Friday, March 01, 2024

The Leapfrog Hypothesis for AI Consciousness

The first genuinely conscious robot or AI system would, you might think, have relatively simple consciousness -- insect-like consciousness, or jellyfish-like, or frog-like -- rather than the rich complexity of human-level consciousness. It might have vague feelings of dark vs light, the to-be-sought and to-be-avoided, broad internal rumblings, and not much else -- not, for example, complex conscious thoughts about ironies of Hamlet, or multi-part long-term plans about how to form a tax-exempt religious organization. The simple usually precedes the complex. Building a conscious insect-like entity seems a lower technological bar than building a more complex consciousness.

Until recently, that's what I had assumed (in keeping with Basl 2013 and Basl 2014, for example). Now I'm not so sure.

[Dall-E image of a high-tech frog on a lily pad; click to enlarge and clarify]

AI systems are -- presumably! -- not yet meaningfully conscious, not yet sentient, not yet capable of feeling genuine pleasure or pain or having genuine sensory experiences. Robotic eyes "see" but they don't yet see, not like a frog sees. However, they do already far exceed all non-human animals in their capacity to explain the ironies of Hamlet and plan the formation of federally tax-exempt organizations. (Put the "explain" and "plan" in scare quotes, if you like.) For example:

[ChatGPT-4 outputs for "Describe the ironies of Hamlet" and "Devise a multi-part long term plan about how to form a tax-exempt religious organization"; click to enlarge and clarify]

Let's see a frog try that!

Consider, then the Leapfrog Hypothesis: The first conscious AI systems will have rich and complex conscious intelligence, rather than simple conscious intelligence. AI consciousness development will, so to speak, leap right over the frogs, going straight from non-conscious to richly endowed with complex conscious intelligence.

What would it take for the Leapfrog Hypothesis to be true?

First, engineers would have to find it harder to create a genuinely conscious AI system than to create rich and complex representations or intelligent behavioral capacities that are not conscious.

And second, once a genuinely conscious system is created, it would have to be relatively easy thereafter to plug in the pre-existing, already developed complex representations or intelligent behavioral capacities in such a way that they belong to the stream of conscious experience in the new genuinely conscious system. Both of these assumptions seem at least moderately plausible, in these post-GPT days.

Regarding the first assumption: Yes, I know GPT isn't perfect and makes some surprising commonsense mistakes. We're not at genuine artificial general intelligence (AGI) yet -- just a lot closer than I would have guessed in 2018. "Richness" and "complexity" are challenging to quantify (Integrated Information Theory is one attempt). Quite possibly, properly understood, there's currently less richness and complexity in deep learning systems and large language models than it superficially seems. Still, their sensitivity to nuance and detail in the inputs and the structure of their outputs bespeaks complexity far exceeding, at least, light-vs-dark or to-be-sought-vs-to-be-avoided.

Regarding the second assumption, consider a cartoon example, inspired by Global Workspace theories of consciousness. Suppose that, to be conscious, an AI system must have input (perceptual) modules, output (behavioral) modules, side processors for specific cognitive tasks, long- and short-term memory stores, nested goal architectures, and between all of them a "global workspace" which receives selected ("attended") inputs from most or all of the various modules. These attentional targets become centrally available representations, accessible by most or all of the modules. Possibly, for genuine consciousness, the global workspace must have certain further features, such as recurrent processing in tight temporal synchrony. We arguably haven't yet designed a functioning AI system that works exactly along these lines -- but for the sake of this example let's suppose that once we create a good enough version of this architecture, the system is genuinely conscious.

But now, as soon as we have such a system, it might not be difficult to hook it up to a large language model like GPT-7 (GPT-8? GPT-14?) and to provide it with complex input representations full of rich sensory detail. The lights turn on... and as soon as they turn on, we have conscious descriptions of the ironies of Hamlet, richly detailed conscious pictorial or visual inputs, and multi-layered conscious plans. Evidently, we've overleapt the frog.

Of course, Global Workspace Theory might not be the right theory of consciousness. Or my description above might not be the best instantiation of it. But the thought plausibly generalizes to a wide range of functionalist or computationalist architectures: The technological challenge is in creating any consciousness at all in an AI system, and once this challenge is met, giving the system rich sensory and cognitive capacities, far exceeding that of a frog, might be the easy part.

Do I underestimate frogs? Bodily tasks like five-finger grasping and locomotion over uneven surfaces have proven to be technologically daunting (though we're making progress). Maybe the embodied intelligence of a frog or bee is vastly more complex and intelligent than the seemingly complex, intelligent linguistic outputs of a large language model.

Sure thing -- but this doesn't undermine my central thought. In fact, it might buttress it. If consciousness requires frog- or bee-like embodied intelligence -- maybe even biological processes very different from what we can now create in silicon chips -- artificial consciousness might be a long way off. But then we have even longer to prepare the part that seems more distinctively human. We get our conscious AI bee and then plug in GPT-28 instead of GPT-7, plug in a highly advanced radar/lidar system, a 22nd-century voice-to-text system, and so on. As soon as that bee lights up, it lights up big!

Tuesday, February 20, 2024

Could Someone Still Be Collecting a Civil War Widow's Pension? A Possibility Proof

In 1865, a 14-year-old boy becomes a Union soldier in the U.S. Civil War. In 1931, at age 90, he marries an 18-year-old woman, who continues to collect his Civil War pension after he dies. Today, in early 2024, she is one hundred and ten years old, still collecting that pension.

I was inspired to this thought by reflecting about some long-dead people my father knew, who survive in my memory through his stories. How far back might such second-hand memories go? Farther than one might initially suppose -- in principle, back to the 1860s. An elderly philosopher, alive today, might easily have second-hand memories of William James (d. 1910) or Nietzsche (d. 1900), maybe even Karl Marx (d. 1883) or John Stuart Mill (d. 1873).

Second-hand memories have a quality to them that third-hand memories and historical accounts lack. Through my father's and uncle's stories, I feel a kind of personal connection to Timothy Leary (d. 1996), B.F. Skinner (d. 1990), and Abraham Maslow (d. 1970), even though I never met them, in a way I don't to other scholars of the era. It hasn't been so long since their heyday in the 1950s - 1960s, when my father and his brother knew them -- but I might still have several decades in me. My son David, currently a Cognitive Science PhD student at Institut Jean Nicod at ENS in Paris, has also heard such stories, and he could potentially live to see the 22nd century. (My daughter Kate was too young when my father died to have made much of his academic stories.)

The idea that the U.S. might still be paying a Civil War widow's pension is not as ridiculous as it seems. According to this website, the last pension-recieving Union widow died in 2003. According to this website, it was 2008. The last recipient of a Civil War children's benefit died from a hip injury in 2020.

GPT-4 representation of an elderly civil war widow in a cityscape in 2020:

Friday, February 16, 2024

What Types of Argument Convince People to Donate to Charity? Empirical Evidence

Back in 2020, Fiery Cushman and I ran a contest to see if anyone could write a philosophical argument that convinced online research participants to donate a surprise bonus to charity at rates statistically above control. (Chris McVey, Josh May, and I had failed to write any successful arguments in some earlier attempts.) Contributions were not permitted to mention particular real people or events, couldn't be narratives, and couldn't include graphics or vivid descriptions. We wanted to see whether relatively dry philosophical arguments could move people to donate.

We received 90 submissions (mostly from professional philosophers, psychologists, and behavioral economists, but also from other Splintered Mind readers), and we selected 20 that we thought represented a diversity of the most promising arguments. The contest winner was an argument written by Matthew Lindauer and Peter Singer, highlighting that a donation of $25 can save a child in a developing country from going blind due to trachoma, then asking the reader to reflect on how much they would be willing to donate to save their own child from going blind. (Full text here.)

Kirstan Brodie, Jason Nemirow, Fiery, and I decided to follow up by testing all 90 submitted arguments to see what features were present in the most effective arguments. We coded the arguments according to whether, for example, they mentioned children, or appealed to religion, or mentioned the reader's assumed own economic good fortune, etc. -- twenty different features in all. We recruited approximately 9000 participants. Each participant had a 10% chance of winning a surprise bonus of $10. They could either keep the whole $10 or donate some portion of it to one of six effective charities. Participants decided whether to donate, and how much, before knowing if they were among the 10% receiving the $10.

Now, unfortunately, proper statistical analysis is complicated. Because we were working with whatever came in, we couldn't balance argument features, most arguments had multiple coded features, and the coded features tended to correlate between submissions. I'll share a proper analysis of the results later. Today I'll share a simpler analysis. This simple analysis looks at the coded features one by one, comparing the average donation among the set of arguments with the feature to average donation among the set of arguments without the feature.

There is something to be said, I think, for simple analysis even when they aren't perfect: They tend to be easier to understand and to have fewer "researcher degrees of freedom" (and thus less opportunity for p-hacking). Ideally, simple and sophisticated statistical analyses go hand-in-hand, telling a unified story.

So, what argument features appear to be relatively more versus less effective in motivating charitable giving?

Here are our results, from highest to lowest difference in mean donation. "diff" is the dollar difference in mean donation, N is the number of participants who saw an argument with that feature, n is the number of arguments containing that feature, and p is the statistical p-value in a two-sample t test (without correction for multiple comparisons). All analyses are tentative, pending double-checking, skeptical examination, and possibly some remaining data clean-up.

Predictive Argument Features, Highest to Lowest

Does the argument appeal to the notion of equality?
$3.99 vs $3.39 (diff = $.60, N = 395, n = 4, p < .001)

... mention human evolutionary history?
$3.93 vs $3.39 (diff = $.55, N = 4940, n = 5, p < .001)

... specifically mention children?
$3.76 vs $3.26 (diff = $.49, N = 4940, n = 27, p < .001)

... mention a specific, concrete benefit to others that $10 or a similar amount would bring (e.g., 3 mosquito nets or a specific inexpensive medical treatment)?
$3.75 vs $3.44 (diff = $.41, N = 1718, n = 17, p < .001)

... appeal to the diminishing marginal utility of dollars kept by (rich) donors?
$3.69 vs $3.29 (diff = $.40, N = 2843, n = 27, p < .001)

... appeal to the massive marginal utility of dollars transferred to (poor) recipients?
$3.65 vs $3.25 (diff = $.40, N = 3758, n = 36, p < .001)

... mention, or ask the participant to bring to mind, a particular person who is physically or emotionally near to them?
$3.74 vs $3.34 (diff = $.34, N = 318, n = 3, p = .061)

... mention particular needs or hardships such as clean drinking water or blindness?
$3.56 vs $3.23 (diff = $.30, N = 4940, n = 49, p < .001)

... refer to the reader's own assumed economic good fortune?
$3.58 vs $3.31 (diff = $.27, N = 3544, n = 35, p < .001)

... focus on one, single issue? (e.g. trachoma)
$3.61 vs $3.40 (diff = $.21, N = 800, n = 8, p = .07)

... remind people that giving something is better than nothing? (i.e. corrective for drop-in-the-bucket thinking)
$3.56 vs $3.40 (diff = $.15, N = 595, n = 6, p = .24)

... appeal to the views of experts (e.g. philosophers, psychologists)?
$3.47 vs $3.39 (diff = $.07, N = 2629, n = 27, p = .29)

... reference specific external sources such as news reports or empirical studies?
$3.47 vs $3.40 (diff = $.07, N = 1828, n = 18, p = .41)

... explicitly mention that donation is common?
$3.46 vs $3.41 (diff = $.05, N = 736, n = 7, p = .66)

... appeal to the notion of randomness/luck (e.g., nobody chose the country they were born in)?
$3.43 vs $3.41 (diff = $.02, N = 1403, n = 14, p = .80)

... mention religion?
$3.35 vs $3.42 (diff = -$.07, N = 905, n = 9, p = .48)

... appeal to veil-of-ignorance reasoning or other perspective-taking thought experiments?
$3.29 vs $3.23 (diff = -$.14, N = 4940, n = 8, p = .20)

... mention that giving could inspire others to give? (i.e. spark behavioral contagion)
$3.29 vs $3.43 (diff = -$.14, N = 896, n = 9, p = .20)

... explicitly mention and address specific counterarguments?
$3.29 vs $3.45 (diff = -$.15, N = 1829, n = 19, p = .048)

... appeal to the self-interest of the participant?
$3.22 vs $3.49 (diff = -$.30, N = 2604, n = 22, p < .001)

From this analysis, several argument features appear to be effective in increasing participant donations:

  • mentioning children and appealing to the equality of all people,
  • mentioning concrete benefits (one or several),
  • mentioning the reader's assumed economic good fortune and the relatively large impact of a relatively small sacrifice (the "margins" features), and
  • mentioning evolutionary history (e.g., theories that human beings evolved to care more about near others than distant others).
  • Mentioning a particular near person might also have been effective, but since only three arguments were coded in this category, statistical power was poor.

    In contrast, appealing to the participant's self-interest (e.g., that donating will make them feel good) appears to have backfired. Mentioning and addressing counterarguments to donation (e.g., responding to concerns that donations are ineffective or wasted) might also have backfired.

    Now I don't think we should take these results wholly at face value. For example, only five of the ninety arguments appealed to evolutionary history, and all of those arguments included at least two other seemingly effective features: particular hardships, margins, or children. In multiple regression analyses and multi-level analyses that explore how the argument features cluster, it looks like particular hardships, children, and margins might be more robustly predictive -- more on that in a future post. ETA (Feb 19): Where the n < 10 arguments, effects are unlikely to be statistically robust.

    What if we combine argument features? There are various ways to do this, but the simplest is to give an argument one point for any of the ten largest-effect features, then perform a linear regression. The resulting model has an intercept of $3.09 and a slope of $.13. Thus, the model predicts that participants who read arguments with none of these features will donate $3.09, while participants who read a hypothetical argument containing all ten features will donate $4.39.

    Further analysis also suggests that piling up argument features is cumulative: Arguments with at least six of the effective features generated mean donations of $3.89 (vs. $3.37), those with at least seven generated mean donations of $4.46 (vs. $3.38), and the one argument with eight of the ten effective features generated a mean donation of $4.88 (vs. $3.40) (all p's < .001). This eight-feature argument was, in fact, the best performing argument of the ninety. (However, caution is warranted concerning the estimated effect size for any particular argument: With approximately only 100 participants per argument and a standard deviation of about $3, the 95% confidence intervals for the effect size of individual arguments are about +/- $.50.)

    ------------------------------------------------------

    Last month, I articulated and defended the attractiveness of moral expansion through Mengzian extension. On my interpretion of the ancient Chinese philosopher Mengzi, expansion of one's moral perspective often (typically?) begins with noticing how you react to nearby cases -- whether physically nearby (a child in front of you, about to fall into a well) or relationally nearby (your close family members) -- and proceeds by noticing that remote cases (distant children, other people's parents) are similar in important respects.

    None of the twenty coded features captured exactly that. ("Particular near person" was close, but neither necessary nor sufficient: not necessary, because the coders used a stringent standard for when an argument invoked a particular near person, and not sufficient since invoking a particular near person is only the first step in Mengzian extension.) So I asked UCR graduate student Jordan Jackson, who studies Chinese philosophy and with whom I've discussed Mengzian extension, to read all 90 arguments and code them for whether they employed Mengzian extension style reasoning. He found six that did.

    In accord with my hypothesis about the effectiveness of Mengzian extension, the six Mengzian extension arguments outperformed the arguments that did not employ Mengzian extension:

    $3.85 vs $3.38 (diff = $.47, N = 612, n = 6, p < .001)

    Among those six arguments are both the 2020 original contest winner written by Lindauer and Singer and also the best-performing argument in the present study -- though as noted earlier, the best-performing argument in the current study also had many other seemingly effective features.

    In case you're curious, here's the full text of that argument, adapted by Alex Garinther, and quoting extensively, from one of the stimuli in Lindauer et al. 2020

    HEAR ME OUT ON SOMETHING. The explanation below is a bit long, but I promise reading the next few paragraphs will change you.

    As you know, there are many children who live in conditions of severe poverty. As a result, their health, mental development, and even their lives are at risk from lack of safe water, basic health care, and healthy food. These children suffer from malnutrition, unsanitary living conditions, and are susceptible to a variety of diseases. Fortunately, effective aid agencies (like the Against Malaria Foundation) know how to handle these problems; the issue is their resources are limited.

    HERE'S A PHILOSOPHICAL ARGUMENT: Almost all of us think that we should save the life of a child in front of us who is at risk of dying (for example, a child drowning in a shallow pond) if we are able to do so. Most people also agree that all lives are of equal moral worth. The lives of faraway children are no less morally significant than the lives of children close to us, but nearby children exert a more powerful emotional influence. Why?

    SCIENTISTS HAVE A PLAUSIBLE ANSWER: We evolved in small groups in which people helped their neighbors and were suspicious of outsiders, who were often hostile. Today we still have these “Us versus Them” biases, even when outsiders pose no threat to us and could benefit enormously from our help. Our biological history may predispose us to ignore the suffering of faraway people, but we don't have to act that way.

    By taking money that we would otherwise spend on needless luxuries and donating it to an effective aid agency, we can have a big impact. We can provide safe water, basic health care, and healthy food to children living in severe poverty, saving lives and relieving suffering.

    Shouldn't we, then, use at least some of our extra money to help children in severe poverty? By doing so, we can help these children to realize their potential for a full life. Great progress has been made in recent years in addressing the problem of global poverty, but the problem isn't being solved fast enough. Through charitable giving, you can contribute towards more rapid progress in overcoming severe poverty.

    Even a donation $5 can save a life by providing one mosquito net to a child in a malaria-prone area. FIVE DOLLARS could buy us a large cappuccino, and that same amount of money could be used to save a life.

    Friday, February 09, 2024

    Grade Inflation at UC Riverside, and Institutional Pressures for Easier Grading

    Recent news reports have highlighted grade inflation at elite universities: Harvard gave 79% As in 2020-2021, as did Yale in 2022-2023, compared to 67% in 2010-2011. At Harvard, the average GPA has risen from 2.55 in 1950 to 3.05 in 1975 to 3.36 in 1995 to 3.80 now. At Brown, 67% of grades were As in 2020-2021, 10% Bs, and only 1% Cs. It's not just elite universities, however: Grades have risen sharply since at least the 1980s across a wide range of schools.

    I decided to look at UC Riverside's grade distributions since 2013, since faculty now have access to a tool to view this information. (It would be nice to look back farther, but even the changes since 2013 are interesting.)

    The following chart lists grade distributions quarter by quarter for the regular academic year, from 2013 through the present. The dark blue bars at the top are As, medium blue Bs, light blue Cs, and red is D, F, or W.

    [click to enlarge and clarify]

    Three things are visually obvious from this graph:

  • First, there's a spike of high grades in Spring 2020 -- presumably due to the chaos of the early days of the pandemic.
  • Second, the percentage of As is higher in recent years than in earlier years.
  • Third, the percentage of DFWs has remained about the same across the period.
  • In Fall 2013, 32% of enrolled students received As. In Fall 2023, 45% did. (DFW's were 9% in both terms.)

    One open question is whether the new normal of about 45% As reflects a general trend independent of the pandemic spike or whether the pandemic somehow created an enduring change. Another question is whether the higher percentage of As reflects easier grading or better performance. The term "inflation" suggests the former, but of course data of this sort by themselves don't distinguish between those possibilities.

    The increase in percentage As is evident in both lower division and upper division classes, increasing from 32% to 43% in lower division and from 33% to 49% in upper division.

    How about UCR philosophy in particular? I'd like to think that my own department has consistent and rigorous standards. However, as the figure below shows, the trends in UCR philosophy are similar, with an increase from 26% As in Fall 2013 to 41% As in Fall 2024:

    [click to enlarge and clarify]

    Lower division philosophy classes at UCR increased from 25% As in Fall 2013 to 40% As in Fall 2023, while upper division classes increased from 26% to 47% As.

    Smoothing out quarter-by-quarter differences, here is the percentage of As, Fall 2013 - Spring 2014 vs Winter 2023 - Fall 2023 for Philosophy and some selected other disciplines at UCR for comparison:
    Philosophy: 27% to 43% (28% to 42% lower, 25% to 46% upper)
    English: 20% to 33% (15% to 28% lower, 38% to 64% upper)
    History: 28% to 52% (23% to 52% lower, 48% to 52% upper)
    Business: 28% to 46% (20% to 24% lower, 29% to 49% upper)
    Psychology: 32% to 51% (33% to 51% lower, 31% to 51% upper)
    Biology: 22% to 38% (28% to 36% lower, 17% to 41% upper)
    Physics: 26% to 39% (26% to 37% lower, 40% to 41% upper)

    As you can see, in some disciplines at some levels, the percentage of As has almost doubled over the ten-year time period.

    UCR is probably not unusual in the respects I have described. However, if other people have similar analyses for their own institutions, I'd be interested to hear, especially if the pattern is different.

    I doubt, unfortunately, that students are actually performing that much better. UCR philosophy students in 2023 were not dramatically better at writing, critical thinking, and understanding historical material than were students in 2013. I conjecture that the main cause of grade inflation is institutional pressures toward easier grading.

    I see two institutional pressures toward higher grades and more relaxed standards:

    Teaching evaluations: Generally students give better teaching evaluations to professors from whom they expect better grades.[1] Other things being equal, a professor who gives few As will get worse evaluations than one who gives many As. Since professors' teaching is often judged in large part on student evaluations, professors will tend to be institutionally rewarded for giving higher grades, ensuring happier students who give them better evaluations. Professors who are easier graders, if this fact is known among the student body, will also tend to get higher enrollments.

    Graduation rates: At the institutional level, success is often evaluated in terms of graduation rates. If students fail to complete their degrees or take longer than expected to so do because they are struggling with classes, this looks bad for the institution. Thus, there is institutional pressure toward lower standards to ensure high levels of student graduation and "success".

    There are fewer countervailing institutional pressures toward higher rigor and more challenging grading schemes. If classes are too unrigorous, a school might risk losing its WASC accreditation, but few well-established colleges and universities are at genuine risk of losing their accreditation.

    At some point, the grade "A" loses its strength as a signal of excellence. If over 50% of students are receiving As, then an A is consistent with average performance. Yes, for some inspiring teachers and some amazing student groups, average performance might be truly excellent! But that's not the typical scenario.

    I have one positive suggestion for how to deal with grade inflation. But before I get to it, I want to mention one other striking phenomenon: the variation in the grade distributions between terms for what is nominally the same course. For example, here is the distribution chart for one of the lower division classes in UCR's Philosophy Deparment:

    [click to enlarge and clarify]

    The distribution ranges from 11% As in Fall 2014 to 72% As in Fall 2020.

    Some departments in some universities have moved to standardized curricula and tests so that the same class in each term is taught and graded similarly. In philosophy, this is probably not the right approach, since different instructors can reasonably want to focus on different material, approached and graded differently. Still, that degree of term-by-term variation in what is nominally the same class raises issues of fairness to students.

    My suggestion is: sunlight. Let course grade distributions be widely shared and known.

    Sunlight won't solve everything -- far from it -- but I do think that in looking at students' teaching evaluations, seeing the professor's grade distribution provides valuable context that might disincentivize cynical strategies to inflate grades for good evaluations. I've evaluated teaching for teaching awards, for visiting instructors, and for my own colleagues, and I'm struck by how rare it is for information about grade distributions even to be supplied in the context of evaluating teaching. A full picture of a professor's teaching should include an understanding of the range of grades they are distributing and, ideally, random samples of tests and assignments that earn As and Bs and Cs. This situates us to better celebrate the work of professors with high standards and the students in their classes who live up to those high standards.

    Similarly, grade distributions should be made available at the departmental and institutional level. In combination with other evidence -- again, ideally random samples of assignments awarded A, B, and C -- this can help in evaluating the extent to which those departments and institutions are holding students to high standards.

    Student transcripts, too, might be better understood in the context of institutions' and departments' grading standards. This would allow viewers of the transcript to know whether a student's 3.7 GPA is a rare achievement in their institutional context, or simply average performance.

    --------------------------------------------------

    [1] A recent study suggests that grade satisfaction might be the primary driver of the correlation between students' expected grades and their course evaluations, rather than grading leniency per se -- these can come apart when a student is satisfied with their grade as a result of their hard work for it -- but grading leniency is an instructor's easiest path to generating student grade satisfaction, generating the institutional pressure.

    Friday, February 02, 2024

    Swallows and Moles in Philosophy

    In his review (in the journal Science -- cool!) of my recently released book, The Weirdness of the World, Edouard Machery writes:

    There are two kinds of philosophers: swallows and moles. Swallows love to soar and to entertain philosophical hypotheses at best loosely connected with empirical knowledge. Plato and Gottfried Leibniz are paradigmatic swallows. Moles, on the contrary, rummage through mundane facts about our world and aim at better understanding it. Aristotle, William James, and Hans Reichenbach are paradigmatic moles. Eric Schwitzgebel is unabashedly a swallow.

    Machery admits to having a mole's-eye view of the swallows. He praises the book, but he is frustrated by my admittedly wild speculations about radical skepticism, group consciousness, an infinite future, etc.

    Machery's goal in his own recent book Philosophy Within Its Proper Bounds was, he says, "to curtail the flights of fancy with which contemporary philosophers are enamored". The Weirdness of the World celebrates such flights of fancy -- so naturally, Machery and I are going to disagree about the value of wild philosophical speculation.

    Reading Machery's contrast of swallows and moles, I was immediately reminded of how the ancient Chinese philosopher Zhuangzi opens his Inner Chapters:

    There is a fish in the Northern Oblivion named Kun, and this Kun is quite huge, spanning who knows how many thousands of miles. He transforms into a bird named Peng, and this Peng has quite a back on him, stretching who knows how many thousands of miles. When he rouses himself and soars into the air, his wings are like clouds draped across the heavens. The oceans start to churn, and this bird begins his journey toward the Southern Oblivion....

    The quail laughs at him, saying, "Where does he think he's going? I leap into the air with all my might, but before I get farther than a few yards I drop to the ground. My twittering and fluttering between the branches is the utmost form of flying! So where does he think he's going? (Ziporyn trans., p. 3-4).

    Zhuangzi is the swallowiest of swallows, soaring far beyond mundane empirical facts, wondering if life might be a dream, speculating about trees who measure eight thousand years as a single autumn, and celebrating "spirit men" with skin like ice and snow who eat only wind and dew, riding upon the air and clouds.

    Zhuangzi's quail, however, raises a good point: It's much clearer where you're going if you confine yourself to small hops between familiar branches. The Peng is neither practical nor grounded, and Zhuangzi's philosophy is arguably the same. Zhuangzi's friend Huizi scolds him: "Your words are... big and useless, which is why they are rejected by everyone who hears them" (Ziporyn trans., p. 8).

    In defense against Machery and the quail critique, I offer three thoughts:

    First, if anyone is going to speculate about wild possibilities concerning the fundamental nature of things, philosophers should be among them.

    It would be a sad, gray world if our reasoning was always confined to "proper bounds" and we couldn't reflect on issues like dream skepticism, group consciousness, and infinitude. Shouldn't it be part of the job description of philosophy to explore such ideas, considering what can or should be made of them?

    Such speculations needn't be entirely unconstrained by empirical facts, even if empirical science fails to deliver decisive answers. In The Weirdness of the World my speculations always start from empirical observation. My discussion of dream skepticism engages with the science of dreams; my discussion of group consciousness engages with the science of consciousness; my chapter on the possible infinite future -- collaborative with physicist and philosopher of physics Jacob Barandes -- is grounded in the standard working assumptions of mainstream physics. Scientifically informed philosophers are as well-positioned as anyone to speculate about wild hypotheticals that naturally intrigue us (at least some of us). To stand athwart such speculations, saying "Thou shalt not enter this epistemic wilderness!" is to reject an intrinsically valuable form of human philosophical curiosity.

    Second, we can distinguish two types of swallow: those confident that their wild hypotheses are correct and those who merely entertain and explore such hypotheses.

    Maybe Plato was convinced of the reality of Forms and the recollection theory of memory. Maybe Leibniz was convinced that the world was composed of monads in pre-established harmony. But Zhuangzi was a self-undermining skeptic who appears to have taken none of his wild speculations as established fact.

    I don't argue that the United States definitely has conscious experiences; I argue that if we accept standard materialist approaches to consciousness, they seem to imply that it does and that therefore we should take the idea seriously as a possibility. I don't argue that this is a dream or a short-term simulation; I argue that our ordinary culturally-given understanding of the world and mainstream scientific assumptions combine to justify assigning a non-trivial (maybe about 0.1%) credence to both of those possibilities. Barandes and I don't argue that there definitely is an infinite future in which future counterparts of you enact almost every possible action, but only that it follows from "certain not wholly implausible assumptions".

    When soaring in speculation far beyond the mundane local tree branches, doubt is appropriate. The most natural critique of swallows is that they appear to believe wild things on thin evidence. That critique is harder to sustain when the swallow explicitly treats the speculations as speculations only, rather than as established facts.

    Third, the swallow and the mole can collaborate -- even in the work of a single philosopher. As Jonathan Birch comments in my Facebook post linking to Machery's book review, two of Edouard's paradigmatic examples of moles -- Aristotle and William James -- are probably not best thought of as pure moles, but rather as swallow-moles. They dug around quite a bit in mundane empirical facts, yes. But they sometimes also soared with the swallows. Aristotle speculated on the existence of a supraphysical unmoved mover responsible for the existence of the physical world. James speculated about metaphysical "neutral monism" concerning mind and matter and celebrated religious belief beyond the evidence.

    I too have done a fair bit of mundane empirical work -- for example, on the moral behavior of ethics professors (e.g., here and here), on introspective method (e.g., here and here), and on the consequences of exposure to ethical argumentation (e.g., here and here). Even when I am not myself running the empirical studies, much of my work engages with nitty-gritty empirical detail (e.g., on the history of reports of coloration in dreams, on the cognitive capacites of garden snails, on the accuracy of visual imagery reports, and on psychological measures of well-being).

    Often, I think, deep empirical mole-digging is valuable for one's subsequent speculative soaring. Digging into the details of cosmological models enables better informed speculation about the distant future. Digging into the details of the behavior of ethics students and professors enables better informed speculation about the general relation between ethical reflection and ethical behavior. Digging into the details of dream reports enables better informed speculation about dream skepticism. As Zhuangzi imagines, a low-lying fish can transform into a soaring phoenix.

    No single researcher needs to do both the digging and the soaring, even if some of us enjoy both types of task. But it's valuable to have a whole ecosystem of moles and swallows, foxes and hedgehogs, ants and anteaters, truth philosophers and dare philosophers, and so on.

    I'm honored that Machery counts me among the swallows. I celebrate his moleishness. Let's dig and soar!

    Thursday, January 25, 2024

    Imagining Yourself in Another's Shoes vs. Extending Your Concern: Empirical and Ethical Differences

    [new paper in draft]

    The Golden Rule (do unto others as you would have others do unto you) isn't bad, exactly -- it can serve a valuable role -- but I think there's something more empirically and ethically attractive about the relatively underappreciated idea of "extension" found in the ancient Chinese philosopher Mengzi.

    The fundamental idea of extension, as I interpret it, is to notice the concern one naturally has for nearby others -- whether they are relationally near (like close family members) or spatially near (like Mengzi's child about to fall into a well or Peter Singer's child you see drowning in a shallow pond) -- and, attending to relevant similarities between those nearby cases and more distant cases, to extend your concern to the more distant cases.

    I see three primary advantages to extension over the Golden Rule (not that these constitute an exhaustive list of means of moral expansion!).

    (1.) Developmentally and cognitively, extension is less complex. The Golden Rule, properly implemented, involves imagining yourself in another's shoes, then considering what you would want if you were them. This involves a non-trivial amount of "theory of mind" and hypothetical reasoning. You must notice how others' beliefs, desires, and other mental states relevantly differ from yours, then you must imagine yourself hypothetically having those different mental states, and then you must assess what you would want in that hypothetical case. In some cases, there might not even be a fact of the matter about what you would want. (As an extreme example, imagine applying the Golden Rule to an award-winning show poodle. Is there a fact of the matter about what you would want if you were an award winning show poodle?) Mengzian extension seems cognitively simpler: Notice that you are concerned about nearby person X and want W for them, notice that more distant person Y is relevantly similar, and come to want W for them also. This resembles ordinary generalization between relevant cases: This wine should be treated this way, therefore other similar wines should be treated similarly; such-and-such is a good way to treat this person, so such-and-such is probably also a good way to treat this other similar person.

    (2.) Empirically, extension is a more promising method for expanding one's moral concern. Plausibly, it's more of a motivational leap to go from concern about self to concern about distant others (Golden Rule) than to go from concern from nearby others to similar more distant others (Mengzian Extension). When aid agencies appeal for charitable donations, they don't typically ask people to imagine what they would want if they were living in poverty. Instead, they tend to show pictures of children, drawing upon our natural concern for children and inviting us to extend that concern to the target group. Also -- as I plan to discuss in more detail in a post next month -- in the "argument contest" Fiery Cushman and I ran back in 2020, the arguments most successful in inspiring charitable donation employed Mengzian extension techniques, while appeals to "other's shoes" style reasoning did not tend to predict higher levels of donation than did the average argument.

    (3.) Ethically, it's more attractive to ground concern for distant others in the extension of concern for nearby others than in hypothetical self-interest. Although there's something attractive about caring for others because you can imagine what you would want if you were them, there's also something a bit... self-centered? egoistic? ... about grounding other-concern in hypothetical self-concern. Rousseau writes: "love of men derived from love of self is the principle of human justice" (Emile, Bloom trans., p. 235). Mengzi or Confucius would never say this! In Mengzian extension, it is ethically admirable concern for nearby others that is the root of concern for more distant others. Appealingly, I think, the focus is on broadening one's admirable ethical impulses, rather than hypothetical self-interest.

    [ChatGPT4's rendering of Mengzi's example of a child about to fall into a well, with a concerned onlooker; I prefer Helen De Cruz's version]

    My new paper on this -- forthcoming in Daedalus -- is circulating today. As always, comments, objections, corrections, connections welcome, either as comments on this post, on social media, or by email.

    Abstract:

    According to the Golden Rule, you should do unto others as you would have others do unto you. Similarly, people are often exhorted to "imagine themselves in another's shoes." A related but contrasting approach to moral expansion traces back to the ancient Chinese philosopher Mengzi, who urges us to "extend" our concern for those nearby to more distant people. Other approaches to moral expansion involve: attending to the good consequences for oneself of caring for others, expanding one's sense of self, expanding one's sense of community, attending to others' morally relevant properties, and learning by doing. About all such approaches, we can ask three types of question: To what extent do people in fact (e.g., developmentally) broaden and deepen their care for others by these different methods? To what extent do these different methods differ in ethical merit? And how effectively do these different methods produce appropriate care?

    Tuesday, January 16, 2024

    The Weirdness of the World: Release Day and Introduction

    Today is the official U.S. release day of my newest book, The Weirdness of the World!

    As a teaser, here's the introduction:

    In Praise of Weirdness

    The weird sisters, hand in hand,
    Posters of the sea and land,
    Thus do go about, about:
    Thrice to thine and thrice to mine
    And thrice again, to make up nine.
    Peace! the charm’s wound up.
    —Shakespeare, Macbeth, Act I, scene iii

    Weird often saveth
    The undoomed hero if doughty his valor!
    —Beowulf, X.14–15, translated by J. Lesslie Hall


    The word “weird” has deep roots in old English, originally as a noun for fate or magic, later evolving toward its present use as an adjective for the uncanny or peculiar. By the 1980s, it had fruited as the choicest middle-school insult against unstylish kids like me who spent their free time playing with figurines of wizards and listening to obscure science fiction radio shows. If the “normal” is the conventional, ordinary, and readily understood, the weird is what defies that.

    The world is weird -- deeply, pervasively so, weird to its core, or so I will argue in this book. Among the weirdest things about Earth is that certain complex bags of mostly water can pause to reflect on the most fundamental questions there are. We can philosophize to the limits of our comprehension and peer into the fog beyond those limits. We can contemplate the foundations of reality, and the basis of our understanding of those foundations, and the necessary conditions of the basis of our understanding of those foundations, and so on, trying always to peer behind the next curtain, even with no clear method and no great hope of a satisfying end to the inquiry. In this respect, we vastly outgeek bluebirds and kangaroos and are rightly a source of amazement to ourselves.

    I will argue that careful inquiry into fundamental questions about consciousness and cosmology reveals not a set of readily comprehensible answers but instead a complex blossoming of bizarre possibilities. These possibilities compete with one another, or combine in non-obvious ways. Philosophical and cosmological inquiry teaches us that something radically contrary to common sense must be true about the fundamental structures of the mind and the world, while leaving us poorly equipped to determine where exactly the truth lies among the various weird possibilities.

    We needn’t feel disappointed by this outcome. The world is richer and more interesting for escaping our understanding. How boring it would be if everything made sense!

    1. My Weird Thesis

    Consider three huge questions: What is the fundamental structure of the cosmos? How does human consciousness fit into it? What should we value? What I will argue in this book -- with emphasis on the first two questions but also sometimes touching on the third -- is (1) the answers to these questions are currently beyond our capacity to know, and (2) we do nonetheless know at least this: Whatever the truth is, it’s weird. Careful reflection will reveal that every viable theory on these grand topics is both bizarre and dubious. In chapter 2 (“Universal Bizarreness and Universal Dubiety”), I will call this the Universal Bizarreness thesis and the Universal Dubiety thesis. Something that seems almost too preposterous to believe must be true, but we lack the means to resolve which of the various preposterous-seeming options is in fact correct. If you’ve ever wondered why every wide-ranging, foundations-minded philosopher in the history of Earth has held bizarre metaphysical or cosmological views (I challenge you to find an exception!) -- with each philosopher holding, seemingly, a different set of bizarre views -- chapter 2 offers an explanation.

    I will argue that every approach to cosmology and consciousness has implications that run strikingly contrary to mainstream “common sense” and that, partly in consequence, we ought to hold such theories only tentatively. Sometimes we can be justified in simply abandoning what we previously thought of as common sense, when we have firm scientific grounds for thinking otherwise; but questions of the sort I explore in this book test the limits of scientific inquiry. Concerning such matters, nothing is firm -- neither common sense, nor science, nor any of our other epistemic tools. The nature and value of scientific inquiry itself rely on disputable assumptions about the fundamental structure of the mind and the world, as I discuss in chapters on skepticism (chapter 4), idealism (chapter 5), and whether the external world exists (chapter 6).

    On a philosopher’s time scale -- where a few decades ago is “recent” and a few decades from now is “soon” -- we live in a time of change, with cosmological theories and theories of consciousness rising and receding in popularity based mainly on broad promise and what captures researchers’ imaginations. We ought not trust that the current range of mainstream theories will closely resemble the range in a hundred years, much less the actual truth.

    2. Varieties of Cosmological Weirdness

    To establish that the world is cosmologically weird, maybe all that is needed is relativity theory and quantum mechanics.

    According to relativity theory, if your twin accelerates away from you at very high speed, then returns, much less time will have passed for the traveler than for you who stayed here on Earth -- the so-called Twin Paradox. According to the most straightforward interpretation of quantum mechanics, if you observe what we ordinarily consider to be a chance event, there’s also an equally real, equally existing version of you in another “world” who shares your past but for whom the event turned out differently. (Or maybe your act of observation caused the event to turn out one way rather than the other, or maybe some other bizarre thing is true, depending on the correct interpretation of quantum mechanics, but it’s widely accepted that there are no non-bizarre interpretations.) So if you observe the chance decay of a uranium atom, for example, there’s another world branching of from this one, containing a counterpart of you who observes the atom not to have decayed. If we accept that view, then the cosmos contains a myriad of different, equally real worlds, each with different versions of you and your friends and everything you know, all splitting off from a common past.

    I won’t dwell on those particular cosmological peculiarities, since they are familiar to academic readers and well handled elsewhere. However, some equally fundamental cosmological issues are typically addressed by philosophers rather than scientific cosmologists.

    One is the possibility that the cosmos is nowhere near as large as we ordinarily assume -- perhaps just you and your immediate environment (chapter 4) or perhaps even just your own mind and nothing else (chapter 6). Although these possibilities might appear unlikely, they are worth considering seriously, to assess how confident we ought to be in their falsity, and on what grounds. I will argue that it’s reasonable not to entirely dismiss such skeptical possibilities. Alternatively, and more in line with mainstream physical theory, the cosmos might be infinite, which brings its own train of bizarre consequences (chapter 7).

    Another possibility is that we live inside a simulated reality or a pocket universe, embedded in a much larger structure about which we know virtually nothing (chapters 4 and 5). Yet another possibility is that our experience of three-dimensional spatiality is a product of our own minds that doesn’t reflect the underlying structure of reality (chapter 5) or that our sensory experience maps only loosely onto the underlying structure of reality (chapter 9).

    Still another set of questions concerns the relationship of mind to cosmos. Is conscious experience abundant in the universe, or does it require the delicate coordination of rare events (chapter 10)? Is consciousness purely a matter of having the right physical structure, or might it require something non-physical (chapter 2)? Under what conditions might a group of organisms give rise to group-level consciousness (chapter 3)? What would it take to build a conscious machine, if that is possible at all -- and what should we do if we don’t know whether we have succeeded (chapter 11)?

    In each of our heads there are about as many neurons as stars in our galaxy, and each neuron is arguably more structurally complex than any star system that does not contain life. There is as much complexity and mystery inside as out.

    The repeated theme: In the most fundamental matters of consciousness and cosmology, neither common sense, nor early twenty-first-century empirical science, nor armchair philosophical theorizing is entirely trustworthy. The rational response is to distribute our credence across a wide range of bizarre options.

    Each chapter is meant to be separately comprehensible. Please feel free to skip ahead, reading any subset of them in any order.

    3. Philosophy That Closes versus Philosophy That Opens

    You are reading a philosophy book -- voluntarily, let’s suppose. Why? Some people read philosophy because they believe it reveals profound, fundamental truths about the way the world really is and the one right manner to live. Others like the beauty of grand philosophical systems. Still others like the clever back-and-forth of philosophical dispute. What I like most is none of these. I love philosophy best when it opens my mind -- when it reveals ways the world could be, possible approaches to life, lenses through which I might see and value things around me, which I might not other wise have considered.

    Philosophy can aim to open or to close. Suppose you enter Philosophical Topic X imagining three viable, mutually exclusive possibilities, A, B, and C. The philosophy of closing aims to reduce the three to one. It aims to convince you that possibility A is correct and the others wrong. If it succeeds, you know the truth about Topic X: A is the answer! In contrast, the philosophy of opening aims to add new possibilities to the mix -- possibilities that you hadn’t considered before or had considered but too quickly dismissed. Instead of reducing three to one, three grows to maybe five, with new possibilities D and E. We can learn by addition as well as subtraction. We can learn that the range of viable possibilities is broader than we had assumed.

    For me, the greatest philosophical thrill is realizing that something I’d long taken for granted might not be true, that some “obvious” apparent truth is in fact doubtable -- not just abstractly and hypothetically doubtable, but really, seriously, in-my-gut doubtable. The ground shifts beneath me. Where I’d thought there would be floor, there is instead open space I hadn’t previously seen. My mind spins in new, unfamiliar directions. I wonder, and the world itself seems to glow with a new wondrousness. The cosmos expands, bigger with possibility, more complex, more unfathomable. I feel small and confused, but in a good way.

    Let’s test the boundaries of the best current work in science and philosophy. Let’s launch ourselves at questions monstrously large and formidable. Let’s contemplate these questions carefully, with serious scholarly rigor, pushing against the edge of human knowledge. That is an intrinsically worthwhile activity, worth some of our time in a society generous enough to permit us such time, even if the answers elude us.

    My middle-school self who used dice and thrift-shop costumes to imagine astronauts and wizards is now a middle-aged philosopher who uses twenty-first-century science and philosophy to imagine the shape of the cosmos and the magic of consciousness. Join me! If doughty our valor, mayhap the weird saveth us.

    Friday, January 12, 2024

    Demographic Trends in the U.S. Philosophy Major, 2001-2022 -- Including Total Majors, Second Majors, Gender, and Race

    I'm preparing for an Eastern APA session on the "State of Philosophy" next Thursday, and I thought I'd share some data on philosophy major bachelor's degree completions from the National Center for Education Statistics IPEDS database, which compiles data on virtually all students graduating from accredited colleges and universities in the U.S., as reported by administrators.

    I examined all data from the 2000-2001 academic year (the first year in which they started recording data on second majors) through 2021-2022 (the most recent available year).

    Total Numbers of Philosophy Majors: The Decline Has Stopped

    First, the sharp decline in philosophy majors since 2013 has stopped:

    2001:  5836
    2002:  6529
    2003:  7023
    2004:  7707
    2005:  8283
    2006:  8532
    2007:  8541
    2008:  8778
    2009:  8996
    2010:  9268
    2011:  9292
    2012:  9362
    2013:  9427
    2014:  8820
    2015:  8184
    2016:  7489
    2017:  7572
    2018:  7667
    2019:  8074
    2020:  8209
    2021:  8328
    2022:  7958

    (The decline between 2021 and 2022 reflects a general decline in completions of bachelor's degrees due to the pandemic that year, rather than a trend specific to philosophy.)

    In general, the humanities have declined sharply since 2010, and history, English, and foreign languages and literature continue to decline.  This graph shows the trend:
    [click image to enlarge and clarify]

    The decline in the English major is particularly striking, from 4.5% of bachelor's degrees awarded in 2000-2001 to 1.8% in 2021-2022.  Philosophy peaked at 0.60% in 2005-2006 and has held steady at 0.39%-0.40% since 2015-2016.

    Philosophy Relies on Double Majors

    [Expanded and edited for clarity, Jan 15] Breaking the data down by first major vs second major, we can see that over time an increasing proportion of students have philosophy as their second major.  In some schools, the distinction between "first major" and "second major" is meaningful, with the first indicating the primary major.  In other schools the distinction is not meaningful.  In the 2021-2022 academic year, 24% of students who took a bachelor's degree in philosophy had it listed as their second major.

    [click image to enlarge and clarify]

    From these numbers we can estimate that philosophy students are at least moderately likely to be double majors.  While it's impossible to know what percentage of students who took philosophy as their first major also carried a second major, a ballpark estimate might assume that about half of students with philosophy plus one other major list philosophy first rather than second.  If so, then approximately half of all philosophy majors (48%) are double majors.  Overall, across all majors, only 5% of students double majored.

    The ease of double majoring is likely to influence the number of students who choose philosophy as a major.

    Gender Disparity Is Decreasing

    NCES classifies all students as men or women, with no nonbinary category and no unclassified students.  Since the beginning of the available data in the 1980s through the mid-2010s, the percentage of women among philosophy bachelor's recipients hovered steadily between 30% and 34%, not changing even as the total percentage of women increased from 51% to 57%.  However, the last several years have seen a clear decrease in gender disparity, with women now earning 41% of philosophy degrees.

    [click image to enlarge and clarify]

    Black Students Remain Underrepresented in Philosophy Compared to Undergraduates Overall, and Other Race/Ethnicity Data

    NCES uses the following race/ethnicity categories: U.S. nonresident, race/ethnicity unknown, Hispanic or Latino (any race), and among U.S. residents who are not Hispanic or Latino: American Indian or Alaska Native, Asian, Black or African American, Native Hawaiian or Other Pacific Islander, White, and two or more races.  Before 2007-2008, Native Hawaiian or Other Pacific Islander was included with Asian, but inconsistently until 2010-2011.  The two-or-more races option was also introduced in the 2007-2008 academic year, again with inconsistent reporting for several years.

    I've charted these categories below.  As you can see, for most categories, the percentages are similar for philosophy and for graduates overall, except that non-Hispanic White is slightly higher for philosophy and non-Hispanic Black significantly lower. In 2021-2022, non-Hispanic Black people were 14% of the U.S. population age 18-24, 10% of bachelor's degree recipients, and 6% of philosophy bachelor's recipients.

    [as usual, click the figures to expand and clarify]

    I interpret the sharp increase in multi-racial students as reflecting reporting issues and an increasing willingness of students to identify as multi-racial.

    It's also worth noting that although philosophy majors are approximately as likely to be Hispanic/Latino as graduates overall, Hispanic/Latino students are underrepresented among bachelor's degree recipients relative to the U.S. population age 18-24 (17% vs 23%). Non-Hispanic American Indian / Alaska Native students are also underrepresented among overall graduates (0.46% vs. 0.84% of the population age 18-24), and maybe particularly so in philosophy (0.37% vs 0.46% in the most recent year).

    Friday, January 05, 2024

    Credence-Weighted Robot Rights?

    You're a firefighter in the year 2050 or 2100. You can rescue either one human, who is definitely conscious, or two futuristic robots, who might or might not be conscious. What do you do?

    [Illustration by Nicolas Demers, from my newest book, The Weirdness of the World, to be released Jan 16 and available for pre-order now.]

    Suppose you think there's a 75% chance that the robots have conscious lives as rich as those of human beings (or, alternatively, that they have whatever else it takes to have "full moral status" equivalent to that of a human). And suppose you think there's a 25% chance that the robots are the moral equivalent of toasters, that is, mere empty machines with no significant capacity for conscious thought or feeling.

    Arguably, if you save the robots and let the human die, you maximize the total expected number of humanlike lives saved (.75 * 2 + .25 * 0 = 1.5 expected lives saved, vs. one life for sure if you save the human). Decision-theoretically, it looks similar to choosing an action with a 75% chance of saving two people over an action that will save one person for sure. Applying similar reasoning, if the credences are flipped (25% chance the robots are conscious, 75% they're not), you save the human.

    Generalizing: Whatever concern you have for an ordinary human, or whatever you would give on their behalf, multiply that concern by your credence or degree of belief that the robot has human-like consciousness (or alternatively your credence that it has whatever features justify moral consideration similar to that of a human). If you'd give $5 to a human beggar, give $3 to a robot beggar in the same situation, if you think it's 60% likely the robot has human-like consciousness. If an oversubscribed local elementary school has a lottery for admission and resident human children each get a 50% chance of admission, resident robot children of disputable consciousness would get a proportionately reduced chance.

    Call this approach credence-weighted robot rights.

    I see a least three problems with credence-weighted robot rights:

    (1.) Credence-weighted robot rights entail that robots will inevitably be treated as inferior, until we are 100% confident that they are our equals.

    Of course it's reasonable to treat robots as inferior to humans now. We should save the person, not the robot, in the fire. And of course if we ever create robots who are beyond all reasonable doubt our equals, we should treat them as such. I'm hypothesizing instead a tricky in-between case -- a period during which it's reasonably disputable whether or not our machines deserve full moral status as our equals, a period during which liberals about robot consciousness and robot rights regard robots as our fully-conscious moral peers, while conservatives about robot consciousness and robot rights regard them as mindless machines to be deployed and discarded however we wish.

    If we choose a 75% chance of rescuing two people over a sure-fire rescue of one person, we are not treating the unrescued person as inferior. Each person's life is worth just as much in our calculus as that of the others. But if we rescue five humans rather than six robots we regard as 80% likely to be conscious, we are treating the robots as inferior -- even though, by our own admission, they are probably not. It seems unfortunate and less than ethically ideal to always treat as inferiors entities we regard as probably our equals.

    (2.) Credence-weighted robot rights would engender chaos if people have highly variable opinions. If individual firefighters make the choices based on their personal opinions, then one firefighter might save the two robots while another saves the one human, and each might find the other's decision abhorrent. Stationwide policies might be adopted, but any one policy would be controversial, and robots might face very different treatment in different regions. If individual judges or police were to apply the law differently based on their different individual credences, or on the variable and hard-to-detect credences of those accused of offences against robots, that would be unfair both to the robots and to the offenders, since the penalty would vary depending on who happened to be the officer or judge or whether they travel in social circles with relatively high vs. low opinions of robot consciousness. So presumably there would have to be some regularization by jurisdiction. But still, different jurisdictions might have very different laws concerning the demolition or neglectful destruction of a robot, some treating it as 80% of a homicide, others treating it as a misdemeanor -- and if robot technologies are variable and changing, the law, and people's understanding of the law, might struggle to keep up and to distinguish serious offences from minor ones.

    (3.) Chaos might also ensue from the likely cognitive and bodily diversity of robots. While human cognitive and bodily variability typically keeps within familiar bounds, with familiar patterns of ability and disability, robots might differ radically. Some might be designed with conscious sensory experiences but no capacity for pain or pleasure. Others might experience intense pain or pleasure but lack cognitive sophistication. Others might have no stable goals or model their goals wholly on instructions from a human to whom they are gladly, perhaps excessively subservient, insufficiently valuing their own life. Still others might be able to merge and divide at will, or back themselves up, or radically reformat themselves, raising questions about the boundaries of the individual and what constitutes death. Some might exist entirely as computational entities in virtual paradises with little practical connection to our world. All this raises the question of what features are necessary for, and what constitutes, "equal" rights for robots, and whether thresholds of equality even make sense. Our understanding might require a controversial multidimensional scalar appreciation of the grounds of moral status.

    Other approaches have their own problems. A precautionary principle that grants fully human equal rights as soon as it's reasonable to think that robots might deserve them risks sacrificing substantial human interests for machines that very likely don't have interests worth the sacrifice (letting a human die, for example, to save a machine that's only 5% likely to be conscious), and it perhaps makes the question of the grounds of moral status in the face of future robots' cognitive diversity even more troubling and urgent. Requiring proof of consciousness beyond reasonable doubt aggravates the issue of treating robots as subhuman even if we're pretty confident they deserve equal treatment. Treating rights as a negotiated social construction risks denying rights to entities that really do deserve rights, based on their intrinsic conscious capacities, if we collectively choose as a matter of social policy not to grant those rights.

    The cleanest solution would be what Mara Garza and I have called the Design Policy of the Excluded Middle: Don't create AI systems whose moral status is dubious and confusing. Either create only AI systems that we recognize as property without human-like moral status and rights, and treat them accordingly; or go all the way to creating AI systems with a full suite of features that enable consensus about their high moral status, and then give them the rights they deserve. It's the confusing cases in the middle that create trouble.

    If AI technology continues to advance, however, I very much doubt that it will do so in accord with the Design Policy of the Excluded Middle -- and thus we will be tossed into moral confusion about how to treat our AI systems, with no good means of handling that confusion.

    -------------------------------------------------------------

    Related:

    The Weirdness of the World, Chapter 11 (forthcoming), Princeton University Press.

    The Full Rights Dilemma for AI Systems of Debatable Moral Personhood, Robonomics, 4 (2023), #32.

    How Robots and Monsters Might Break Human Moral Systems (Feb 3, 2015)

    Designing AI with Rights, Consciousness, Self-Respect, and Freedom (2020) (with Mara Garza), in S. Matthew Liao, ed., The Ethics of Artificial Intelligence, Oxford University Press.

    Monday, January 01, 2024

    Writings of 2023

    Each New Year's Day, I post a retrospect of the past year's writings. Here are the retrospects of 2012, 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, and 2022.

    The biggest project for the past few years has been my new book The Weirdness of the World, available for pre-order and scheduled for U.S. release on January 16. This book pulls together ideas I've been publishing since 2012 concerning the failure of common sense, philosophy, and empirical science to explain consciousness and the fundamental structure of the cosmos, and the corresponding bizarreness and dubiety of all general theories about such matters.

    -----------------------------------

    Books forthcoming:

    The Weirdness of the World (under contract with Princeton University Press).
      See description above.
    Books under contract / in progress:

    As co-editor with Jonathan Jong, The Nature of Belief, Oxford University Press.

      Collects 15 new essays on the topic, by Sara Aronowitz, Tim Crane and Katalin Farkas, Carolina Flores, M.B. Ganapini, David Hunter, David King and Aaron Zimmerman, Angela Mendelovici, Joshua Mugg, Bence Nanay, Nic Porot and Eric Mandelbaum, Eric Schwitzgebel, Keshav Singh, Declan Smithies, Ema Sullivan-Bissett, amd Neil Van Leeuwen.
    As co-editor with Helen De Cruz and Rich Horton, a yet-to-be-titled anthology with MIT Press containing great classics of philosophical SF.


    Full-length non-fiction essays, published 2023:

    Revised and updated: "Belief", Stanford Encyclopedia of Philosophy.

      A broad-ranging review of the main philosophical approaches to belief.
    "Borderline consciousness: When it's neither determinately true nor determinately false that consciousness is present", Philosophical Studies, 180, 3415–3439.
      Being conscious is not an on-or-off phenomenon but has gray zones. Our failure to conceive, in a certain way, of such in-between cases is no evidence against their existence.
    "Creating a large language model of a philosopher" (with David Schwitzgebel and Anna Strasser), Mind and Language [online article mila.12466, print forthcoming].
      We trained GPT-3 on the corpus of Daniel Dennett, and even Dennett experts had trouble distinguishing its answers to philosophical questions from Dennett's actual answers.
    "The full rights dilemma for AI systems of debatable moral personhood", Robonomics, 4 (32).
      We might soon create AI systems where it's a legitimately open question whether they have humanlike consciousness and deserve humanlike rights. There are huge moral risks however we respond to such cases.
    "What is unique about kindness? Exploring the proximal experience of prosocial acts relative to other positive behaviors" (with Annie Regan, Seth Margolis, Daniel J. Ozer, and Sonja Lyubomirsky), Affective Science, 4, 92-100.
      Participants assigned to do kind acts for others reported a greater sense of competence, self-confidence, and meaning while engaging in those acts across the intervention period.


    Full-length non-fiction essays, finished and forthcoming:

    "Dispositionalism, yay! Representationalism, boo!" in J. Jong and E. Schwitzgebel, eds., The Nature of Belief, Oxford.

      Presents three problems for hard-core representationalism about belief: The Problem of Causal Specification, the Problem of Tacit Belief, and the Problem of Indiscrete Belief.
    "Repetition and value in an infinite universe", in S. Hetherington, ed., Extreme Philosophy, Routledge.
      Standard decision theory fails when confronted with the possibility of infinitely many consequences of our actions. Still, it's reasonable to prefer that the universe is infinite rather than finite.
    "The ethics of life as it could be: Do we have moral obligations to artificial life?" (with Olaf Witkowski), Artificial Life.
      Creators of artificial life should bear in mind the conditions under which artificial systems might come to be genuine targets of moral concern.


    Full-length non-fiction essays, in draft and circulating:

    "The prospects and challenges of measuring morality" (with Jessie Sun).

      Could we create a "moralometer" -- that is, a valid measure of a person's general morality? The conceptual and methodological challenges would be formidable.
    "The washout argument against longtermism" (commentary on William MacAskill's book What We Owe the Future).
      We cannot be justified in believing that any actions currently available to us will have a non-negligible positive influence on the billion-plus-year future.
    "Let's hope we're not living in a simulation" (commentary on David Chalmers's book Reality+).
      If we are living in a simulation, there's a good chance it's small or brief and we are radically mistaken about the past, future, and/or distant things.
    "Consciousness in Artificial Intelligence: Insights from the science of consciousness" (one of 19 authors, with Patrick Butlin and Robert Long).
      Some mainstream scientific theories of consciousness imply that we might be on the verge of creating AI systems that genuinely have conscious experiences.
    "The necessity of construct and external validity for generalized causal claims: A critical review of the literature on quantitative causal inference" (with Kevin Esterling and David Brady).
      We develop a formal model of causal specification which clarifies the necessity of construct validity and external validity for deductive causal inference.
    "Inflate and explode".
      Illusionists and eliminativists about phenomenal consciousness illegitimately build objectionable presuppositions into the notion of "phenomenal consciousness" and defeat only this artificially inflated notion. (I wrote this a few years ago and I'm undecided about whether to trunk this one or revise it.)


    Selected shorter non-fiction:

    "Uncle Iroh, from fool to sage -- or sage all along? (with David Schwitzgebel), in J. De Smedt and H. De Cruz, eds., Avatar: The Last Airbender and Philosophy (2023), Wiley Blackwell.

      Uncle Iroh is a Zhuangzian sage, and ordinary viewers immediately glimpse the sageliness behind his veneer of foolishness.
    "Dehumanizing the cognitively disabled: Commentary on Smith's Making Monsters" (with Amelie Green), Analysis Reviews (forthcoming).
      We describe Amelie Green's experience witnessing the dehumanization of the cognitively disabled in care homes, comparing it with Smith's treatment of racial dehumanization.
    "Introspection in group minds, disunities of consciousness, and indiscrete persons" (with Sophie R. Nelson), Journal of Consciousness Studies, 30 (2023), #9-10, 288-303.
      We describe a hypothetical AI system that defies the usual sharp lines between cognitive systems, conscious experiencers, and persons.
    "Quasi-sociality: Towards asymmetric joint actions with artificial systems" (with Anna Strasser), in A. Strasser, ed., How to Live with Smart Machines? (forthcoming), Xenemoi.
      AI systems might soon occupy the gray area between being asocial tools and being real, but junior, social partners.
    "AI systems must not confuse users about their sentience or moral status", Patterns, 4 (2023), #8, 100818.
      AI systems should be designed to either be clearly nonsentient tools or (if it's ever possible) clearly sentient entities who deserve appropriate care and protection.
    "How far can we get in creating a digital replica of a philosopher?" (with Anna Strasser and Matt Crosby), in R. Hakli, P. Mäkelä, J. Seibt, eds., Social Robots in Social Institutions: Proceedings of Robophilosophy 2022. Series Frontiers of AI and Its Applications, vol. 366 (2023), IOS Press.

    "Don't make moral calculations based on the far future", The Latecomer (Dec 19, 2023).

      An epistemic critique of "longtermism".

    "Could the Universe Be Finite? (with Jacob Barandes), Nautilus (Dec 15, 2023).

      Well, probably not.

    "Is it time to start considering personhood rights for AI chatbots?" (with Henry Shevlin), Los Angeles Times (Mar 5, 2023).

      Reflections on the hazards of confusion about the moral status of AI systems


    Science fiction stories

    "Larva, pupa, imago", Clarkesworld, issue 197, (2023).

      The life-cycle and worldview of a cognitively enhanced future butterfly.


    Some favorite blog posts

    "The black hole objection to longtermism and consequentialism" (Apr 13).

    "'There are no chairs' says the illusionist, sitting in one" (Apr 24).

    "We shouldn't 'box' superintelligent AIs" (May 21).

    "The fundamental argument for dispositionalism about belief" (Jun 7).

    "The Summer Illusion" (Jul 10).

    "One reason to walk the walk: To give specific content to your assertions" (Sep 8).

    "Percent of U.S. philosophy PhD recipients who are women: A 50-year perspective" (Nov 3).


    Happy New Year!