A Budding Researcher in AI

Hello, my name is Isabell and I'm currently a student at the University of Michigan. I've been working under a research professor in the Center of Complex Studies. This blog will host a majority of the work that I have done over the past year within the UROP program, along with other technology centric research related topics.

[isaresearcher]

What Kind of Ethical Implications are Associated with Disclosive Ethics? (Part 2 Social Movements)

Online Social Movements

Online social movements, stem from political mobilization efforts that started appearing on the internet in 1996. These organizations in political arenas, used online tools of mobilization reach voters who were outside the current efforts being used to mobilize voters (Bimber 1998).

Online social movements, are built on the foundation of instigating social change though the act of creating an online environment that shifts the power to underprivileged networks of individuals suppressed by a social hierarchies through the collective validity to create an advantage for the suppressed group expressing societal concern (Arquilla 2001). Online social movement structures have framework that works similar to embedded values within a system. The structure works as a support system to specific value set through promoting the need for an inherent social value within an information system and society or demoting the less desirable value (Floridi 2012).

SocialMovements.Black_Bodies()

Viral social media video clips circulating the internet displaying evidence of the racially injustice killings of African American individuals by authorities in American justice systems have increased significantly since the emergence of social media. News outlets and online social media platforms capitalizing on embedded values, such as justice and equality, distribute specific content to online users based on relevance. The embedded value of justice and equality act as the motivating factor for bringing attention to these stories in the online environment but these platforms are built from an economic standpoint to “commodify social life” also. In the work of Safiya Umoja Noble, an assistant professor at the University of Southern California, she discussed that these media platforms use emotional value to create more traffic to gain users, and this is highly prevalent when evaluating online content pertaining to African American police killings.  African American police killings increase traffic to these specific platforms and when the story becomes old news, there little done to address societal misdeeds of injustice in the social environments in which these events take place. The videos circulated are seldom used as solidified evidence in the court of law to convict wrongdoers and the value these viral videos have in changing corrupt systems only last as long as the news outlet endorses its presence online.

References:

Arquilla, John, and David F. Ronfeldt. Networks and Netwars the Future of Terror, Crime, and Militancy. Rand, 2001. pp.1

Bimber, B. (1998). The Internet and Political Mobilization: Research Note on the 1996 Election Season. Social Science Computer Review, 16(4), 391–401. https://doi.org/10.1177/089443939801600404

Floridi, Luciano. The Cambridge Handbook of Information and Computer Ethics. Cambridge University Press, 2012 Chapter 3

Noble, Safiya Umoja. “Critical Surveillance Literacy in Social Media: Interrogating Black Death and Dying Online.” Black Camera, vol. 9, no. 2, 2018, p. 147., doi:10.2979/blackcamera.9.2.10.

[isaresearcher]

What Kind of Ethical Implications are Associated with Disclosive Ethics? (Part 1 Algorithmic Implications)

Search Engines

Users of a search engines may have implied knowledge that search engine criterias have the embedded value of popular opinion within proactive search suggestions. The built in trend capturing algorithms implemented to accommodate these values have a tendency to “perpetuate the status quo” through data analysis driven by online traffic (Trevisan 2013). This has the potential to insert societies pre-existing biases into the search engine systems unconsciously through the act of a certain user’s interaction with the system within a given society. Although the organization never intended to perpetuate the pre-existing bias within the framework of the system’s creation, the algorithm of the system becomes decontextualized from the system user’s interactions, a concept known as emerging bias (Friedman 1996).

SearchEngines.Algorithm_Bias()

As a result, algorithms used in systems which depend on ranking, tend to deafen the voices of more alternative viewpoints in mass media through data analysis because they are created with the intent to bring more popular topics and websites to the top of the search algorithm queue. Most of the data points these systems are relying on come from humans with “biased data” learned from a specific society (Snow 2018; Cathy 2016). Users implicitly have knowledge about the value of popularization inherently built into these systems, but this knowledge acts as a moral vice on the creation moral neutrality in specific search engines. The implied morally opaque practice within the system becomes a new normal and this occurs through the act of the seeming absolute knowledge of the online practice of popular opinion.

SearchEngines.Algorithm_Protection()

 Since the growing concern in society for individuals to have access to certain information, such as issues pertain to self harm or outward hostility, companies in charge of creating and maintaining search engine criterias have put in an effort to curb access to harmful results without restricting the public's ability to search for keywords. Search engines have inherent set of conditions that prevent users from being able to obtain information and instructions through the act of flagging these keywords in the system as harmful and restricts the user’s ability to obtain content related to the keyword (Goldman 2008).

References:

Cathy O’Neil. Weapons of Math Destruc- tion: How Big Data Increases Inequality and Threatens Democracy. Broadway Books, 2016 pp.22-30

Friedman, Batya, and Helen Nissenbaum. “Bias in Computer Systems.” ACM   Transactions on Information Systems, vol. 14, no. 3, 1996, pp. 330–347.,  doi:10.1145/230538.230561.

Goldman, E., 2008. “Search Engine Bias and the Demise of Search Engine Utopianism,” in Web Search: Multidisciplinary Perspectives, A. Spink and M. Zimmer (eds.), Berlin: Springer-Verlag, pp.42-43

Snow, Jackie. “Bias Already Exists in Search Engine Results, and It's Only Going to Get Worse.” MIT Technology Review, MIT Technology Review, 1 Mar. 2018, www.technologyreview.com/s/610275/meet-the-woman-who-searches-out-search-engines-bias-against-women-and-minorities/.

Trevisan, Filippo. “Social Engines and Social Science: A Revolution in the Making.” SSRN Electronic Journal, 2013, doi:10.2139/ssrn.2265348.

[isaresearcher]

Computer Ethics: Disclosive Ethics

Disclosive ethics refers to an exploratory subfield of information ethics founded by Philip Brey in 2000. The methodology of disclosive ethics is utilized as an approach to discuss ethical issues that arise in perceived normative technology environments in information driven atmospheres of analysis. (Brey 2000). 

One of the key facets to disclosive ethics mindset involves the ability to approach perceived normative environments with a level of scepticism in order to expose the unperceived biases of the designers of a specific informational system. Designer bias that may have been knowingly or unknowingly embedded into the creation process of that information system, have the ability to shape the way a user interacts with a system which has the capacity of effecting societal cultural belief systems. This social implication is one of the fundamental reasons behind Brey proposal to create a subfield of computer ethics exclusively focusing on broadening the scope of moral analysis within computer practices to the issues that have not been exposed. 

Disclosive ethics is a deviation from the current standard model of applied ethics, which does not account for morally opaque ethical issues in perceived normative information systems and acts as a complementary approach in developing the limitations of the standard model of applied ethics.

foundationOne = DisclosiveEthics.Morally_Opaque_Practices()

Morally opaque practices (or the existence thereof in computer related practices) are a set of computational practices or values embedded within the design of a system that user’s of the system may be aware or unaware of in an online information atmosphere. This creates an environment of perceived moral neutrality on the front end of an online platform although this may not be the case on the back end. The existence of a morally opaque practice can theoretically be traced back to two probable causes within the initial design of a system (Brey 2000). The principle of the existence of a morally opaque practice within the design of a system is one of the foundational beliefs of disclosive ethics.

foundationTwo = DisclosiveEthics.Implied_knowledge() 

The extent of moral opacity related to this computer related practice is based off of implied knowledge or unknown knowledge that a front end user may have about the design of a platform. This morally opaque practice can act as a hidden practice although there may be a slight awareness in the user using the system. The user is left unaware of the logistics of the running practice.

foundationThree = DisclosiveEthics.Undetectable_Embedded_Practices()

A practice is built into the design of a system, through algorithms or a set rule of conditions, so that the user may appear to be using the system in a normal fashion but this normal fashion may be tainted when used outside the bounds of a embedded virtues in the design process that designers perceive as protecting a societal virtue. When a computer system has potential biases that are left unaddressed, intentionally or unintentionally, as the use of that system becomes more embedded into the standard of normal daily interaction within the world, the underlying bias becomes more persuasive as the collective norm (Friedman 1996). In this fashion, the system appears to be value neutral and no apparent or implied knowledge of the practice can be detected by the user. 

References:

Brey, Philip. “Disclosive Computer Ethics.” ACM SIGCAS Computers and Society, vol. 30, no. 4, 2000, p. 10., doi:10.1145/572260.572264

Friedman, Batya, and Helen Nissenbaum. “Bias in Computer Systems.” ACM Transactions on Information Systems, vol. 14, no. 3, 1996, pp. 330–347., doi:10.1145/230538.230561.

[isaresearcher]

Why do Bayesian Networks Work with Abstract Framework Analysis? A Philosophic Perspective (Section 1)

Bertrand Russell had envisioned the representation of abstract thought through the means of atoms, described as factual ground truths within a larger thought framework representing piece of a larger factual proposition, true or untrue according to an individual perspective. In one of Bertrand’s lecture discussing the concept defined as Atomism, the belief that a factual statement can be partitioned it into compositional logic segments of smaller factual propositions. Russell states that larger factual statements can be divided into segments with a simplistic approach but the smaller representations created from the factual statement are built with a level of complexity themselves (Russell 1918). Therefore an atom within it itself is derived from the existence of compounded fact observed to exist in totality, true or false. Obtaining a factual proposition within a factual statement as a point of absolute true or absolute false, has a level of vagueness associated with the construction of it. Russell deems this quantifiably of vagueness as irrelevant in the analysis of abstract thought. This idea of vagueness appearing as a relative probability of error falls along the lines of Quine’s argument that the existence of a discrete objectification of a factual truth can be defined by continuation of evidence and further add to the claim for the model of abstract thought that Quine envisioned within the “web of belief”. 

 Critique of theories and the characterization behind quantifying abstract belief networks and scientific theory found in Quine’s “Two Dogmas of Imperialism”, reject the “distinct boundary” between the synthetic characterization of a network, defined as the continuation of observable factors within the network, and analytic characterization of a network, the ability to quantify of an objective propositional truth within a working network. Quine, idealized the most pragmatic model for the quantification of belief networks and scientific theory, involves the retraction of this synthetic and analytic boundary with network characterization (Quine, 1951). The sum of the synthetic and analytic characterization lend itself well as a logical bedrock for formal analytical representations abstract frameworks of belief and theory.  The removal of synthetic characterization would uproot the analytical nature of understanding the translation of logic, and would call for a systematic redefinition of representation within models for abstract frameworks. In this way, the analytical characterization of abstract frameworks is disjoint and incomplete without the synthetic characterization (Carlson 2015). Quine later went on to describe theory this as a “web of belief” in later works (Quine 1978).

Brian Skyrms and Karel Lambert, outline the foundation for the principles that should embody an empirical framework representation of Quine’s propose “web of belief ” on the grounds of a coherent belief representation in “Laws of Nature: Essays on the Philosophical, Scientific and Historical Dimensions”. The laws of necessity laid out in the beginning portion of this work elaborate on the core components of the “web of belief” Quine envisioned in his later work. Skyrms and his counterpart argued that the proposed method for building this “web of belief” empirically, involve the ability to encase degree of belief within a closed and finite field of values through methods of Boolean algebra, with non-negative correlations to the degrees of belief within all aspects of the framework. The robust beliefs,  seen as the factual ground truths in an initial network, act as the core structure to the observable network. Experiments within this network are proposed applications of evidence applied to factual atoms, creating the new core structure through actions of “successful induction” on the likely outcomes of the probabilistic network over random correlation. (Pearl 1988)  Skyrms further elaborates on the foundational pillars of inference models which are coherent with the definition of inferential continuation within empirical representations and do not disrupt the nature of the Kolmogov axioms.  These axioms, stated within chapter four, are some of the axioms that lay the groundwork for the construction of node representation and combinational logic characterizations of abstract reasoning inferential models such as the Bayesian Network.

Further elaborating on the ideas brought forth by Skyrms and Lambert within inferential models dealing with abstract reasoning, Pearl introduces the idea of the need for a model that does not explicitly enumerate all possible states of a substantially large subset of uncertain probable worlds but rather a model which can capture with maximal accuracy the characteristic behaviors within a “web of belief” with a certain amount of likelihood. The visible interaction of the application of proposed transition between the current state and the next state, only plausibly computational under a restrictive independence assumption for efficiently local detached computation. The updating occurring between each state implied by the definition of preset sub-functions that are applied to the observable state with a certain amount of strength m. In the case of Bayesian reasoning, the strength m is defined by the conditional probability where m has a direct implication on the truth value of the inferential piece derived from pieces of given knowledge (Pearl 1988). These pieces embody the concept of a rule network, or a belief network in probability theory, similar to concepts laid out by Brian Skyrms and Karel Lambert.  Through these foundations, the observable evidence acting in a manner in which there is a creation of an altered truth, made through a local inference decision making system (Pearl 1988).

This chain of inference, leading to ideas within the empirical representation of reasoning that led to the formation of thought in using systems such as the Bayesian network. The philosophical perspective of abstract frameworks maps well the the structure of the Bayesian network and this is why those who are proponents of Bayesianism in Philosophy, argue that the Bayesian network is proposed as a great approximation methods for abstract reasoning.  

References:

Carlson, Matthew. “Logic and the Structure of the Web of Belief.” Journal for the History of Analytical Philosophy, vol. 3, no. 5, 2015, doi:10.15173/jhap.v3i5.28.

Pearl, Judea. 1988.  Probabilistic Reasoning in Intelligent Systems.  San Mateo, CA: Morgan Kaufman.

Russell, Bertrand. “THE PHILOSOPHY OF LOGICAL ATOMISM [with Discussion].” The Monist, vol. 28, no. 4, 1918, pp. 495–527. JSTOR, www.jstor.org/stable/27900704. Accessed 11 Mar. 2021.

Skyrms, Brian and Karel Lambert, 1995. “The Middle Ground: Resiliency and Laws in the Web of Belief,” in Friedel Weinert, ed.,Laws of Nature: Essays on the Philosophical, Scientific and Historical Dimensions. Berlin: De Gruyter

Quine, Willard Van Orman and J. S. Ullian, 1978.  The Web of Belief.  New York: Random House.

Quine, Willard Van Orman.  1951, 1953.  “Two Dogmas of Empircisim.” The Philosophical Review 60 (1): 20-43.  Reprinted in From a Logical Point of View, Cambridge MA: Harvard University Press, 1953.

[isaresearcher]

The Foundations of my Research (i.e snips from my Abstract)

The research that I am currently doing is built on the foundations of Bayesian inference and the logical implications inferred under such representations. These are endeavors that I hope will works to create momentum towards creating the collective of possible universal models, precisely quantifiable and possibly universal, for empirically representing abstract reasoning and uncertainty. The Bayesian network model offers itself as one method of quantifying and characterizing the behavior of abstract thought and theoretical frameworks. 

Through the observation of behavioral patterns relating to the continual influence of evidential pieces applied to an observable outer Bayesian belief network, composed of initialized conditionals and atomized representations of factual truth propositions, the hope is that the behaviors captured at these sub-structural points within a Bayesian network will give insight to a deeper understanding of the behaviors and patterns of internal evidence application and external evidence application applied to scientific theory and belief frameworks.  

Abstract frameworks represented through the Bayesian network models also work towards creating the foundational groundwork for the ability for network expansion. Prior research, I have been a part of in the process of publication discusses the convergence of nodes displayed in a network which can be seen as constructed from atomic propositions, each of which is either true or false.  The creation of an atom of absolute truth or absolute false. This research aims to represent through the mathematics and philosophical foundations of Bayesianism, the behaviors of the classification of sub-network gaps in relation to evidence activation of applied to sub-network portions of an observable framework. Through the characterization these sub-networks, identification of sub-networks gaps within the observable framework can be expanded allowing the potential for network growth.

[isaresearcher]

Scientific Theories as Bayesian Nets: Structure and Evidence Sensitivity

If you are interested in reading the paper we produced in the first year of UROP, references to the paper that we finished are at the following links:

University of Pittsburg archives:

http://philsci-archive.pitt.edu/18705/

Research Gate:

https://www.researchgate.net/publication/349197857_Scientific_Theories_as_Bayesian_Nets_Structure_and_Evidence_Sensitivity

Academia:

https://www.academia.edu/45050368/Scientific_Theories_as_Bayesian_Nets_Structure_and_Evidence_Sensitivity

[isaresearcher]

UROP - Undergraduate Research Program

The Undergraduate research program is a educational program that the University of Michigan created to allow undergraduate students to become a part of the diverse research community within the university. This program promotes the journey of undergraduate researchers who are looking to start their research from square one. The process of finding a research group or lab can be done outside of UROP but this program helps individuals without a background in research navigate their way through the research process and the research environment through seminars, workshops, and assignments.  

Personally, I was able to be a part of this program for two years and this program greatly changed my perspective of research and the research process when it comes to multidisciplinary work. The research I ended up being a part of helped build the knowledge base of Bayesian Networks applied to abstract frameworks. This research was conducted through the Center of Complex Systems at the University of Michigan. Although I am very much a computer science major, the work was based in philosophy, which is also another foundational discipline in the study of AI. I was very grateful for the opportunities UROP led me to and at the end of my first year our research group had completed a paper with foundations in Bayesian Networks relating to shift in abstract and theoretical frameworks. This paper, “Scientific Theories as Bayesian Nets: Structure and Evidence Sensitivity”, was published in the Philosophy of Science. I continued working with the team after my first year of UROP, which led me to a second year in UROP. I joined the Research Scholars, which is a program dedicated to fostering the space for independent research endeavors (I will speak more about this in later posts). I’m currently working on a research paper with foundations in Bayesian Networks. I’m going to publish the groundwork for the first half of my paper, explaining the history of Bayesian Networks on this blog, the paper has implementations in progress. 

I will also be posting a podcast created by two other Research Scholars and I that talks about the research process. This blog will contain a majority of the work and I have done and been a part of. (There may be some over lap between my proposal and essay but this is fine, an abstract is like a proposal on occasion). This blog is for scientific purposes.   

(Also don’t mind the tab space Tumblr apparently doesn’t mess this that.)