Readings for 10/17/18
Readings are up for next week :)
Following up on class discussion yesterday, here is the paper that I thought of assigning but was concerned about overloading people:
https://www.microsoft.com/en-us/research/wp-content/uploads/2016/05/Bishop-MBML-2012.pdf
This is not assigned, but it's an excellent paper and feedback is welcome.
Also, someone asked about Gaussian processes yesterday. In thinking about the whole context of that discussion, I thought it might be helpful to recap some terminology.
A random process is something that creates non-deterministic (uncertain) outcome
Examples:
Following up on class discussion yesterday, here is the paper that I thought of assigning but was concerned about overloading people:
https://www.microsoft.com/en-us/research/wp-content/uploads/2016/05/Bishop-MBML-2012.pdf
This is not assigned, but it's an excellent paper and feedback is welcome.
Also, someone asked about Gaussian processes yesterday. In thinking about the whole context of that discussion, I thought it might be helpful to recap some terminology.
A random process is something that creates non-deterministic (uncertain) outcome
Examples:
- Rolling dice (fair dice or not)
- Flipping a coin (fair coin or not)
- Random sampling
Counterexamples
- Flipping a coin with two heads (because there is only 1 outcome, it is deterministic/certain)
A random variable maps the outcome of a random process to a number.
Example: Flipping a coin
- Random process: flipping a coin
- Random variable X where 1 is heads and 0 is tails
- The letter X is arbitrary
- The numbers 1/0 are arbitrary but conventional
Example: Sampling age
- Random process: randomly sampling
- Random variable X where age is a positive nonzero integer (e.g. 19)
Sometimes you will hear terms like Gaussian process, which does not refer to a specific random process (like flipping a coin) but is rather a category of random processes that meet certain criteria.
Frequently you will hear Gaussian process in the context of regression, i.e. Gaussian process regression. In this approach you assume that your data has been generated by unknown functions with noise (the random process) and the Gaussian process regression assumptions (which include several things like mean, covariance function/kernel, etc) allow you do define a probability distribution over these unknown functions, at which point you can do some Bayesian inference. In other words, the unknown functions are a random variable, and you estimate the probabilities of these functions using your data and your assumptions.
Debes (2018):
ReplyDeleteIn this essay, Remy develops a dynamic, dialogic theory of interpersonal understanding in light of developments in social epistemology. I’d like to go back to something that was mentioned a few classes ago as something unambiguously good. Predictive policing initiatives are effective in reducing crime rates. I don’t think anyone thinks that is a bad thing. However, there can be problems with overpolicing minority neighborhoods and proliferating racial profiling. Has anyone asked the locals—in the dialogic sense of this essay, in the sense of understand-ing their perspectives—about their perspectives on such initiatives? Are the poor problems to be managed or citizens to be listened to? Why isn’t big data and machine learning being used to solve some of the empirically-verifiable causes of crime—like social inequality? Is the answer more police, or more socially equitable policies and initiatives? In face of the retreat of the latter, the former becomes concerning.
Debes (2017):
This gives a brief historical overview of the concept of dignity. Does technology—big data and AI being the latest iteration, the one we’re interested in—benefit or harm society and individuals? Does it uphold human dignity, or does it attenuate it? When the English Luddites of the 19th century destroyed the textile machinery that was being introduced into their trade, it was not because of some irrational technophobia. It was because that machinery was being used, in the name of progress and industry, to implement exploitative labor practices. The question is not whether technology, or any particular piece of technology, is inherently good or bad. That is a senseless question. It is a question of how the institutions, classes, and individuals in power are using that technology in particular, historically-embedded contexts. There is no blanket judgment to be made, only local ones. How are particular uses of AI and big data attenuating or supporting human dignity?
Debes- Understanding Persons and the Problem of Power
ReplyDeleteThe framework for understanding other people first consisted of two problems and now has added a third problem, “the problem of power.” Oppressed people must makes sense of their oppression but often do not have the means to do this task. Then, the powerful make it harder for the oppressed to make sense of their oppression because the oppressed are called into question in regards to being “knowers” and their knowledge is often not treated as knowledge. This concept is fascinating to me because if we could answer some of the overarching questions here, we could potentially reduce conflict on all levels. The idea of “Can empowered people ever genuinely understand oppressed people? (p. 56)” is remarkable. Conflict is often viewed as negative; however, I believe that conflict doesn’t always have to be negative. Conflict that promotes understanding and ends with both parties better off is, to me, not a bad thing. If we could answer the question above and find a way to understand others, our conflict could lead to improvements. On page 58, the author describes how “we typically try to settle by appeal to ‘facts’. . . these facts depend on what we make of them --what ‘we’ understand ‘our’ social world to be.” The author goes on to distinguish a difference in understanding a person’s experiences and understanding those experiences as inherently “his” or “hers” or “yours” or “mine” and how this is different than claiming that you just understand. Honestly, I think understanding others could become a black box in the sense that how this process works could be unknown and impossible to understand. When we think of all the different facets of communication (both verbal and nonverbal) it is no wonder that we have trouble understanding others, but that doesn’t mean we shouldn’t try. The last sentence of this chapter really left an impression on me and, I believe, really captures how to address the problem of power when trying to understand. We understand others when the other specific person feels understood.
Debes- Intro to Dignity A History
I did not realize that dignity did not always mean what it means today. It is interesting that there is a stark difference in the historical context of dignity as a social status while the modern day concept appears as a moral status. Also interesting is the two types of claims we make about dignity: substantive & formal. I have found that not having a clear, concise argument free definition of a particular word makes it that much harder to describe and study it. I think this is the case of dignity especially since we still use it to have multiple meanings today (i.e., human dignity vs. dignitaries). I am interested to see how Debes ties this to Big Data and/or AI. How does the concept of dignity play into these categories?
Debes (2018) describes the hermetic injustice between the oppressed and the oppressors. Introduces the problem of power. It makes clear that beliefs are bits of cognition and we never perfectly understand anyone, but sufficiently understand a particular person here and now. Debes (2017) is the Introduction to Dignity A History book. Defines dignity as the fundamental moral worth or status belonging to all persons equally. Describes a history of dignity in its moralized sense in 4 platitudes.
ReplyDeleteThis comment has been removed by the author.
DeleteThe history of dignity is complex, and extends far beyond the etymology of the word itself. Artifacts of the evolution of dignity can be found in plain sight. For example, for someone to be “dignified,” the elicited connotations of meritocracy and social esteem more closely resemble the vestigial Roman sentiment rather than the modern one. Integrating the present and extrapolating the future of dignity within the context of this course, I think it might be useful to ask what requisites of dignity are applicable to humanity specifically, rather than to intelligence or rationality itself. It doesn’t seem like respect for autonomy, as Macklin is quoted as saying, because we like our tools (i.e. AI) to be autonomous, and in fact exploit their autonomy as long as the scope of their autonomy is limited, especially if they are sufficiently “rational” as probabilistic decision-making tools. I think this goes back to Kant’s original assertions regarding other humans as ends in and of themselves rather than as means to any end, but then the subjectivity of what can be a means or a tool (the approaching equality of a general AI and human workers) seems to raise suspicion. In short, is dignity an inherent trait of humanity or rationality? I think it’s almost certainly a false dichotomy, but I’m not sure how to further pull it apart.
ReplyDeleteRegarding the chapter on understanding and power, I’m curious about the implicit contradiction in section 3.5.1. It seems that one of the best things we can do as fellow humans is to try to understand each other, especially in attempts to transcend power inequalities. However, to completely understand someone, at least implied here, would be to minimize their personhood since you could essentially speak for them at any given moment with comprehensive understanding of them or their perspective. So then should we just be glad that we can never seem to reach understanding across power gradients while maintaining the goal of total understanding? Aside from this, I would like to point out how these considerations of understanding speak to the need for representation of marginalized groups in politics and beyond.
Debes- Intro to Dignity A History
ReplyDeleteThis chapter introduces the history of dignity. I am interested in the modern day concept. Just as the writer said we don't have a unified concept of dignity. What concept might correlate more with the AI? Personality, integrity, or self-esteem? In other words, AI would cause what kind of dignity problems?
Debes- Understanding Persons and the Problem of Power
With the development of AI, I think there may be another kind of problem of power, people know about AI and owe AI products vs. people know little about AI and can't afford AI products. I watched a TV series Humans in which lifelike humanoids cause a lot of social problem. I am curious whether AI products can help us understand others. And, how to make AI products better understand people's ideas and needs?
Debes in Dignity A History, describes how the concept of dignity has evolved since the 16th century to the present. He shows us definitions from several dictionaries. I think this article is especially relevant in areas of the world where human rights are not respected as they are in the United States. I didn't know that Kant owed part of his ideas of dignity and human rights to Pufendorf. I would like to see Debes offer up his own definition of human dignity.
ReplyDeleteDebes Understanding Persons and the Problem of Power tries to explain how oppressed people often lack the power to free themselves from their oppression. Debes inquires about the nature of the problem of power and its solution. I agree with Medina that those in power should be open-minded and receptive to the views of those oppressed about what it's like to live without power.
Davis
Debes (2018)
ReplyDeleteThis article was very interesting, and had what I believe to be very relevant comments about oppression and power. The section on page 58 about oppressed voices called to mind the various opinion pieces you see that involve an empowered individual taking on some characteristics of an oppressed minority (male signing emails with a typically female name, a white woman wearing a hijab, etc) and "pointing out" the hardships they face in these instances. Even if the meaning of these articles are good, it still shows that we only believe the struggles that the oppressed face when it is those in power who are speaking about it. While I do not know the best way to teach this kind of empathy, sympathy, or basic understanding of others that is highlighted in this reading, I wonder if not shying away from our history of being a nation of oppressors is one potential solution. Instead of trying to sweep dark history under the rug, teach it in schools from the perspective of those who lived it, and instead of hiding from our past, helping us to overcome it.
Debes 2017:
Like others have mentioned, I too did not realize that the definition of dignity had changed as much as it had throughout recent history. I have noticed, however, that people today will sometimes use it more closely to its original meaning, stating that, for example, they have too much dignity to do a job that heavily involves manual labor, essentially belittling the individual who already has said job. The same individuals who are too "dignified" to take simple cashiering jobs become outrageously offended by concepts like self checkout machines that are "taking away people's jobs," and I often wonder if they see the hypocrisy in their statements.
Debes (2017):
ReplyDeleteDebes argues/points out several times that this volume aims at focusing on the “history of dignity in its moralized sense.” I think there is a lot of value in walking us through the history of the different meanings dignity has taken on to make a point that it will probably continue to change as time continues to evolve. I also like that he points out that the volume does not aim to “defend any given theory of dignity [but rather] aims to analyze past theories and their defenses” (p 5) In regard to AI, dignity is probably put in a context different than distinctions pointed out in this writing, of which I am curious to hear about. I guess when talking about big data/the world of AI, I never voluntarily put “dignity” into the same category. So reading this particular article made me really step out of my narrow-minded way of thinking and consider the connection.
Debes (2018):
Speaking directly to understanding and power, I think Debes makes some bold claims. He definitely acknowledges some hard question in considering if empowered people can ever genuinely understand oppressed people and even what it is to understand a human. The section discussing truly understanding the perspective of another individual was very thought-provoking to me, but held some solid insight—of considering a person’s position, rather than just understanding their claim or leaving the person altogether out of the understanding at all. So in somehow connecting this to AI, I wonder are those that are oppressed, underserved, marginalized, etc. unable to understand those in power because of the gap of knowledge they cannot fill in? and vise versa for those in power—is it a lack of truly understanding those that are oppressed that makes a void in using the technology of AI to solve some of the problems present in the lives of the oppressed? Or is there an alternative problem/void at hand?
Intro to Dignity (2017)
ReplyDeleteI've heard of multiple words that have drastically changed meaning over time, but this is the most in-depth history I have read about any word meaning alteration. I did not know that dignity had a different original meaning, and I do agree with what we currently define as dignity as a basic human right. Thinking about the future of dignity, would we expand the definition to include non-human creatures or machines? Particularly, does anyone believe that someday robots or AI technology will have enough autonomy and even consciousness to possess their own dignity? I'm sure this is/will be a big debate, not only about dignity but about giving AI all types of human rights.
Understanding others and the problem of power (2018)
I think it is important for us to realize that it is often extremely hard, or near impossible, for oppressed people to get out of oppressive situations. It reminded me of something I was reading a couple nights ago about how even people who aren't oppressed may not leave sub-optimal conditions for an improved condition due to cost, effort, and wanting to stay in their comfort zone rather than evoking change. If a person if oppressed and doesn't have much power to free themselves, it would feel hopeless. I'm curious as to how AI, big data, or other forms of technological advancement could be used to empower people who are oppressed, teach others about the struggles of oppressed people, and ultimately just improve peoples' lives.
Debes (2018)
ReplyDeleteDebes makes an interesting point about understanding a person that I don’t quite agree with. He proposes that when we claim to understand utterances, actions, beliefs, feelings, etc of other people that this is simply “understanding-about” a person, which he contrasts with a “holistic understanding”. While he makes sure to state that holistic understanding doesn’t really entail “total understanding”, he defines it as perspectival understanding, that holistic understanding involves having a person’s point of view in mind and understanding that. I fail to see how this is different than an instance of “understanding-about”. After all, isn’t a person’s perspective or point of view just an instance of or collection of beliefs and/or feelings? It seems as though what he’s trying to get at by defining a “holistic understanding” of a person is to know what it means to understand the person rather than a specific instance of their actions, beliefs, etc. I would propose to think about this a bit differently in what we could call “predictive understanding”. I think what makes a person who they are is the whole collection, or sum, of utterances, actions, beliefs, feelings, etc., and in line with that, we could define the degree of understanding one has of a person to be the degree to which one is able to correctly predict that person’s utterances, actions, beliefs, feelings, etc. This is similar to what is, in philosophy of mind, called mindreading, but entails more than just mental states. He makes one more point on this understanding of a perspective that I find difficult. It does seem prudent to point out the importance of taking into account the person’s “take” on the facts if one is to understand someone. I fail, however, to see how this is different than taking into account the person’s beliefs. Isn’t one’s “take” on the facts, simply their belief about those facts? Indeed, I think that if one understood the other person’s beliefs and even, in his example, had those same beliefs, that seems to pretty much include the person in the understanding. The problem is that he seems to want to define “holistic understanding” in some ethereal way that escapes simply being able to state or predict the other person’s utterances, actions, beliefs, etc. Either that, or he is simply getting at empathy, but empathy seems, to me, to be even more shallow because it only entails feeling what someone else feels rather than understanding all of the other aspects such as actions and beliefs.
Debes (2017)
I like the historical review of the moralized and social status/merit meaning of dignity in this article. He pointed out that while conceptions of dignity have mostly shifted to the moralized meaning, we still use the merit meaning in certain context such as “where is your sense of dignity?”. One way I thought to make sense of this is the following: whatever substantive claim we make about dignity, everything that has those properties does have dignity (moralistic meaning), if everything with said properties does indeed have dignity, they ought conduct themselves in a manner becoming of a thing with dignity (poise, gravitas, beauty, grace, integrity, etc), and to do otherwise is to display to the world one’s degree of disbelief that they themselves do have dignity (merit meaning). After reading this article though, I’m quite curious how he goes on to prove the platitudes (which seem a lot like historical origins of the respectively contemporary descriptions of dignity) false or highly exaggerated. They all seem, to me, to offer some modicum of truth or insight into how we currently conceptualize dignity (which he did say would likely be the case). Now I would very much like to read the main article.
This comment has been removed by the author.
DeleteI am curious though as to what the tie in to big data and AI will be from both of the concepts in these articles. On the AI front, I suppose we could conceive of trying to understand an AI in the manner laid out as similar to understanding persons of a different power class, etc. Also, some of the following questions seem as though they could stem from the concepts discussed, such as "what level of AI do we need to consider those systems as having dignity?", "what level of understanding of an intelligent system do we need to make the call in the previous question?", or "if we don't fully understand an intelligent enough system and don't view it as having dignity, will we be prone to making some of the past mistakes of oppression and what would that power structure look like?" Certainly there is some substantive claim about dignity that fits humans that will likely someday also fit an artificially intelligent system.
DeleteDebes – Intro to Dignity
ReplyDeleteSome interesting information was presented here regarding changing views of dignity throughout time. While I respect the questioning of definitions and implications of dignity, I am much more in line with their description of Macklin’s view that it “is a useless concept. It means no more than respect for persons or their autonomy” (p. 9). That is not to say “dignity” is not an important concept, but that the word itself is not of importance, rather our attempts at preserving or protecting those things which we currently view as dignity, such as respect for individuals’ autonomy. I also see a bit of strength in the Marxist arguments (though, I’m much more fond of Pierre Bourdieu, myself) in how dignity may serve as a means to appease the subjugated. Afterall, if dignity involves any element of autonomy then certainly being forced to endure undesirable workplace conditions—whether in the form of listening to your boss drone on about mission statements, working in unsafe and physically harmful conditions, or having to perform acts viewed as “dirty” by society (e.g., cleaning bodily fluids)—is a prime example of a threat to this, tied in very closely with marginalization, as those groups viewed less favorably are more often forced to accept unacceptable conditions.
Debes – Understanding Persons and the Problem of Power
This chapter raised some interesting points about power but seemed to suggest that discussion was the best means to achieve understanding, while maintaining that understanding can never be achieved. I argue that instead of pretending we can ever understand another; a more socially useful metric is prediction. We are at an uncomfortable point in society, where we can predict behavior somewhat reliably. Note, that this is absolutely a function of the complexity of the issue. For example, we can predict that if someone likes A, B, and C, they will probably like X (e.g., in music preferences), though there is less certainty. If we have a prediction with a much larger space, prediction becomes almost impossible (such as predicting who you will go on to marry, if you should go on to marry someone even). However, we can make fairly reliable predictions in certain domains that often hold true (e.g., politics). That is, big data, psychology, marketing, and many other approaches do support that “that the mere accumulation of descriptive facts about a person’s experiences can suffice for understanding her” (p. 65). While some can argue that this is not understanding them, I would argue that we can know things about people that they do not even know. This certainly is influenced by changes over time, as discussed in the article, but I would argue that we cannot understand anyone in the purest sense, as seemingly echoed by the chapter, but relying on their subjective interpretations of who they are and why they do what they do, regardless of dialogue, is not a fruitful endeavor for anything beyond forming a better relationship with that one individual.
A polarizing, but hopefully light hearted, example was a discussion with a friend about leggings/yoga pants (I am not sure the full taxonomy of these clothing-types). They argued that their decision was not influenced by trends, style, fashion, etc…, but by comfort. I then asked them what they thought about wind pants and they shared my beliefs that wind pants were one of the comfiest trends in clothing we have ever experienced. YET, wind pants are not commonly worn, certainly never by this individual. That does not mean she is “a phony” or anything, just that outside influences shape us in ways that we don't recognize or like to admit. My argument is not to dismiss anyone’s opinions, but to suggest that understanding will never occur, and is muddied by biases, heuristics, and incomplete insight so, instead, we ought to focus on how we can see beyond the subjective into the objective.
I completely agree on the point of understanding people. In fact, I think in an attempt to dismiss knowing descriptive facts about a person as a method to holistically understand them, Debes actually talked about those same descriptive facts just in different words. I think the prediction metric is probably the best metric by which to claim understanding, especially if you also understand the mechanism of prediction. I mean, to back up the assertion that we sometimes know things about people that they don't know, I see plenty of proof in daily life, where both friends and loved ones often know exactly what one needs in situations of distress even when one might be willing to do something that won't benefit them or is destructive, and in neuroscience, where, by reading brain activity, we can actually know when a person will act before they even consciously know it themselves. I also agree that any full or true understanding of another person can never actually occur because that would likely require actually having the same biology and experiences and hence, the same brain, as that person as well as perhaps even knowing the things they cannot know about themselves.
Delete