els and algorithms; in worst, they might enable the users to make a wrong decision. Users trust is violated when these algorithmic systems produce an outcome that is harmful, biased, and unethical. As a consequence, sometimes users end up stop using such product or services [6,7]. Thus, designing transparent algorithmic user interfaces are getting more and more attention among the research community [8] to make the algorithmic aspects explicit and more transparent.

Previous research have advocated for transparent recommendation systems in various domains [9,10], transparent statistical research practices [11], transparent debugging [12,13], and transparent journalism [14,15]. While others have examined and emphasized the importance of transparent data collection process [16,17]; Microsoft datasheets for datasets presents one example [18,17] to achieve trans-parency and accountability during Machine Learning (ML) lifecycle for both dataset creator and dataset consumer.

Similarly, various techniques are also available to make the underlying algorithmic assumptions more open, interpretable, and easy to understand; explanation is one of them [2,19]. Recently, researchers also explored the potential of socio-technically inspired perspective such as Social Transparency (ST) [20], perhaps, due to the social nature of interpretability [21]. Explanation tool or Explainers, also known as interpretability tool are available as open-source Python packages to describe both white box and black box models [18,22,23,24]. They provide an easy interpretation of the model's mechanism and outcome in a trustworthy, transparent, and safer manner [25]. Applying these explainers requires the user to call pre-defined functions, integration with complex workflows [21], and is often "critiqued for its techno-centric view" [20]; and, applying them requires programming. Furthermore, as these tools are publicly available and free to use, research shows that even expert data scientists overuse the explainers (InterpretML [22]) prediction by overly trusting them, and sometimes use them without proper understanding [26].

Even though various design guidelines [8,27,7] and principles [13] exists, including explanatory prototypes [13,12], for designing transparent algorithmic UI's; however, design approaches considering users personality, cognitive style, problem-solving strategies are still unexplored. Research in education, psychology, marketing, and other domains indicate that there exists a significant difference in the way different users use and process information [28,29,30]. We do not know how these different cognitive style or mental processes will play out in the case of designing a transparent algorithmic system. Likewise, how much transparency is even enough or desired across different critical audience [31] is also unknown.

We also do not know how to measure a particular user groups transparency needs. To bridge these gaps, we propose a playful exploration approach called "tinkering" [32,30] as a way of designing transparent algorithmic UI's by examining Facebook News Feed. There are a couple of benefits of our proposed approach: first, it is possible, that the exploratory nature of algorithm features (matrices) will not overwhelm the user by providing the user sufficient algorithmic control in the personalized interactive news feed experience; second, by enabling the measuring ability at the interface level of how much transparency is desired for each groups, we include the possibility of transparent interface designs that are inclusive (e.g., gender [33,34].) We do not intend to modify or suggest a new facebook ranking algorithm, instead, our main objective is to encourage a different perspective in the design of transparent algorithmic user interfaces. We end our discussion by suggesting potential future research directions.

Algorithmic systems are everywhere; they range from search engine [35], social media news feed, to video/music or product recommendation systems. These systems has the ability to impact and influence the way we perceive, interact and experience the world around us. Much of the blessings associated with these systems are not free from its perils [36], in many cases, the algorithms are not fair [6].

For example, research showed that Google search algorithm displayed biased and racist content when queried for certain keywords such as "black girls" [35]. The absence of context associated with the search results makes algorithmic interpretation even more difficult [16,35]. Researchers also discovered biases in image annotation [4] in computer vision across various facets such as race, gender, and weight [3]. As a consequence, researchers advocated for making data's economic value transparent [19], because users will likely stop using a technology due to the lack of opacity about how their generated data is actually used. Algorithmic transparency has also received a considerable amount of attention in data science work practices [37], medical AI applications [38] among many others. The lack of transparency causes mistrust [6] and dissatisfaction in these systems. There exists an increasing opportunity to establish trust through transparency with the advancement of digital media and computer technology [39]. For simplicity, in this paper, we examined algorithmic transparency in the case of Facebook news feed because it is a well-studied sociotechnical interactive system [2,40,41]. Also, very little is known about how the news feed curation works [41], thus, we wanted to propose an early stage transparent algorithmic news feed prototype, to imagine how a transparent news feed might look like.

Previous research have shown the significance of explanation (e.g., how, what, why) to achieve transparency in algorithmic systems such as Facebook news feed curation [19,2]. Explanation enables the user to become more aware, make judgment about the correctness of the output and mechanism; it also supports interpretability, accountability of algorithmic decision making action [2]. Transparency and interpretability are related through explanation, these relationships are shown in Figure 2. The way existing interpretability tool works is also through explanation [18]. Explana-tion tools or explainers such as white box and black box API's/packages [18] are available to describe a wide range of ML models. White box explainers (glass box or generalized additive models or GAM's [26]) works directly on data to explain models such as linear regression that is easy to understand; while the black box explainers (post-hoc explanations) requires input and ML models output to explain models that are harder to explain such as neural networks [25]. The explainers are open-source, available in Microsoft's Azure ML packages (e.g., SHAP [42], LIME [32], eli5 [22]), Google Cloud API (e.g., What If Tool [23]) platforms and also in Python Sklearn libraries. Figure 1 shows the output of calling the SHAP summary plot function visualization. These explainers operate on tabular, text, and image data [18,25]. While these tools and visualizations have helped data scientists to understand the model's output in some cases, but it also depends on the explainers used. Research showed that due to the free availability of these tools, data scientists misuse them by overly trusting them [26,5]. These tools are mainly used by data scientists and ML practitioners. However, they faced numerous challenges such as model instability (e.g., LIME, SHAP), tools not scaling with large dataset, difficulty in tool integration with their workflow [21]; thus, these tools may not be accessible to technical nonexperts (e.g., legal professionals) [26,21], because applying them requires more than basic programming skill and experience (e.g., knowledge of ML models, built-in methods, see Fig 1). Moreover, we might be able to build transparent data science tools using these explainers, however, we cannot apply them for designing transparent news feed or sociotechnical systems (e.g., transparent twitter), because most of these news feed uses proprietary algorithm [40]. Research also showed that explanation may enhance user's positive attitude towards a system, but not necessar- The right image shows the output of SHAP interaction values, which makes a calls to the function shap.TreeExplainer(model).shap_interaction_values() on tree models, to display the interaction among various demographic variables.

ily trust [7]. These limitations encouraged us to discover a distinct way of designing algorithmic transparency at the interface level. Thus, in this study, we propose a cognitive style based approach through incorporating tinkering ability into the design.

Tinkering is a cognitive style or a "mindset" to approach problem-solving through "experimentation and discovery", [43]; it is associated with exploratory behavior, trial and error method, deviation from instructions when learning [44]. Tinkering is an act of playful experimentation that enhances motivation, influence learning, innovation [45,44], impact task completion and performance [46,47,30]. Even though tinkering is often associated with making activities under playful conditions [48], tinkering behavior has shown to improve learning and educational benefits in domains such as engineering, robots, and programming (e.g., debugging, block-based) [48,30,46]. However, tinkering on its own may [44,48,45] or may not be a beneficial strategy for problemsolving [30], for example, Beckwith et al. [30] proposed that effective tinkering happens when it is associated with pause and reflection about software features. We applied tinkering to design transparent algorithmic system with the hope that its exploratory nature will make the overall algorithmic transparency experience less overwhelming. Informed by several existing research [28,30,49,50], the association of gender with respect to tinkering cannot be overlooked, discussed below.

Previous research has identified gender differences in the tinkering attitude [30,44,28], and confirmed that females tend to tinker or explore new software features (e.g., spreadsheet) less compared to the males for problemsolving software [30], also in Computer Science education (e.g., programming assignment) [44]. Numerous studies have showed that tinkering is a mental or psychological trait that distinguishes how different genders (males and females) approach a given task (e.g., making an Arduino project) [44,47,28,30,48]; tinkering is also one of the facets of gender-inclusive design [49]. Gender-inclusiveness [49] de-sign do not suggest building a different version of the same software for a different group of user [28], rather, it advocates for designs that support different gender groups equally [51]. Gender inclusivity relies upon five facets of gender differences: motivations, computer self-efficacy, tinkering, information processing style, risk aversion, that can impact the use of problem solving software. While "Inclusive Design considers the full range of human diversity with respect to ability, language, culture, gender, age, and other forms of human difference. " [52], gender is one aspect of inclusive design. Informed and inspired by previous research [51,34,30,28], in this work, we will focus on only gender inclusiveness.

In this study, we discuss tinkering as a means of designing transparent algorithmic UI's, because of it's inherent exploratory nature might be less overwhelming [13] to the user while providing personalized user experience; tinkering also adds an additional ability in the design to detect gender differences in the design. Detecting gender issues early on during the design process improves the usability of the software for everyone, including marginalized users [28,30].

Transparent algorithmic system for various interactive domain applications will work differently; not only at the user interface level but also at the algorithmic level. For simplicity of design and illustration, we focused on only a single interactive domain, the Facebook news feed. We first provide the rationale behind designing tinkering approach to transparent UI design, followed by feature description, and a brief discussion of complete transparent news feed prototype, Glass News Feed. Finally, we show how tinkering based approach can be applied to determine how much algorithmic transparency is desired across different user groups, which essentially helps in the determination of gender differences in the interface design.

Transparent algorithmic UI's primary objective is to reveal how it works by -explaining the mechanism it uses to produce an outcome [2]. Even though previous research suggests design guidelines for them, however, they did not addressed the cognitive aspects as a design element in the UI design [8,27,13]. We decided to investigate this gap by proposing tinkering based transparent algorithmic UI (see Figure 2). Tinkering is not only related to problem-solving but it is also associated with decision making [28], with regards to software feature (e.g., new or existing) exploration; thus, it allows us with the ability to measure algorithmic transparency needs (how much) across diverse population at the UI level. Following Beckwith et al. [30], we applied the term tinkering as users exploratory behavioral action and practice with software, here news feed features. Allowing the user to "playfully experiment" with the transparent algorithmic features serves two potential purposes: first, the design provides algorithmic information in a manner that is not overwhelming to the user while providing personalized interactive experience; second, tinkering-based design has the ability to be tested for gender inclusiveness issues.

Facebook is one of the most widely used sociotechnical systems. Undoubtedly, Facebook has opened numerous opportunities for work, business, collaboration and communication by connecting people worldwide; nonetheless, it has also caused various problems ranging from privacy threats, mental illness, addiction, to users trust violation. Facebook news feed has been well studied in the literature for users perception and understanding of news feed transparency [2,40,41]. Facebook news feed works by allowing users to share content and consume content through automated selection and ranking algorithm. The news feed provides content that are relevant, interesting, informative, having high quality [41]. Users are usually unaware of how the underlying algorithmic curation works [2,53,41]. Therefore, as a case study, we turned our attention to design transparent algorithmic news feed using a tinkering based approach.

Glass News Feed Algorithm Features: Facebook news feed algorithm applies users past action and behavior data to provide content. Even though existing Fecebook news feed provides certain amount of very high-level control about what the user sees and why (e.g., sort, hide, block, follow/unfollow, limited profile) [40], however, a transparent news feed requires other non-trivial control, which revolves around answering "how" question in addition to answering "what" and "why", but, Facebook news feed blog does not explain that very clearly [41]. For simplicity, we turned users actions in the news feed into transparent algorithmic interface feature (see Figure 3); for example, i) counts such as like and reaction count (e.g., happy, love) in photos, status, videos, ii) list such as friends list, family, acquaintance, and iii) other features such as liked pages (e.g., product, business), public groups user follow, [41], including descriptive features such as notes, tags can used as features. The real future transparent application might apply a different set of features in various categories. We also enabled the ability for the user to be able to create their own feature set and For simplicity, a feature can be in any of the following states: selected (✔) indicating on [30], unselected (as empty) indicating off, feature explanation: How (?) option [40,30]; a click on the drop down menu activates these options for selection.

Figure 4: Transparent Glass News Feed prototype applying Facebook user activities such as like counts, groups, emojis count, etc., as features to provide personalized interactive experience (left).

When "Refresh" button is clicked, the personalized news feed is displayed (right).

explore the news feed outcome. We showed only some of these features in the proposed prototype.

Tinkering Capability: We enabled tinker-ing capability in the design through incorporating Facebook data as interface features (see Figure 3); these features can be frequently turned on and off by the user for exploration, as described in [30,28]. Each of the features can be in one of the states: i) when checked, marked by ✔, meaning feature selected for current exploration, ii) when un-checked, indicated by an empty box, meaning not currently under exploration, and iii) a question mark (?) to provide feature-related explanation [40]. These capabilities are hidden under the drop down menu, this button is activated when clicked, otherwise, it remains inactive to make sure that these extra abilities does not overwhelm the user. Tinkering count for any particular user can be measured by simply counting the number of features that were turned on and off during a session.

Subtle Explanation: Transparency cannot be implemented without providing some kind of explanation. Thus, inspired by Rader et al. [2], we subtly added "How" explanation. "How" explanation, "Informs participants that the ranking algorithm uses data collected about users and their behaviors to calculate score score for each story". Explanation "How" was indicated by a question (?) mark, and gets activated when clicked to indicate more information (see Fig 3), by showing other metadata information about the queried feature. This explanation feature becomes really essential when user creates their own "userdefined" feature set for experimentation with the news feed. This ability of defining userdefined feature set ensures enough flexibility for exploration without overwhelming the user with all possible tinkering options.

The complete very first prototype of Glass news feed is presented in Figure 4. Tinkering or feature set exploration window is depicted on the left, and the corresponding outcome is shown on the right. For simplicity, we assumed that the features will appear on the news feed itself, though, it is possible to design the exploration window in many different ways, for different applications.

The design of tinkering enabled transparent algorithmic UI was inspired by the design techniques suggested in problem-solving domain [30]. We added tinkering capabilities in the Glass news feed design for feature set exploration and experiment with corresponding news feed output. The Glass news feed feature sets were derived from relevant research [41], and was kept to a minimum number to avoid causing information overload. We incorporated the ability to add a "user-defined" feature set to provide some flexibility. The entire interactive experience is built on the concept of "playful experimentation" while giving users enough control without hurting their interface experience [13].

The resulting news feed is displayed on the news feed with confidence or accuracy information (top right corner in Figure 4). The algorithmic outcome (news feed after refresh) intentionally provides minimal information such as confidence accuracy, because we have no idea what specific selection or ranking algorithm Facebook originally uses for news feed curation [41]. For similar reason, we did not apply visualization, however, it is a possibility [18,5,13]. This is again our very first trying of tinkering approach to achieve algorithmic transparency in interactive systems.

Glass News Feed Though, our main motivation for applying tinkering approach to design transparent algorithmic systems was to enable the exploratory nature of tinkering to unfold in the interface design, we suspect that the playful cognitive style might also be able to reduce cognitive load in the transparent systems [20,15]. Another direct outcome from applying tinkering approach is that it allows the designer to check for gender issues in the design. The way our proposal is able to detect and measure gender differences is through measuring (counting) "how much" tinkering (on/off) an user engaged during an episode, consistent with prior study of tinkering in problemsolving [30] domain. Similarly, whether our design suffers from gender issues or not can be measured by collecting users tinkering frequency, tinkering episode and tinkering rate.

For any particular task (in a user study), i) tinkering frequency is the number of features a user have turned on and off; ii) tinkering episode can be defined as a fixed amount of time for task completion; iii) tinkering rate is the ratio of the previous two measures (i and ii). Depending on the number of user groups taking part in the study, tinkering measures for each user groups can be passed to statistical or ML models for quantitative analysis. We did not show these measures in this study, rather, these are some of the potential areas for future exploration. Transparency is critical for designing interactive social media news feed for trust building and system acceptance. However, too much openness may make the system vulnerable to various kinds of exploitation, harm, and misuse. Thus, how to balance such competing, yet a necessary aspect of a transparent news feed requires further inquiry.

In this study, we proposed a tinkering based approach towards designing transparent algorithmic user interface. There are several limitations to this study that is worth mentioning. First, the Glass news feed design was inspired by relevant research in Facebook news feed [41] and tinkering [30,28]. While background research related to tinkering was broad and detailed, Facebook news feed research was limited, because Facebook news feed uses proprietary algorithm [40]. A useful workaround suggested in [2] can be beneficial for designing other socio-technical transparent systems, by content analysis of blog posts or related sources. Second, we proposed transparent algorithmic prototype for social media news feed only, there are other algorithmic domain applications such as recommender systems, data science tools, data journalism tools, that can be designed and tested using similar strategy suggested in this study. Our design was also very limited in features and capabilities. Future work might take our design concept, expand (features/matrics) it, and test with the various users to see how tinkering plays out in achieving transparency. Third, we addressed tinkering approach to the design of transparent algorithmic system, however, there are other facets of cognitive styles such as riskaversion, information processing style (e.g., [49,34]) which might influence the use of transparent systems (especially for females), we did not address these complex relationships while designing our proposal. Thus, future work should examine other cognitive styles of problem solving and their influence on tinkering when designing transparent algorithmic system. Additionally, most previous studies investigated genders (males and females) influence in design research, thus, we need to expand our understanding by including marginalized LGBTQ+ communities in our design process. Finally, gender is one dimension in the broad spectrum of inclusive design [52], thus, future studies should investigate other diversity dimensions (e.g., race, class, language) while designing transparent systems.

The demand for transparent algorithmic user interfaces is on the rise. Previous research applied explanations associated with text and visualization techniques to improve the interpretability of ML models. These specialized tools are mainly used by technical experts such as data scientists and cannot be easily adapted for developing other transparent domain applications such as socio-technical systems. Furthermore, sample transparent UI prototypes in diverse domains exists, however, we do not know how to design a transparent interactive Facebook news feed that do not hurt the UX. Also, how much transparency is even desired across diverse population and how to measure that is also unknown. Thus, in this study, we proposed a very first tinkering based transparent algorithmic Glass News Feed UI prototype with the potential to navigate these multiple scenarios. This proposal can be easily expanded and adapted to design transparent algorithmic systems in various domain applications (e.g., transparent algorithmic tools for the journalists [54]), which essentially requires further examination with various groups of users to understand its technical feasibility, ethical and societal implications (e.g., benefits, harms).

I would like to thank anonymous reviewers for their valuable comments and feedback.