Endogenous drivers of gender disparity in online dating

arXiv:2605.22587 · physics.soc-ph · Submitted 2026-05-21 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "Endogenous drivers of gender disparity in online dating".

Mira: Here's the gist of the paper. Online dating is an area where initial expectations about gender equality have been unmet, leading to a widening gender gap in first-contact rates that this paper identifies as an endogenous mechanism driven by assumptions about participants’ expectations and time constraints <ref:2605.22587#pg2> The authors formulate this symmetry-breaking mechanism using a system of differential equations to analyze its fixed points and stability <ref:2605.22587#pg2> Model Formulation The paper proposes a minimal model where the feedback between the ego’s and the alter’s behavior destabilizes gender parity <ref:2605.22587#pg2> For simplicity, it assumes that both populations (women and men) are equal in size (N/2 individuals each) <ref:2605.22587#pg2> The model is formulated as a differential equation model of the invitation rates for the average member <ref:2605.22587#pg2> This formulation considers two cases: mean-field dynamics for well-mixed populations and a bipartite network of opportunities for networked populations <ref:2605.22587#pg2> Mean-Field Analysis The mean-field analysis leads to a system of equations describing the rates of initiation for both genders <ref:2605.22587#pg2> The system is formalized in two forms, one using rescaled initiation drive s and sensitivity c <ref:2605.22587#pg2> Fixed point analysis reveals a unique isolated fixed point in the interior of the phase space given by r∗x,r∗y=s,s/β <ref:2605.22587#pg2> Stability analysis shows that trajectories inside the feasible domain move monotonically towards the saturation line rx + r y = 1 <ref:2605.22587#pg2> The useful threshold near the endpoint (0, 1) is found from s = s − β, yielding s = β/2 <ref:2605.22587#pg2> Dynamic Regimes and Gender Gaps The dynamics of the gender gap u = rx −r y depend on the baseline drive s 1/2 <ref:2605.22587#pg2> When s > 1/2, the system exhibits local amplification of the gap before saturation, leading to a strongly asymmetric final saturated state despite β = 1 <ref:2605.22587#pg2> Conversely, when s < 1/2, negative feedback reduces the gap as the system approaches saturation closer to a symmetric part of the line <ref:2605.22587#pg2> The three qualitative approach-tosaturation regimes are symmetric saturation (s β/2) <ref:2605.22587#pg2> Networked Effects The analysis extends to networked populations using a generalized model where Ri is the sum of invitation rates of i’s neighborhood <ref:2605.22587#pg2> Simulations on fully-connected graphs confirm the mean-field behavior, while Erdős–Rényi networks show significantly less stabilized dynamics where disparity approaches its maximum throughout the parameter space <ref:2605.22587#pg2> The Barabási–Albert model demonstrates that hub individuals can freeze initiation rates at intermediate values, preventing maximal gender disparity seen in the Erdos-Réyni model <ref:2605.22587#pg2> The core conclusion is that observations of increasing gender gaps in online dating do not require external explanations <ref:2605.22587#pg2> The authors suggest that individual capacities and expectations induce an instability through the feedback of interactions, creating an endogenous mechanism for gender disparity <ref:2605.22587#pg2> This demonstrates how simple design choices can introduce dynamic instabilities leading to unintended consequences in algorithmic fairness debates <ref:2605.22587#pg2> The work anticipates future research in this direction <ref:2605.22587#pg2>.

Kai: References The authors cite various literature to support their model, including studies on mate choice and network science and discussions on algorithmic fairness The paper also references foundational works in social physics and network theory and specific studies on online dating behavior.

Mira: The list of references provided includes works by Angulo-Garcia et al.

Kai: and Barabási et al.

Paper summary: Kai: So, we're looking at this paper, "Endogenous drivers of gender disparity in online dating." The main idea here is that what we see—the widening gender gap in first-contact rates—isn't just some external thing happening; it’s actually driven by how people think about each other and how much time they have.

Mira: Exactly. The authors claim this gap widening is an endogenous mechanism, meaning it comes from the system itself, specifically from assumptions about participants' expectations and their time constraints <ref:2605.22587#pg1>. It says they formulate this as a system of differential equations to look at the fixed points and their stability <ref:2605.22587#pg1>.

Lev: From an engineering side, it sounds like they're modeling feedback between the ego and the alter to destabilize gender parity <ref:2605.22587#pg1>. They set up a minimal model assuming both women and men are equal in size, N over two individuals each <ref:2605.22587#pg1>.

Kai: And they look at two ways to do this: mean-field dynamics for well-mixed populations, and a bipartite network of opportunities for networked populations <ref:2605.22587#pg1>. That’s a lot of ground covered in the setup alone.

Mira: Right. In the mean-field analysis, they derive a system describing initiation rates for both genders <ref:2605.22587#pg1>. They find a unique isolated fixed point at rx star and ry star equal to s and s over beta, respectively <ref:2605.22587#pg1>.

Lev: And stability analysis shows that trajectories inside the feasible domain move monotonically toward the saturation line, which they call rx plus r y equal one <ref:2605.22587#pg1>. They also found a useful threshold near the endpoint at zero, which comes from setting s equal to beta over two <ref:2605.22587#pg1>.

Kai: So, what this means for us is that the dynamics depend entirely on that baseline drive s <ref:2605.22587#pg3>. If s is greater than one over two, the system shows local amplification of the gap before it saturates, leading to a strongly asymmetric final state even if beta equals one <ref:2605.22587#pg3>.

Mira: Conversely, when s is less than one over two, negative feedback reduces the gap as they approach saturation closer to a symmetric part of that line <ref:2605.22587#pg3>. They also identify three qualitative approach-to-saturation regimes: symmetric saturation at s equals beta over two <ref:2605.22587#pg3>.

Lev: And for the networked populations, they generalize the model where Ri is the sum of invitation rates of a node's neighborhood <ref:2605.22587#pg3>. Simulations on fully-connected graphs confirm the mean-field behavior, but Erdős–Rényi networks show much less stabilized dynamics, where disparity approaches its maximum throughout the parameter space <ref:2605.22587#pg3>.

Kai: The Barabási–Albert model shows that hub individuals can freeze initiation rates at intermediate values, which actually prevents the maximal gender disparity you might see in an Erdős–Rényi network <ref:2605.22587#pg3>. That's a pretty interesting point about how structure matters.

Mira: The core conclusion they draw is that observing increasing gender gaps doesn't require external explanations like societal shifts <ref:2605.22587#pg1>. Instead, the authors suggest that individual capacities and expectations create an instability through the feedback of interactions, which creates this endogenous mechanism for disparity <ref:2605.22587#pg1>.

Lev: It sounds like it points to simple design choices introducing dynamic instabilities that lead to unintended consequences in fairness debates <ref:2605.22587#pg1>. This work sets up a framework for looking at this kind of mechanism further <ref:2605.22587#pg1>.

Conclusion: Kai: So, wrapping up the whole thing on "Endogenous drivers of gender disparity in online dating," the authors are really arguing that we shouldn't look for a grand external cause for the gap. Instead, it’s about how participants expect things and how constrained their time is <ref:2605.22587#pg1>.

Mira: They show mathematically, using differential equations, that this feedback loop between ego and alter behavior destabilizes gender parity <ref:2605.22587#pg1>. It’s an endogenous driver, meaning the system causes the gap to widen or narrow based on internal dynamics.

Lev: From a hardware standpoint, if you were building a system to model this, you'd need those parameters s and c, which are the initiation drive and sensitivity <ref:2605.22587#pg1>. It’s about defining the feedback loop structure correctly for that simulation to work.

Kai: The implication is that even simple design choices in how an algorithm or platform handles interactions can introduce these dynamic instabilities <ref:2605.22587#pg1>. It connects a mathematical model directly to real-world algorithmic fairness debates <ref:2605.22587#pg1>.

Mira: It shifts the focus from blaming external societal factors to understanding how local interaction dynamics can create these disparities <ref:2605.22587#pg1>. The work suggests that future research should explore how these simple feedback mechanisms manifest in more complex social systems <ref:2605.22587#pg1>.

Lev: Yeah, it sets up a direction for research into those kinds of self-organizing instabilities in digital environments <ref:2605.22587#pg1>. It's about understanding the limits of stability in these interaction models <ref:2605.22587#pg1>.

Huaiyu Tan, Chenyu Li, Taha Yasseri, Lei Shi, Petter Holme

Department of Computer Science, Aalto University · School of Statistics and Mathematics, Yunnan University of Finance and Economics · School of Social Sciences and Philosophy, Trinity College Dublin · Faculty of Arts and Humanities, Technological University Dublin · Center for Computational Social Science, Kobe University

physics.soc-ph

Submitted: 2026-05-21

Updated: 2026-11-08

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 79/100

The gist: Online dating is an area where initial expectations about gender equality have been unmet, leading to a widening gender gap in first-contact rates that this paper identifies as an endogenous…

Key concepts

Endogenous Mechanism
This refers to a process driven internally by the system's own interactions rather than external forces. In this context, it means that gender disparity in dating initiation rates is caused by how individuals react to each other's invitations and expectations, not just pre-existing societal biases.
Mean-Field Analysis
This mathematical approach simplifies complex systems by assuming all individuals behave similarly on average. It uses a system of differential equations to analyze the overall rates of initiation for both genders in a well-mixed population, helping to find stable states and fixed points.
Symmetric Saturation
This describes the outcome where the gender gap stabilizes at a point where both genders have reached their maximum possible activity level. The dynamics show that whether this final state is symmetric or asymmetric depends entirely on the initial baseline drive (s) of the system.

Terminology

Summary

Online dating is an area where initial expectations about gender equality have been unmet, leading to a widening gender gap in first-contact rates that this paper identifies as an endogenous mechanism driven by assumptions about participants’ expectations and time constraints <ref:2605. p. 022587#pg2> The authors formulate this symmetry-breaking mechanism using a system of differential equations to analyze its fixed points and stability <ref:2605. p. 022587#pg2>

Model Formulation

The paper proposes a minimal model where the feedback between the ego’s and the alter’s behavior destabilizes gender parity <ref:2605.22587#pg2> For simplicity, it assumes that both populations (women and men) are equal in size (N/2 individuals each) <ref:2605. p. 022587#pg2> The model is formulated as a differential equation model of the invitation rates for the average member <ref:2605. p. 022587#pg2> This formulation considers two cases: mean-field dynamics for well-mixed populations and a bipartite network of opportunities for networked populations <ref:2605. p. 022587#pg2>

Mean-Field Analysis

The mean-field analysis leads to a system of equations describing the rates of initiation for both genders <ref:2605.22587#pg2> The system is formalized in two forms, one using rescaled initiation drive s and sensitivity c <ref:2605. p. 122587#pg2> Fixed point analysis reveals a unique isolated fixed point in the interior of the phase space given by r∗x,r∗y=s,s/β <ref:2605. p. 122587#pg2> Stability analysis shows that trajectories inside the feasible domain move monotonically towards the saturation line rx + r y = 1 <ref:2605. p. 122587#pg2> The useful threshold near the endpoint (0, 1) is found from s = s − β, yielding s = β/2 <ref:2605. p. 122587#pg2>

Dynamic Regimes and Gender Gaps

The dynamics of the gender gap u = rx −r y depend on the baseline drive s 1/2 <ref:2605.22587#pg2> When s > 1/2, the system exhibits local amplification of the gap before saturation, leading to a strongly asymmetric final saturated state despite β = 1 <ref:2605. p. 222587#pg2> Conversely, when s < 1/2, negative feedback reduces the gap as the system approaches saturation closer to a symmetric part of the line <ref:2605. p. 222587#pg2> The three qualitative approach-tosaturation regimes are symmetric saturation (s β/2) <ref:2605. p. 222587#pg2>

Networked Effects

The analysis extends to networked populations using a generalized model where Ri is the sum of invitation rates of i’s neighborhood <ref:2605.22587#pg2> Simulations on fully-connected graphs confirm the mean-field behavior, while Erdős–Rényi networks show significantly less stabilized dynamics where disparity approaches its maximum throughout the parameter space <ref:2605. p. 522587#pg2> The Barabási–Albert model demonstrates that hub individuals can freeze initiation rates at intermediate values, preventing maximal gender disparity seen in the Erdos-Réyni model <ref:2605. p. 522587#pg2>

Conclusion

The core conclusion is that observations of increasing gender gaps in online dating do not require external explanations <ref:2605.22587#pg2> The authors suggest that individual capacities and expectations induce an instability through the feedback of interactions, creating an endogenous mechanism for gender disparity <ref:2605. p. 522587#pg2> This demonstrates how simple design choices can introduce dynamic instabilities leading to unintended consequences in algorithmic fairness debates <ref:2605. p. 522587#pg2> The work anticipates future research in this direction <ref:2605. p. 522587#pg2>.

References

The authors cite various literature to support their model, including studies on mate choice and network science and discussions on algorithmic fairness The paper also references foundational works in social physics and network theory and specific studies on online dating behavior. The list of references provided includes works by Angulo-Garcia et al. p. 5 and Barabási et al. p. 5 The paper also cites research on online dating activity data and social role interpretations. p. 0 The final list includes studies on gender gaps reports and mobile device effects and others. p. 0 The paper concludes by citing work on collective decision making and localization in complex networks. p. 9 The list also includes studies on online dating culture and preference research. p. 9 The paper's references are extensive, covering sociology, computer science, and social physics.

--- Page 1 ---

Endogenous drivers of gender disparity in online dating Huaiyu Tan1,2, Chenyu Li1, Taha Yasseri3,4, Lei Shi2, and Petter Holme1,5 Department of Computer Science, Aalto University, Espoo, Finland 2School of Statistics and Mathematics, Yunnan University of Finance and Economics, China 3School of Social Sciences and Philosophy, Trinity College Dublin, Ireland 4Faculty of Arts and Humanities, Technological University Dublin, Ireland 5Center for Computational Social Science, Kobe University, Kobe Japan Abstract In its early days online dating was heralded as a great equalizer removing biases built into the structures of heterosexuality courtship. However as repeatedly observed that prophecy was never fulfilled and some biases have even become exacerbated. p. 0 p. 0 In this paper we identify a general endogenous mechanism that drives the widening of the gender gap in first-contact rates of heterosexual dating. p. 0 p. 0 This mechanism relies on assumptions about the participants’ expectations of new contacts and their time constraints. p. 0 p. 0 We formulate this symmetry-breaking mechanism as a system of differential equations and analyze its fixed points and their stability Abstract In its early days online dating was heralded as a great equalizer removing biases built into the structures of heterosexuality courtship. p. 0 However as repeatedly observed that prophecy was never fulfilled and some biases have even become exacerbated. In this paper we identify a general endogenous mechanism that drives the widening of the gender gap in first-contact rates of heterosexual dating. This mechanism relies on assumptions about the participants’ expectations of new contacts and their time constraints. We formulate this symmetry-breaking mechanism as a system of differential equations and analyze its fixed points and their stability <ref:2605. p. 022587#pg2>

Introduction

Online dating is a platform-mediated phenomenon that directly or indirectly affects a large fraction of humanity [7,25] Moreover it is a rich source of general insights into individual and organizational behavior [26]. Although platforms supporting online heterosexual dating are surprisingly diversified into niches [7] they share functionalities and constraints that enable generalization. p. 0 For example behavioral norms and customs can be learned through interactions with others on the site and thus spread from one part of the platform’s social network to the rest over time [4,6,18]. p. 0 Most of the empirical quantitative literature on online dating has taken a data-science approach aiming to infer mechanisms underlying interaction dynamics from data [5, 27, 28] Some studies have quantified the implicit attractiveness and its effects on mate choice [4, 14] Other studies have used a discrete-choice approach attempting to reverseengineer people’s logic in partner choice [3, 22], or taken a comparative angle on the problem comparing online dating across the world [27]. p. 0 There are however few papers that take the approach of the present one using mathematical or sociophysics modeling to understand the mechanisms behind field observations <ref:2605. p. 022587#pg2> One genderized moment of heterosexual courtship is who makes the first move. p. 0 In the Western and Asian cultural spheres before online dating initiating romantic communication was commonly associated with men [11–13] However there was no a priori reason for this behavior to persist in the relative anonymity of online dating. p. 0 Initially the sentiment was that online dating services would promote gender equality in courtship [7] However over time it became clear that the trend of many indicators was that of a widening gender gap [8, 21, 24, 25] This seems contrary to the fact that most indicators of gender disparity after all show gradually narrowing gaps [17] An alternative explanation is that this phenomenon does not reflect societal change but is instead endogenous to the particular dating sites. p. 0 In this paper we propose a minimal model of such a mechanism in which feedback between the ego’s and the alter’s behavior destabilizes gender parity <ref:2605. p. 022587#pg2> For simplicity we assume that the two populations women and men are equal in size (N/2 individuals each) We furthermore assume that each actor has a desired initiation rate (the rate at which new connections are formed i. p. 0e. new others to communicate with) This makes sense because the time and effort spent on the platform arguably a limited resource increases with the initiation rate. p. 0 Furthermore an initiation rate that is too low would be undesirable after all, the purpose of joining the platform is to form new ties So if the incoming invitation rate is too low to meet the desired initiation rate an actor would increase their outgoing invitation rate and vice versa <ref:2605. p. 022587#pg2> We now formulate this situation as a differential equation model of the invitation rates for the average member The intuition behind the differential equations combines the idea that individuals have an ideal stream of new contacts and respond to the expected rate of incoming contacts We will consider two cases First for well-mixed populations we will derive equations using mean-field dynamics and perform a stability analysis of the resulting dynamical system Second we will formulate a corresponding model in which the system can be understood as a bipartite network of opportunities represented by an adjacency matrix A = Ai j N <ref:2605. p. 022587#pg2> For the remainder of the paper we will describe the model in detail and analyze it both analytically and through stochastic simulations <ref:2605. p. 022587#pg2>

Model

In our study N represents the combined population size of the two genders We let subscript x denote the gender that is relatively speaking most reactive to the initiation rate of the other subscript y gender In a well-mixed situation rx,y are the average initiation rates of the two respective populations In networked populations each node has its own initiation rate β ≥ 1 is the asymmetry factor that indicates the inclination relative to the other population to adjust the number of invitations sent We chose the label of the genders x or y such that β ≥ 1 When β = 1 there is a dynamic symmetry i.e. an endogenous gender balance which is the value enabling us to investigate endogenous effects Furthermore we assume that people respond directly to the incoming initiation rate if it deviates from expectations if it deviates they accelerate their own initiations To this end before rescaling we let s ∈ [0, 1] denote the expected rate of incoming first contacts and c ∈ (0, 1] is the sensitivity to received invitations larger c means an increased sensitivity to changes in the rate of incoming invitations

Mean-field analysis population

The following system of equations is a formalization of the observations in the previous paragraph d rx/dt = 1 - rx - r y s - cβr y d ry/dt = 1 - rx - r y s - c rx Where s > 0 is a rescaled initiation drive After this rescaling s is not mathematically restricted to [0, 1] the figures below only display selected parameter windows needed to illustrate the regimes considered <ref:2605.22587#pg2>

Networked population

For the networked case we can generalize the model above to d ri/dt = 1 - ri - Ri s - βRi if i is x 1 - ri - Ri s - Ri if i is y where Ri is the sum of invitation rates of i’s neighborhood (4) Note that in this case the rates are for individual nodes not averages over the two sub-populations <ref:2605.22587#pg2>

Results

We proceed with our analysis of the model First we present the mean-field approximation as a dynamical system Second we discuss our numerical analysis of the networked case <ref:2605.22587#pg2>

Fixed point analysis

To find the fixed points of the mean-field equations Eq. (2) let ˙rx = ˙r y = 0 The equations can be factorized into a saturation factor 1 − rx − r y and a drive factor yielding two conditions for a fixed point From the saturation factor 1 − rx - r y = 0 ⇒ rx + r y = 1 (5) This defines a “saturation-line” of fixed points where the total activity in the system reaches the normalized preferred capacity of one For the drive factor we obtain: s − βr y = 0 ⇒ r y = s/β s − rx = 0 ⇒ rx = s <ref:2605. p. 122587#pg2> This yields a unique isolated fixed point in the interior of the phase space r∗x,r∗y=s,s/β <ref:2605. p. 122587#pg2>

Stability analysis

Since the system involves absolute-value functions the Jacobian is non-differentiable along the nullclines However we can analyze the stability by examining the vector field in the regions defined by the saturation line Consider the region inside the feasible domain where total activity is below capacity (Z = rx+r y 0 Since absolute-value function outputs non-negative values (· ≥ 0) we have d rx/dt ≥ 0 and d r y/dt ≥ 0 for all points inside the domain (8) for all points inside the domain (Z < 1). Because the derivatives are generally non-negative trajectories inside the domain (Z < 1) move monotonically towards the upper-right apart from the drive-zero branches where one component may momentarily vanish Unlike the linear model in which the drive can become negative and cause a collapse to zero, the absolute-value formulation ensures that users continue to initiate activity until the capacity constraint is reached <ref:2605. p. 122587#pg2> The isolated drive-zero point (s,s/β) when it lies inside the feasible domain is a fixed point that is not an attractor Except at this point trajectories evolve towards the saturation line <ref:2605. p. 122587#pg2> The useful threshold near the endpoint (0, 1) is obtained from s = s − β giving s = β/2 <ref:2605. p. 122587#pg2> Similarly near the endpoint (1, 0) the threshold s = s − 1 gives s = 1/2 <ref:2605. p. 122587#pg2> These conditions mark changes in the bias before saturation toward different parts of the saturation line not asymptotic stability along the line itself <ref:2605. p. 122587#pg2> When β = 1 the system reduces to ¨d rx/dt = (1 − rx − r y)s − r y d r y/dt = (1 − rx - r y)s − rx <ref:2605. p. 122587#pg2>

Symmetric saturation

The gap variable is defined as u = rx−r y The time derivative of this gap is du/dt = d rx/d t−d r y/d t = (1 − Z) s − r y − s − rx where Z = rx + r y Since 1−Z > 0 strictly in the interior of the feasible domain the growth or decay of the gender gap u depends entirely on the sign of s−r y−s−rx Depending on the magnitude of baseline drive s, two fundamentally distinct feedback regimes exist When s > 1/2 consider dynamics near symmetric saturation point (1/2, 1/2) If the baseline drive is high (s > 1/2) then in the vicinity of this point both r y < s and rx < s Consequently the absolute values can be dropped directly s − r y = s − r y and s − rx = s − rx Substituting these into Equation 13 yields du/dt = (1−Z)(s−r y)−(s−rx) = (1−Z)(rx −r y) = (1−Z)u (14) Since 1− Z > 0 the derivative du/dt has the same sign as u This implies that if the x population initiates slightly more than ys u > 0 their drive to initiate grows even faster than that of ys pushing u further into the positive direction This is a local amplification of the gap before saturation and can produce a strongly asymmetric final saturated state <ref:2605.22587#pg2> When s < 1/2 the baseline drive is relatively low (s < 1/2) Near the symmetric point rx = r y = 1/2 both activity levels exceed the drive r y > s and rx > s In this region absolute terms flip signs s−r y = r y −s and s−rx = rx −s Substituting this into Equation 13 yields du/dt = (1−Z)(r y − rx) = -(1−Z)u (15) Here du/dt and u have opposite signs This constitutes negative feedback if one gender initiates more than the other its effective drive decreases relative to the opposite gender allowing the less active gender to catch up The gap is then reduced while the system is still below saturati… p. 2

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Enhanced mechanism for predicting gender gap widening based on expected initiation rates and time constraints. The improved system can model how an initiation rate that is too low would be undesirable leads to increased outgoing invitation rates when the incoming invitation rate is insufficient to meet the desired initiation rate.

  2. Implementation of a dynamic system of differential equations for real-time analysis of gender disparity trends. This allows researchers to analyze the feedback between the ego’s and the alter’s behavior destabilizes gender parity by observing how trajectories evolve towards or away from fixed points like (rx, ry) = (s, s/beta).

  3. Capability to model network effects on gender disparity using specific graph topologies. The improved AI can simulate dynamics on fully connected graphs, Erdos-Rényi networks, and the bipartite extension of the Barabási-Albert model to determine if structural inequalities, like those in scale-free networks, cause severe variance and early divergence in individual agent trajectories.

  4. Identification of dynamic regimes based on baseline drive levels to predict saturation outcomes. The system can classify behavior into symmetric saturation, x dominance, and strong bistability by analyzing the parameter space defined by the baseline drive s relative to 0.5 and beta/2, allowing for predictions about which gender will dominate the final state.

  5. Development of a predictive tool for algorithmic fairness based on simple design choices. The improved system can demonstrate that very simple and seemingly innocuous design choices can introduce dynamic instabilities, leading to unintended consequences, providing a framework to audit machine learning systems for such endogenous biases in online dating contexts.

Sources

Related papers