸’s comment on: Sav­ing hu­man lives cheaply is the most cost-effec­tive way of in­creas­ing an­i­mal welfare?

  Hi Vasco — thanks for inviting me to comment your post. I think we’ve already clarified this in an earlier exchange and found that we’re working from genuinely different aggregation frameworks, and your nematode vs. torture example makes that divergence especially explicit. Since we’d essentially reached an “agree to disagree” already, I’ll leave it here rather than reopening a long back-and-forth.

  Happy to revisit once we have a better empirical handle on ceilings /​ affective capacity.

• Wladimir J. Alonso 16 Feb 2026 12:06 UTC

  5 points

  0 ∶ 0

  on: Turn­ing Farms into Welfare Labs

  Great post, Aaron — I completely agree with your framing of why the lab-to-farm leap feels overdue. Most published welfare research is tied up in universities and controlled settings, which often are not only expensive and slow but often miss how animals actually experience their environments on commercial operations.

  I love your emphasis on starting with engaged farmers — that is a low-friction entry point, especially because so much welfare-relevant data is already being collected in everyday farm management but never shared or analyzed. If even a handful of farmers were willing to share anonymized data, we could begin to extrapolate welfare trends across operations and turn those insights into sector-wide benchmarks that signal progress in a measurable way.

  Crucially, this model can offer upside for participating companies and farms: a feedback loop with welfare experts that helps them refine practices, capture wins, and communicate them externally — for example, “Our data-driven changes reduced mortality (or time in intense pain) by 15% — here’s how.” That kind of concrete progress, grounded in real farm data, can add a powerful real-world layer to academic research and help translate existing knowledge into practice at scale.

• Wladimir J. Alonso 11 Feb 2026 13:40 UTC

  6 points

  0 ∶ 0

  in reply to: Jim Buhler’s comment on: Why Neu­ron Counts Shouldn’t Be Used as Prox­ies for Mo­ral Weight

  Good catch, Jim — and thanks for flagging the terminology. This field is already complicated enough that we really don’t need parallel vocabularies for the same underlying idea. One of the reasons we post publicly is exactly to get this kind of “conceptual linting” from the community.

  I think Birch et al.’s acuity is basically what we mean by resolution: sensitivity to small differences — the ability to discriminate fine gradations (what psychophysics would call “just-noticeable differences”). Where we’re being careful is in separating that from range, which refers to the maximum intensity an organism can plausibly access.

  So in short: acuity ≈ resolution in our usage; it’s distinct from range.

• Wladimir J. Alonso 10 Feb 2026 18:26 UTC

  3 points

  0 ∶ 0

  in reply to: Jim Buhler’s comment on: Why Neu­ron Counts Shouldn’t Be Used as Prox­ies for Mo­ral Weight

  Thanks, Jim — that’s a thoughtful attempt to restate our terms, and it touches on something important you asked in the other thread about bandwidth–acuity vs range–resolution.

  Our range concept maps fairly closely onto the welfare range used in Rethink Priorities’ Moral Weight project — it refers to the upper bound of affective intensity an organism can plausibly access .

  Where we’d adjust is resolution. In the earlier draft your summary makes it sound like resolution is just “precision within a bounded range,” but that framing risks suggesting that resolution is always subordinate to range in motivational function. In fact, from a pure information-encoding perspective, resolution is as versatile as range for enabling intensity-based prioritization, because in principle both could be increased indefinitely: range by extending the extreme ends of the scale, and resolution by subdividing any given range into arbitrarily fine gradations. We develop this point in the The Function and Evolution of Affective Scales section of the Do primitive sentient organisms feel extreme pain? paper

  So the distinction isn’t “range = strength, resolution = detail”; it’s that range and resolution are two orthogonal axes along which affective systems can vary, each capable of supporting graded prioritization. A system with high resolution but modest range could still distinguish and act on nuanced motivational differences without accessing extreme affective intensities at all.

• Wladimir J. Alonso 9 Feb 2026 13:23 UTC

  1 point

  0 ∶ 0

  in reply to: Jim Buhler’s comment on: When Feel­ing Is Worth It: A Cost–Benefit Frame­work for the Evolu­tion of Sentience

  Thanks Jim
  
  On the cost point you raised — “extra integration, valuation, and modulatory capacity are costly only if they decrease fitness in some way, right?” — selection indeed acts on net fitness. Still, it’s both useful and standard to keep costs and benefits analytically separate before recombining them. A trait can be costly in terms of resources or architecture even when it increases fitness overall; brains and immune systems are classic examples.

  On your footnote #3 — “the question of whether organisms with narrower welfare ranges could feel extreme pain” — I think there may be a bit of a contradiction in terms. If an organism has a genuinely narrower welfare range, then by definition (or at least under the operational definitions I’m using), it does not reach disabling or excruciating levels of negative affect. In that framing, the relevant question is precisely where the negative-intensity ceiling lies.

• Wladimir J. Alonso 9 Feb 2026 0:37 UTC

  1 point

  0 ∶ 0

  in reply to: Jim Buhler’s comment on: When Feel­ing Is Worth It: A Cost–Benefit Frame­work for the Evolu­tion of Sentience

  Thanks, Jim — that does get to the crux.

  I think your scenario is plausible in principle: once an alarm is “loud enough,” further increases in intensity could be selectively neutral, so unnecessarily loud alarms might persist by drift, much like neutral variants in molecular evolution.

  My hesitation is about how often extreme felt intensity actually falls into that neutral regime. For neutrality to hold, extra intensity must add no benefit and impose no additional costs or constraints. If affective states are whole-organism control states rather than simple sensory readouts, then escalating intensity plausibly requires extra integration, valuation, or modulatory capacity. In that case, intensity beyond “loud enough” would not be strictly neutral, and drift would be limited.

  So I see neutral drift as a live alternative, but not the default. The framework is meant to clarify when neutrality is plausible versus when selection should instead cap, reshape, or avoid extreme intensity altogether.
  

• Wladimir J. Alonso 8 Feb 2026 23:55 UTC

  3 points

  0 ∶ 0

  in reply to: Vasco Grilo🔸’s comment on: When Feel­ing Is Worth It: A Cost–Benefit Frame­work for the Evolu­tion of Sentience

  Thanks, Vasco — I really appreciate that.

  Yes, some dedicated funding would be very welcome to help expand and accelerate the comparative analysis of affective capacity. I have a few ideas I’d be keen to discuss once the follow-up piece is out and the framework is more fully specified.

  Let’s definitely stay in touch — I’ll make sure to loop back..

• Wladimir J. Alonso 8 Feb 2026 23:38 UTC

  1 point

  1 ∶ 0

  in reply to: Itamar Menuhin-Gruman’s comment on: When Feel­ing Is Worth It: A Cost–Benefit Frame­work for the Evolu­tion of Sentience

  Thanks, Itamar. I’m glad you found the framework useful, and thanks for laying out these concerns.

  (1) On selection in life-or-death situations.
  I’m less convinced that life-or-death contexts should be treated as marginal for evolutionary explanation. Many such hazards (e.g. fire, severe injury, predation) recur across generations, and even small increases in the probability of rapid withdrawal and survival can be strongly selected for. In that sense, excruciating pain in these contexts looks like a straightforward case of ordinary evolutionary logic at work, rather than a special case where costs become irrelevant or selection effectively “switches off.”

  (2) On alternative mechanisms (linear transduction, hormones, etc.).
  I’m very sympathetic to the idea of carefully considering non-adaptive explanations, since some features can indeed arise as byproducts of other selective pressures. I was an avid reader of Stephen Jay Gould, and The Spandrels of San Marco remains a classic reminder of this point. So I agree that high-intensity pain could arise largely as a byproduct of how damage signals scale at the sensory or cellular level, rather than because intensity itself was directly selected for. At the same time, this kind of “linear transduction” may itself carry adaptive value (i.e. function as an exaptation, again in Gould’s sense), since greater damage would naturally call for more urgent behavioral responses to stop it.

  On the point about hormones or other system-wide mechanisms, I may be missing what you have in mind — I’d be very interested in a concrete example and in how you think it would change the cost–benefit picture.

• Wladimir J. Alonso 6 Feb 2026 20:29 UTC

  3 points

  0 ∶ 0

  in reply to: Vasco Grilo🔸’s comment on: When Feel­ing Is Worth It: A Cost–Benefit Frame­work for the Evolu­tion of Sentience

  Thanks a lot, Vasco — and thanks for the upvote!

  You’re absolutely right to push us toward the practical question of how to compare affective capacity across species. That’s ultimately where this line of work needs to go. At the same time, we’ve been deliberately cautious here, because we think this is one of those cases where moving too quickly to numbers or rankings risks making the waters muddier rather than clearer.

  Our sense is that the comparison of affective capacity across species hinges on a set of upstream scientific questions that are still poorly articulated- especially around when sentience arises at all, and when it plausibly extends to very intense affective states. The aim of this piece was to stress-test a way of structuring those questions before turning them into quantitative tools.

  That said, we do see this as complementary to RP’s research agenda on valuing impacts across species. In fact, we think cost–benefit reasoning about sentience and affective intensity can help discipline some of the assumptions that go into moral-weight or welfare-capacity estimates, rather than replacing them.

  We’re currently working on a follow-up that moves closer to a practical comparative framework, and we’re very much treating the present work as groundwork for that. Happy to loop back and share it once it’s ready — and we’d be keen to hear your thoughts then as well.

• Wladimir J. Alonso 6 Feb 2026 20:06 UTC

  3 points

  0 ∶ 0

  in reply to: Jim Buhler’s comment on: When Feel­ing Is Worth It: A Cost–Benefit Frame­work for the Evolu­tion of Sentience

  Thanks a lot for the kind words, Jim — and for the thoughtful pushback.

  I think your point holds if we assume that the only way to implement a very strong alarm is via extreme felt intensity — but that assumption is exactly what we’re questioning.

  I agree that in genuinely catastrophic situations, evolution should tolerate very “loud” alarms. The open question, though, is whether those alarms need to be implemented as extreme affective states, rather than through non-affective or lower-intensity control mechanisms.

  On the benefit side, there seem to be two distinct roles a very strong signal could play. First, triggering an immediate reaction in life-or-death situations. But this doesn’t require affect at all: many organisms (including very simple ones) already show robust threat responses via non-felt control. Even in sentient organisms, immediate escape could in principle be driven by low-intensity affect if thresholds are set low enough, especially where behavioral options are limited.

  Second, overriding other ongoing motivations in organisms with richer behavioral repertoires. Here, stronger affective signals become more plausibly useful, as they can reliably dominate competing drives (foraging, mating, self-maintenance, etc.). One way to achieve this is by expanding affective range rather than relying only on finer discrimination within a narrow range.

  On the cost side, generating and sustaining very high-intensity affective states may plausibly require substantial architectural capacity of the kind we discuss above. In systems with limited computational or neural resources—and especially in organisms with few available behavioral options—extreme felt states may therefore be difficult or unnecessary to implement, regardless of how valuable a very loud alarm would be.