When I learned about Jeremy Bentham for the very first time many years ago, it had nothing to do with his ideas on utilitarianism. I stumbled upon a picture of his auto-icon on the internet and was frankly overwhelmed. Although my immediate reaction at the time was to close the tab, I did, out of curiosity, go back to take a second look. As someone that could be prescribed as clinically private, I found it rather odd and, at the same time, fascinating that a thinker would request such an interesting postmortem “arrangement” for himself; I’d always assumed that the motive behind leaving one’s own body for public display was to achieve everlasting influence or some kind of eternity, in general, and enlightening figures like Bentham could and would accomplish that simply through their works, if they so desired. However, it was indeed my own bigoted bias. As the founder of modern utilitarianism, Bentham had his own utilitarian reasoning for this decision — as if you could hear him shout out maximize my utility after I’m dead straight from his deathbed. He donated his body for dissection to benefit medical science, in a time when cadavers for medical education were very hard to obtain, and indicated that having the rest of him stuffed and mummified into an “auto-icon” could, among other benefits, reduce the need for paintings, statues, and other monuments as remembrances of people, eliminate the danger posed to public health by accumulating corpses, and “diminish the horrors of death” by leaving only its “agreeable associations.”
Bentham lived to 84. His auto-icon and writings were not the only “utility facilities” that he kindly left behind. He had a few quite famous students that carried his spirits on, including James Mill and his son, John Stuart Mill, who was raised to and did eventually become the torchbearer of the school of utilitarianism (probably helpful to mention that both the father and the son were also members of the Associationist School). As you would have already guessed from the name alone, the Mill, the ambitious logic-based ethics machine which is the subject of this long overdue post, is named after J.S. Mill and his wife, Harriet Taylor Mill, for their trailblazing views and efforts in the many fields of individual rights and liberty. J.S. Mill’s thoughts on utilitarianism, liberalism and practical reason are also the foundation and core constituents of the Mill computation engine. I’ll briefly explore some of them later in the article, but, first, we need to answer a basic question.
What’s the Mill?
I’ve mentioned the Mill a couple of times in a few previous posts; it was referenced as a moral authority for the Dawn system and major contributing factor in GEN ai. By definition, it is a machine-run program that somehow has the ability to advise on ethics and produce an output when given a situation. Although it may sound like so, it’s, however, not another The Code of Hammurabi or a huge collection of decision flowcharts, both of which are powerful and effective models that have long been used extensively to determine right or wrong. The Mill may never be that powerful or effective, not even in my wildest imagination, partially because I am deeply aware of my own limits but, more importantly, because it’s not what it’s being built for. Compared to the existing models that are predominantly semantics-dependent and therefore subject to interpretations, the Mill categorically favors syntax (logic). I could also add that the Mill does not intend to make determinations or, further, judgements; actually, its only ambition, albeit a mighty one admittedly, is to strenuously and exhaustively reach for near-perfect proofs.
I’ll use an example to illustrate the differences.
The most controversial player in the NBA right now is Kyrie Irving (it’s a tough life being someone that lives in New York and loves basketball). Recently, he promoted a movie on Twitter that had antisemitic content and, as a result, received huge backlash. However, besides the mainstream sentiment, there are other voices that question the scope of freedom of speech and the meaning of this supposed freedom if freedom does not include free of consequences. To examine this seemingly complex situation with a traditional model, we can say Kyrie’s behavior is wrong because anti-semitism is wrong, disinformation is wrong, and promoting antisemitic disinformation (to a large audience) is wrong. Then in the next step, we’ll further show that antisemitism is wrong because racially-motivated hatred and discrimination is always wrong and dangerous. This reasoning process can go on for a few more rounds, though, in reality, we arrive at the conclusion almost instantly. Meanwhile, on the opposite side, one could counterargue that Kyrie is not wrong because his actions were conducted within his own space (personal Twitter account) and he had the absolute liberty to speak his own mind and express his own opinions. As for his followers that were exposed to the information on his timeline, they were not coerced or tricked; it was by choice. If we follow both perspectives and keep breaking the issue down, the two sides would ultimately reach the stage of a stalemate as the two uncompromisable entities at the center of this discussion, one’s agency and another’s equality, are independent of each other. On a side note, as I see it, it seems that the assumptions that have caused this stalemate are questionable in the first place for they are incorrectly framed and contextualized. Freedom of speech is in the Constitution to restrict the power of the government, as opposed to that of the ordinary people (the communities) in our everyday life. Therefore, it cannot shield anyone from public condemnation. However, it’s worth noting that with the advancement of technology, the significance of freedom of speech might have significantly evolved as well. Back in 1792, a man in Iowa had to be otherworldly charismatic or educated to have his words travel out of his town and get to people all over the country, through either word-of-mouth or handwritten letters (telegraphs would be invented a bit later in the 1830s). Today, that type of “speech” would almost be considered as “muted” in a highly connected world running on social media and, in general, the Internet. In the meanwhile, if Kyrie was standing on top of an unknown mountain in Upstate New York by himself and shouting out statements like Hitler was helping the black people, would that make any difference? If so, is reception part of the definition of “speech”? What about the “echoes” of the speech then? Therefore, when one is citing freedom of speech as defense or, occasionally, offense, it might be interesting to investigate the nature of that assumption and the overall mindset.
I probably went a bit too far here, but it’s indeed upsetting and worrisome that, with the recent cases of Kanye West and Kyrie Irving, both insanely talented at what they do, antisemitism seems to be making another full comeback and this time — it is taking advantage of the toxic Internet culture and dragging another traumatized community into it (although Black Hebrew Israelites have existed for decades).
Getting back to our discussion, how will the Mill hypothetically examine this in a different way?
What’s the Mill running on?
We are all familiar with the classical shield and spear paradox: if we have a spear that can cut through any shield and a shield that can resist any force, what happens when the two meet?
The paradox finds its origin in Han Fei Zi, written by Hanfei, an ancient Chinese philosopher from 300 BC that belonged in the school of Legalism (if anyone is interested in understanding why China is the way it is today, knowing that Legalism, not Confucianism or Taoism, is by far the most influential school of thought that shaped its history might be a good start). In the original text, the paradox is resolved simply by showing that the premise is false — such spear and shield would not both exist (不可同世而立 means those two things cannot co-exist). It’s also where the word contradiction is from, which literally means spear and shield in Chinese. In reality, situations that somewhat resemble this paradox, in much softer versions, of course, often end up in dilemmas. Back to Kyrie’s case, everyone wants and believes that they have autonomy over their bodies. As J.S. Mill puts it, it’s “one of the elements of well-being” (Mill 1859/1975, ch. III). However, when one’s autonomy conflicts with another person’s rights, simply telling either that their spear or shield is an illusion might not be an optimal resolution.
The Mill tries to avoid that by doing the next best thing — simulating the battle between the spear and the shield through computation. In the following sections, I’ll explore the basic structure of a Mill solution and briefly demonstrate how it may hypothetically work in Kyrie’s case. Ideally, a Mill solution would go through four stages: research, conversion, computation and projection. I am skipping the research stage, where input material is rigorously gathered and cleaned up, as it implements a common searching approach that extracts information on/off the Internet.
As I mentioned earlier, the Mill prefers to rely on low-level syntax, instead of semantics, as raw feed, although the dichotomy is not always so absolute. Semantics bring in emotions through associations and emotions are notoriously difficult to quantify. Semantics are also subject to interpretations and become, by nature, indeterministic. Therefore, in the initial stage, the Mill must prepare the input and convert the semantic elements into atomic low-level representations on a best-efforts basis. It is as intimidating as it sounds. As a matter of fact, conversion is indeed the most challenging part of the Mill process, comparable to data preparation in machine learning projects.
I am still in the early phase of developing and formulating the general conversion strategies. As of now, deconstruction (similar to that of the traditional model) and maxim-based mapping are the two key modules. The maxims are predefined and set as the fundamental values for all system operations. The great thing about using maxim sets is that they can be different from one Mill instance to another. For example, the maxims for Dawn, our own resource manager, are designed with commerce in mind:
- Benefit the advancement of civilization.
- Improve living standard and general genuine happiness.
- Protect the environment/ecosystem.
- Contribute to society-wide fairness, equality and freedom.
- Value honesty, quality, aesthetics, diligence and responsibility.
Obviously, this maxim set won’t work in Kyrie’s case, which requires a different set that centers on race and freedom. Let’s mock up one quickly just to see how conversion works.
- Do no harm.
- Speak no lie.
- Discriminate no one.
- Satan worshippers are damned.
- Earth is not flat.
- Kevin Durant plays good basketball.
I am throwing in an axiom (5) because Kyrie Irving is an avid Flat Earther and it just blows my mind every time I think about it. There’s also a fact (6) since Kevin Durant, Kyrie’s friend and teammate, happens to be one of my all-time favorite players. Both 5 and 6 are introduced under the Rule of Assumption to extend the scope of a native maxim set, which plays a big role in the rest of the proof.
Once we have the maxims ready, we can deconstruct the subject matter exhaustively until elements can be represented by maxims with basic logic rules, i.e., double negation, RAA, etc. Note that the subject matter here is a set of semantic objects that are “unpacked” raw information from the research stage, such as the tweet that Kyrie posted, which is essentially the entire content of the movie and the book it is based on, and a statement that he posted the tweet. I’ll skip the actual lengthy process and get to the part where we can show, among others (please note that this is a mock):
- Kyrie Irving believes that the earth is flat.
- Kyrie Irving believes that the Jewish people are damned (deduced from “the Jews are Lucifer worshippers” from the movie/book)
After the subject matter, a messy bunch normally, is deconstructed into tiny bits, those bits will be translated into symbols through mapping. As you could guess, the symbols are largely numbers that will go to work in the next stage of computation. This process corresponds with that of GEN ai. In fact, GEN ai and the Mill share the same root — and that’s not me! They are both applications under Leibniz, a general-purpose framework that transcodes any mundane object into computational representations and provides the universal algorithm for mapping. While GEN ai focuses on the cognitive aspect, the Mill is built for situations that involve moral judgements and ethics. An oversimplified way to differentiate the two would be that GEN is for the intrinsic values and the Mill extrinsic. Apparently, it is named after Leibniz as it was his idea and ambition back in the 17th century and he would absolutely do the same if he were alive today (and probably have already succeeded).
“The only way to rectify our reasonings is to make them as tangible as those of the Mathematicians, so that we can find our error at a glance, and when there are disputes among persons, we can simply say: Let us calculate calculemus, without further ado, to see who is right…”
— Gottfried Wilhelm Leibniz in a letter to Philip Spener, The Art of Discovery 1685, Wiener 51
However, that spirit is not the only thing I am borrowing from him; the framework also depends on a key concept from his genius mind: the identity of indiscernibles. The Leibniz system is only valid if we can assume that the mundane objects are all unique for otherwise the representations are simply meaningless and useless (I am excluding quantum mechanics here due to lack of knowledge). Think about this, if Kyrie Irving cannot be represented as a different being than Steph Curry, who’s also a top-tier point guard, a champion and teammate with Kevin Durant, then it is a fatal system failure. Leibniz is the ultimate solution to the Trinity Problem introduced a while back, a personal agenda that is (unfortunately) far beyond my capabilities. I’d love to elaborate on Leibniz another time, but, for now, let me get back to the Mill.
As part of the design, the mapping process usually leads to multiple translations because we have to take in account entropy and chaos; each of them is of its own specific significance based on the underlying values. Let’s say that after mapping we have:
- Kyrie Irving believes that the earth is flat. => F(23), V(441), 7
- Jewish people are damned => B(149), Q(V(23)), 91.19, 5, 5783568
Here, symbols like F and V are functions with parameters, and numbers like 7 and 91.19 are constants.
Now, we can move on to computation.
What do we actually compute in this stage? Do we apply certain magic aggregate function to the F, V and constants we’ve assembled in the last step?
The answer is no. These inputs prepared by Leibniz are generic; to be able to compute them for the Mill we’ll have to bring in the utilitarian approach.
Utilitarianism is a school of ethics that emphasizes happiness (“utilities”). Generally speaking, if an action promotes happiness, then it is right; if it produce unhappiness, it is wrong. The actor (performer of the action) is not the only source of happiness or unhappiness that is measured; everyone else’s should be taken into consideration. For example, by tweeting the controversial film, Kyrie Irving’s happiness increased, but it caused more unhappiness among the Jewish community. These utilities are the numbers we can use for computation.
As for the utilitarian approach itself, I’m following J. S. Mill’s doctrine and expanding the dimensions of happiness. According to Mill, quality of the pleasure matters; higher, intellectual-level happiness is much more valuable. When an addict takes drugs, he obtains a moment of physical pleasure while the lasting pain that it brings to both him and his loved ones is many times more.
Although the utilitarian approach sits at the heart of the Mill, other influences are consulted as well during computation, from the mandatory categorical imperative (how can you exclude Kant in anything regarding ethics?) to modern thinking on technology, environment and animal rights. Altogether, they are internally identified as the “Counsel,” though I am hoping that it will turn into a collaborative project, at some point, and, eventually, an independent research-based think tank unit.
Now the last question is naturally how we apply the Counsel’s wisdom in the computation process. Since we receive Leibniz formatted results from the previous step, the Counsel has to be Leibniz-ready as well. It may sound counterintuitive as it seems more convenient just to measure happiness directly from “facts.” However, again, semantics are hard to work with. Imagine subtracting a public embarrassment from a national sensation and calculate the utilities from the remainder. Another main reason is on the practical side — it largely improves manageability as transcoded Counsels can always be reused in other cases, and I can also update or synchronize them independently whenever I see fit.
The last stage is projection.
If we think of the conversion stage as a process to break semantics down to lower-level representations, then the projection stage is about building them up through logic again.
However, up till now, we are leaving out the key component of any proof: the hypotheses or — what are we computing for? Normally, a scientific method will have hypotheses set up as the first step to be accepted or rejected later. The Mill has a slightly modified methodology in which setting hypotheses is postponed till the projection stage. Why is that?
Other than the aforementioned semantics-related reasons, this scheme allows greater flexibility and tolerance, both important for ethics. Of course, it is also viable to lay down hypotheses in the beginning and update them later based on computation outcomes. What matters is that the hypotheses need to be individually definitive but, collectively, diverse. Ethics are ambiguous and humanity is complicated; it’s practically impossible to have simple black-and-white answers in most cases, but our biases and ego predispose us to seek them, relentlessly. These flaws of human nature and the lack of awareness thereof are what the Mill is being built to address. I’ll circle back to this point in the next section.
Different from the common last part of a standard methodology where test results are reported, fitted and concluded, the Mill’s projection is dynamic and, even, chaotic. In some way, it is analogous to tangram, a dissection puzzle game, consisting of seven flat polygons which the player needs to put together to form shapes. If we see the computation results as the polygon pieces, then the hypotheses are specific shapes one would like to form. In simpler terms, if you are keen on receiving 2 on the right side of the equation, the left side cannot be 0 + 0 as the law of addition is canonical. Part of its own integrity test, the Mill has to make sure that patterns can only be successfully formed if and only if all the polygons are in the right shape and placed correctly. Granted, the same set of polygons can generate a large number of different patterns, including contradictory ones, but that’s exactly the nature and power of ethical debates. When all raw input objects are exhausted, getting paradoxes is just part of the process or, even, a helpful system signal to call for an update to either the maxims or the hypotheses, or both. It could also mean that we do not have sufficient valid input objects; more research has to be done. In this way, projection implements recursive reasoning (programming).
In Kyrie’s case, first we may set our hypotheses to (make sure the hypotheses cannot contradict with each other):
- Kyrie is antisemitic.
- Kyrie is wrong.
- Punitive measures shall be taken against Kyrie.
Then we’ll see if we can assemble the computation outcomes into these assertions. Bear in mind that the semantic terms must all have had their symbolic translations already. If they can be successfully produced, then we can pack the input, maxims, hypotheses, conclusions and metadata into a final Mill case object. These objects can be reviewed, evaluated and replicated; more importantly, they can be referenced in other cases, e.g., is it racist to dislike Halle Bailey as Little Mermaid, or put together to test larger ethical simulations. However, if, more likely than not, the proof fails (for example, we cannot prove Kyrie Irving is antisemitic because Counsel says tweeting a questionable tweet does not equate with being antisemitic), then again we have to go back to the beginning and reexamine the maxims, input and hypotheses.
The Mill’s workload may seem overwhelming because it truly is; the good news is that it runs on computers and computer science, as a field, is advancing rapidly. While the machines do provide unprecedented computing power that makes multithreading, multivariate and multi-conditional processing effortless, that’s not their only thing the Mill is taking advantage of. What I consider and believe as the real strength is within logic itself fundamentally, which gives logic-based programs a better chance at reducing biases, detecting fallacies and avoiding logic traps. I’ll elaborate in the next section soon.
Why the Mill?
To put it simply, the motivation behind the Mill project is twofold: my own anxiety and the necessity for Dawn.
I’d always thought that despite making a fair share of questionable judgements in my youthful days I was a fairly logical and rational person. It had a lot to do with my indolence, which I rationalized as a way to preserve energy for something that could potentially happen one day when I needed to deliver in full force. It’s a terrible excuse, but, as a result, I would take more time to research and think, instead of rushing to conclusions, as that, in my head, requires much higher energy. However, over the past few years, I’ve started to have self-doubt, and that, to not be able to fully trust my own ability to think, honestly frightens me. Following Descartes’ idea, if I cannot be sure that I can think, then what is this? It may sound a bit exaggerated, but I’ve been genuinely questioning:
- Have I reached a point in life where I’d turn to heuristics too often?
- How many blindspots do I have due to my ego?
- How many views of mine are the way they are only because I am somehow influenced by others or the “narrative”?
For example, during the pandemic, I was mad at the anti-vaxx people. In my mind, it was so outrageous that even trying to understand their stance would be a ridiculous idea. Now I am going to get my second booster next week, but part of me does wonder if I overreacted — in the end, I had no right telling another person what to do with their body. The only way I could possibly justify my fury was the potential consequence of overcapacity at hospitals, a public resource, as a result of more infections.
How did I have such a categorical reaction on a complicated social issue? Besides the stress inevitably caused by the pandemic itself, in retrospect, the Internet (and the social media) played a huge role. It was awfully divisive — you felt like you had to choose from two polarizing camps. It was also aggravating — the information coming your way would only add to the growing animosity toward the opposing camp. By the end of the day, a legitimate debate between the scope of personal freedom and social responsibility, as I see it now, turned into another full-on pseudo-ideology war between the “left” and the “right” that brought insanity, vindictiveness and vileness out of human nature. Unfortunately, although called out by many, it is still the common recipe for the Internet and will likely remain so for a while — because it works, effectively. I previously argued that the whole “AI” and algorithm business, now supported by the thriving field of big data better than ever, is not turning machines humanlike as their fabulous pitch deck says, but the other way around. Huge data sets, harvested unethically, are fed into algorithms, which are essentially probability functions, to generate predictions that are then used to direct user behavior and impact their cognition. Human mind, sadly vulnerable and flawed, is no match for such machine power. With algorithms designed with the main goal to drive traffic and generate clicks, users are more likely to receive information that either gets them defensive or offensive, especially on political and/or controversial issues on ethics (race, gender identity, etc.).
But, it could get worse. Since the probability-based algorithms apparently adhere to the majority rule, it makes consensus marketable because in the age of the centralized Internet “majority” can be paid for. We’ve seen the tip of the iceberg over the past few years, from organized armies of bots to click farms in developing countries. The programming is usually uninspired but efficient; it seeks the more susceptible targets among all and exploits their emotions in exchange for behavioral change (many theories here, such as cognitive dissonance). When such manipulative programming of hired influence invades the public discourse, it drives people to extremes; when it goes to private spaces, it jeopardizes psychological wellbeing.
Therefore, to a large extent, I am building the Mill for myself, a skeptic, on one of those days in the future when I can no longer trust my own moral compass. Mind you, I am not trying to be a saint that follows certain categorical moral principals, as Kant postulates. I honestly don’t think that it is possible; human nature that puts ego first does not allow it and Kant was seriously a racist. What I want to achieve is when my decision or judgement is made out of greed or selfishness I can be made fully aware of those biases (the maxims would reflect exactly that).
The other main reason for making the Mill is that Dawn requires a moral authority to regulate resources. As I explained above, while big data provides unparalleled market insights, it is unreliable and potentially hazardous. Its compatibility with a decentralized system also remains questionable. With Dawn and LOVN (edited due to systemwide renaming, originally LOVA, 11/20/2023) growing, the new market channel, born to be the interface between these two independent structures that represent resources and consumers respectively, needs a valuation mechanism to replace the role of “popularity” in the traditional model, one that cannot be abused by capital and power. This urgency also goes beyond the Internet, a microcosm of our society. While we still surely appreciate others’ kindness and loyalty as individuals, the society, as a system, does not reward values like these, but more often the opposites. Thereby, narcissists become champions, deception prevails over honesty, and greed trumps conscience. Building Dawn is a chance to challenge that from the market perspective and, the Mill, acting as the authoritative mechanism to manage rewards, is the soul of it all.
The Mill for the Dawn system will be detailed further in the next post when I discuss the LOVN app (edited due to systemwide renaming, originally LOVA app, 11/20/2023) and how the embedded Renaissance works for data authentication — I finally found a silly way to code Renaissance Seals into paintings. The LOVN app (edited due to systemwide renaming, originally LOVA app, 11/20/2023) is an important piece of the general infrastructure of a decentralized ecosystem that recognizes and honors intrinsic values. It also packs a proximity-based AR-rendered social networking dimension, Euler.
Right, another madman in history.