Talkboctopus

Title: Predicting extinction cascades using species interaction networks Abstract: Network studies have long explored the relationship between network structure and stability. In particular, network robustness measures the resistance of a network to node extinction. In other words, how likely is it for a failure at one node to spread and cause a catastrophic and cascading collapse of the network? Although these are well-explored dynamics in theoretical and simulated network studies, rarely have predicted extinctions in networks been compared against real-world observed extinctions. In this talk, I'll discuss a project to generate predicted extinction cascades in Gulf of Maine coastal food webs and a series of observational and experimental tests of these predicted extinctions, bringing together theory, simulation, and empirical studies in a highly perturbed ecosystem.

Title: A Bayesian approach to learning mathematical models from data Abstract: For a few centuries, scientists have described natural phenomena by means of relatively simple mathematical models such as Newton's law of gravitation or Snell's law of refraction. Sometimes, they found these models deductively, starting from fundamental considerations; more frequently, however, they derived the models inductively from data. With increasing amounts of data available for all sorts of (natural and social) systems, one may argue that we are now in a position to inductively uncover new interpretable models for these systems. But can this process be authomatized? That is, can we design algorithms that automatically learn, from data, the closed-form mathematical models that generated them? And if so, are the true generating models always learnable? In the talk I will discuss how we can use inference approaches to define what we call a Bayesian machine scientists which can obtain closed-form mathematical models from data. We will see how often, noisy observations result in there being multiple models that can describe our data well. Nonetheless, with our approach we can show that there is a transition occurring between: (i) a learnable phase at low observation noise, in which the true model can in principle be learned from the data; and (ii) an unlearnable phase, in which the observation noise is too large for the true model to be learned from the data by any method.

Abstract: When algorithmic predictions are used to inform social decision-making, these predictions don’t just forecast the world around it: they actively shape it. From online recommender platforms to financial predictions, machine learning systems are in active feedback with the surrounding environment and have the ability to steer the underlying data distributions toward different targets. While traditionally neglected, these causal forces of prediction have been recently formalized in a new risk minimization framework called performative prediction.

Title: Economic Networks: Robustness and Interventions Abstract: We are in a golden age of core ideas from network theory being applied to make progress on major questions in economics. This talk will present two examples of this from my work, both concerning the incentives of firms in economic production. In the first paper I'll discuss, we model the production of complex goods in a large supply network, where specific relationships, or contracts, between firms are essential. A supply network is called fragile if aggregate output is very sensitive to small aggregate shocks. We model both the mechanics and the strategic forces that determine network robustness. By building a new interface between percolation theory and canonical models of economic production, we show that supply networks of intermediate productivity are fragile in equilibrium, even though this is always inefficient. The second paper also focuses on a network of firms, but now the key interaction arises from how the demand for a firm's output is affected by other firms' choices: the firms sell goods that are either consumed together or are substitutes (competitors) for one another. The main questions are: how volatile are prices and measures of economic welfare to economic shocks (e.g., changes in commodity prices), and how should a planner intervene to improve the welfare a market delivers? Our main conceptual contribution is to use a certain basis of the goods space, determined by the network of interactions among suppliers. It consists of principal components. By developing a new approach for using this principal component perspective to analyze the market, we can give new and sharp answers to both main questions. The results permit us to leverage the theory of spectral approximation (recently attracting a lot of interest in the statistical analysis of big data) to design optimal interventions even when the demand system is observed with a lot of noise.

Abstract: Over the past couple of decades, the growth and ubiquity of the Internet, and social media in particular, has drastically changed the world. While things like the rapid spread of knowledge and instant communication have certainly had a positive impact, the Internet has also enabled the spread of disinformation and coordination of both online attacks and physical violence. In this talk, I present a data-driven understanding of the darker side of the Web. Among other things, I will show how we can measure the evolution and splintering of extremist groups over time and how conspiracy theories form and are disseminated online. I will also cover how extremist language changes and novel socio-technical attacks rapidly manifest in response to real-world events. Finally, I will discuss the nuanced relationship that science itself has with online jerks.

Samson Goddy is a software engineer who believes in changing the world in his way. He is an open-source advocate, primarily maintaining the Sugar Labs projects, the organization behind the Linux-flavored interface called Sugar Desktop. He is the co-founder of Open Source Community Africa, a project he cares about because it allows him to reflect his love for open source while building large projects that help the budding open-source community in Africa. He serves as a board member of the Open Source Collective, a project that helps sustain open-source projects through the open collective platform. He is on the board of directors of Sugar Labs, the new organization behind the Sugar Desktop project. He previously served on an oversight board with the organization, primarily focusing on growth and partnerships. Out of his open-source world, he consults with African governments on technology and has led programs for state and federal governments. He currently consults for the International Telecommunication Union, the African Union, and UN Women to bring more women into technology on the continent through the AGGCI project.

Title: Mathematical models of curiosity Abstract: What is curiosity? Is it an emotion? A behavior? A cognitive process? Curiosity seems to be an abstract concept—like love, perhaps, or justice—far from the realm of those bits of nature that mathematics can possibly address. However, contrary to intuition, it turns out that the leading theories of curiosity are surprisingly amenable to formalization in the mathematics of network science. In this talk, I will unpack some of those theories, and show how they can be formalized in the mathematics of networks. Then, I will describe relevant data from human behavior and linguistic corpora, and ask which theories that data supports. Throughout, I will make a case for the position that individual and collective curiosity are both network building processes, providing a connective counterpoint to the common acquisitional account of curiosity in humans. Reference: Perry Zurn & Dani S. Bassett. Curious Minds: The Power of Connection. MIT Press, 2022.

Title: Building a framework for population-scale network analysis Abstract: Traditional network analysis methods become unfeasible when addressing the structure of large-scale networks. First, it is computationally too expensive and even irrelevant to calculate centrality scores for networks of millions of people. Second, and more importantly, one-mode networks (and pairwise interactions) are insufficient as they disregard the crucial role that organizations play in forming social structures. Thus, a more realistic approach to general population structures is crucial to enhance data collection methods, study design in several fields, and policy-making.

Title: Diversity and heterogeneity can improve the efficiency of collective decisions Abstract: Many organisms regularly make decisions regarding foraging, home-site selection, mating, and danger avoidance in groups ranging from hundreds up to millions of individuals. These decisions involve evidence-accumulation processes by individuals and information exchange within the group. Moreover, these decisions take place in complex, dynamic, and spatially structured environments, which shape the flow of information between group mates. We will present a statistical inference model for framing evidence accumulation and belief sharing in groups and some examples of how interactions shape decision efficiency in groups. Our canonical model is of Bayesian agents deciding between two equally likely options by accumulating evidence to a threshold. First passage times and error rates can be accurately estimated using asymptotics for order statistics in the limit of large group sizes. When neighbors only share their decisions with each other, groups comprised of individuals with a distribution of decision thresholds make more efficient decisions than homogeneous ones. To conclude, we will briefly discuss specific examples of the impacts of spatial and communication heterogeneity on collective decision making in foraging animal groups like honey bees and primates.

Title: Something from nothing: transfer learning and the global structure of species interaction networks Abstract: Despite their importance in many ecological processes, collecting data and information on ecological interactions is an exceedingly challenging task. For this reason, large parts of the world have a data deficit when it comes to species interactions, and how the resulting networks are structured. As data collection alone is unlikely to be sufficient, community ecologists must adopt predictive methods. We present a methodological framework that uses graph embedding and transfer learning to assemble a predicted list of trophic interactions of a species pool for which their interactions are unknown. Specifically, we ‘learn’ the information (latent traits) of species from a known interaction network and infer the latent traits of another species pool for which we have no a priori interaction data based on their phylogenetic relatedness to species from the known network. The latent traits can then be used to predict interactions and construct an interaction network. Here we assembled a metaweb for Canadian mammals derived from interactions in the European food web, despite only 4% of common species being shared between the two locations. The results of the predictive model are compared against databases of recorded pairwise interactions, showing that we correctly recover 91% of known interactions. The framework itself is robust even when the known network is incomplete or contains spurious interactions making it an ideal candidate as a tool for filling gaps when it comes to species interactions. We provide guidance on how this framework can be adapted by substituting some approaches or predictors in order to make it more generally applicable.

Title: Systems thinking and structural explanations in responsible AI Abstract: Recently the concepts of trustworthy and responsible AI have gained prominence in industry, academia, government, and civil society as an AI governance rallying cry. These umbrella terms serve as useful, albeit often vacuous, ways to conceptually organize the myriad efforts to establish common standards, rules, and guardrails for the development and use of data-driven decision-making systems. The key gap in the efforts to make algorithmic systems---and the decision making they effect---responsible is an understanding of complex systems and social structures. Furthermore, this gap is revealed when algorithmic systems lead to unintended harms, where 'AI accidents' reveal underlying social structure. Drawing on examples from ranking systems, the mortgage market, algorithmic discovery and algorithmic accidents, I will discuss several recent efforts to bridge understandings of complex social systems with this emerging field.

Abstract: The idea of creating artifacts that evolve by themselves has been at the heart of the Artificial Life research, dating back to the early motives of John von Neumann’s monumental work on self-reproducing automata in the 1940’s. This vein of research is unique and fundamentally different from other more widely studied evolutionary computation research, because basic processes of evolution (heredity, variation, selection) are not given a priori as built-in mechanisms but they need to emerge as a result of interactions among microscopic components. In this talk, I will provide a brief review of how this problem has been approached in ALife using various kinds of methodologies, including classic frameworks (e.g., cellular automata, evolving programs) and more modern ones (e.g., artificial chemistry, AI/ML). I aim to highlight several key ingredients in order for complex systems to show spontaneous evolutionary behaviors by themselves and, in particular, to exhibit open-ended exploration of the possibility space. Link to slides: https://www.slideshare.net/HirokiSayama/how-to-make-things-evolve

Title: The wisdom of crowds in naturalistic conditions Abstract: The ‘wisdom of crowds’ (whereby groups of humans or animals make better decisions than individuals) has attracted a great deal of attention, inspiring many researchers to search for signals of it across many species and contexts. However, the model that this intuition is based on is highly simplistic and its assumptions are easily violated when real animals make decisions in real environments. Here, I describe a series of models that I’ve developed with collaborators that add in features of natural environments to models of collective decision-making. We find that these features tend to substantially alter, and sometimes reverse, our predictions of how the wisdom of crowds behaves in nature, and reveals several new mechanisms by which animals could potentially exploit to improve their decision accuracy when living in a group.

Title: Archaeoecology: Using the archaeological past to understand our present and future Abstract: Archaeology provides rich data of the past 60,000 years of human-environment interaction, yet it remains under-utilized for examining present ecosystems. However, modern methods can harness the explanatory power of the past to calibrate our understanding of the present and predict how we will face challenges in the future. In this vein approaches from complex adaptive systems science including agent-based modeling and network science prove particularly promising. By simulating societies in silico agent-based models and networks have enabled researchers to not only understand previously intractable aspects of the past, but also to use these simulations to predict what can make resilient societies and what lead them toward vulnerabilities to external perturbations. My work has used agent-based modeling, social network analysis, and trophic network analysis (or food web modeling) to examine robustness and vulnerabilities of societies from the American Southwest, to northern Mongolia, to Aboriginal Australia.In this talk I explore the unique ways that socio-environmental modeling can help us understand the lifeways of societies worldwide, and also suggest that understanding how people interacted in their uniquely challenging environments can provide parallels to understanding humanity’s position in ecosystems today. Only through applying a complexity lens can we truly understand how the actions and interactions of people led to the large overarching structures we see today.

Abstract: Science, technology, engineering, and mathematics (STEM) organizations, both in the government and the industry, are facing two daunting challenges. First, the acute failures in developing complex engineered artifacts, which are often manifested in cost and schedule overruns. The second is the lack of diversity in STEM fields. Despite a large fraction of STEM organizations report to have an active diversity policy, there remains a significant underrepresentation of women, racial, and ethnic minorities, in addition to often overlooked deep-level diversity features such as knowledge. While the open innovation (OI) literature hints at its potential to alleviate both of these issues, research on leveraging OI as a sociotechnical mechanism that could serve both as an effective systems architecting tool and as a more inclusive policy instrument for innovation, have been nascent. As such, it is unclear if the crowd can contribute to complex STEM problems with high-degrees of interdependencies or if the subset of the crowd that can deliver transferrable knowledge is actually more diverse compared to the internal workforce of the STEM agencies. This talk outlines an opportunity to re-think system design processes with an awareness of who will solve and how the solvers will engage in the design process. I will demonstrate that joint consideration of problem formulation, organizational knowledge, and outside expertise could significantly improve design process outcomes; and discuss strategies for incorporating solver-awareness into the architecting process. Then, I will present evidence from a unique NASA field experiment, probing into the efficacy and the extent of crowdsourcing to serve the diversity policy goals of STEM organizations. Finally, I will discuss how mixed-methods research approaches could help to develop a novel sociotechnical theory for architecting complex systems.

Talk Title: The Alignment Problem: Machine Learning and Human Values Talk Abstract With the incredible growth of machine learning over recent years has come an increasing concern about whether ML systems' objectives truly capture their human designers' intent: the so-called "alignment problem." Over the last five years, these questions of both ethics and safety have moved from the margins of the field to become arguably its most central concerns. The result is something of a movement: a vibrant, multifaceted, interdisciplinary effort that is producing some of the most exciting research happening today. Brian Christian, visiting scholar at UC Berkeley and author of the acclaimed bestsellers "The Most Human Human" and "Algorithms to Live By," will survey this landscape of recent progress and the frontier of open questions that remain. Short Bio Brian Christian is the author of the acclaimed bestsellers "The Most Human Human" and "Algorithms to Live By," which have been translated into nineteen languages. His most recent book, "The Alignment Problem," has just been published and has been named a Los Angeles Times Finalist for Best Science & Technology Book of the Year. Christian is a visiting scholar at the University of California, Berkeley, where he an affiliate of the Center for IT Research in the Interest of Society and the Center for Human-Compatible AI, and he lives in San Francisco.

Abstract: Self-organization offers a promising approach for designing adaptive systems. Given the inherent complexity of most cyber-physical systems, adaptivity is desired, as predictability is limited. Here I summarize different concepts and approaches that can facilitate self-organization in cyber-physical systems, and thus be exploited for design. Then I mention real-world examples of systems where self-organization has managed to provide solutions that outperform classical approaches, in particular related to urban mobility. Finally, I identify when a centralized, distributed, or self-organizing control is more appropriate.

Abstract: The world is facing a fisheries crisis, with fish stocks and marine ecosystems collapsing due to human over-exploitation. Understanding the interconnectedness among species in harvested ecosystems and the dynamic responses of ecosystems to fishing is critical for informing managing practices to attain fisheries’ sustainability. Two of our recent publications evaluate such interconnectedness among dozens of species in harvested ecosystems using network analysis, mathematical models and computational tools. First, we investigate the combined effects of artisanal fisheries and climate change on an intertidal food web of the Central Coast of Chile. We show that climate change has a stronger effect on the food web than artisanal fisheries. Second, we incorporate economic rules governing fish extraction based on fish price and yield to evaluate how economic dynamics affect food webs and cause species extinctions. Our work exemplifies the importance of studying the effects of fisheries on the entire food web, instead of only focusing on the target species, and of introducing humans as dynamic components into food web models to answer questions of fisheries sustainability.

Abstract: Network ecology is an emerging field allowing researchers to conceptualize and model complex and dynamic ecological systems. Here, I highlight the parallel developments in ecological network studies and spatial network applications and their convergence in the use of network theory. Then, I summarize how ecological, spatial, temporal, and spatio-temporal networks can be formalized into the framework of multilayer networks. Specifically, I present how the spatio-temporal network can be analysed focusing either on the dynamics on the network or of the network. Such distinctions provide the foundations for the formulation of an “ecological network dynamics hypothesis” (ENDH) stating key network topologies that constrain the dynamic of ecological systems. Such an ENDH could in turn permit assessing trade-offs in network topologies and functions that affect the persistence of ecological systems. I then propose the integration of ecological networks with spatial networks to improve our understanding of complex ecological systems. This integration combines filtering the ecological interaction network based on motifs with the delineation of patches in the spatial network using an edge detection algorithm scalable to species dispersal abilities. I conclude that network ecology with its analytical methods will improve our understanding on how ecological systems will respond in spatially dynamic landscapes.

Bio: Timnit Gebru, Expert Research Scientist and AI Ethicist Timnit Gebru is an expert research scientist and AI ethicist , working to reduce the potential negative impacts of AI. Timnit earned her doctorate under the supervision of Fei-Fei Li at Stanford University in 2017 and did a postdoc at Microsoft Research NYC in the FATE team. She is also the cofounder of Black in AI, a place for sharing ideas, fostering collaborations and discussing initiatives to increase the presence of Black people in the field of Artificial Intelligence.

Abstract: While most cells in the body have identical 1-D genomes, cells in different tissues have remarkably different cell shapes, internal structures and functionality. Recent studies suggest that the 3-D structure of the genome differs from cell to cell, and that these differences have profound effects on the function of the underlying DNA, impacting cell structure and functionality. Interestingly, viral infections also affect the 3-D organization of the genome. While the common paradigm suggests that chromosome compaction causes gene silencing, few mechanistic studies on full chromosomes have been performed to support this. Recently, the development of high throughput sequencing technology has produced precise information regarding gene-gene interactions. The Hi-C technique cross links chromosomes with themselves, revealing DNA-DNA interactions. These samples are then fragmented and ligated to produce small loops containing the DNA-to-DNA interaction sequences. Finally, a 2-D contact map is produced. While many suspect that patterns in the 2-D contact map reflect 3-D proximity of genes, this is difficult to discern. We use constrained molecular simulations to produce 3-D reconstructions of full chromosomes highly consistent with Hi-C experimental data for the process of X chromosome silencing, shedding light on the chromosome compaction. Changes in 3-D structure, could, in principle, be used to detect more subtle signatures of viral infections. In addition to full chromosome simulations, results from the first explicit solvent MD simulations of an entire gene locus (using the LANL Trinity supercomputer) will be presented. This pilot study is the first billion-atom simulation published in the biosciences. Simulations such as these have the ability to push HPC resources much more than conventional simulations in the physical sciences.

Abstract: Are the global properties of complex quantum systems simply a consequence of the properties of their individual components? Or are there features that emerge, as a result of self-organisation, uniquely characterising the complex object as a whole? The answer to these questions might unveil up-to-now unknown properties of molecular complexes, proteins, and living organisms, as well as of engineered complex quantum communication networks.

Abstract: Big data has opened up new possibilities and transformed the ways we conduct research in nearly every discipline. However, ethical considerations and education for research has long focused on human subjects, governed in the U.S. by institutional review boards. Data science often falls through the cracks of these regulations, so it is even more imperative that we have strong ethical norms and guidelines. This starts with the reminder that though data scientists may not interact directly with people, the data collected and analyzed very often comes from people, which opens up important considerations around privacy, consent, and harm. Moreover, applications of data science research, particularly with respect to prediction, have the potential for large-scale societal impacts. In considering the broad landscape of technology ethics when it comes to uses and applications of big data, I will argue for a fundamental shift in how we teach ethics to future data scientists and researchers.

Abstract: In a rapidly changing world, severely affected by extreme weather events, epidemic outbreaks, economic shocks and conflicts, it is of fundamental importance to understand where the most vulnerable people are, how many they are, and to identify what it is that makes them more vulnerable than others to these threats. During the last decade, research has shown that data such as digital traces, phone metadata and satellite imagery carry relevant information beyond their original purpose and can be used as a proxy to measure socio-economic characteristics and detect vulnerabilities when traditional data is not available. Following an overview of these studies, the talk will deep dive into the UN World Food Programme’s original work on predicting food security. We will then conclude by discussing challenges and limitations, but also opportunities, that come with these approaches.

Season 1 Episode 2 - August 31, 2020 Speaker: DJ Patil DJ Patil has held a variety of roles in Academia, Industry, and Government. He is Head of Technology for Devoted Health, a Senior Fellow at the Belfer Center at the Harvard Kennedy School, and an Advisor to Venrock Partners. Dr. Patil was appointed by President Obama to be the first U.S. Chief Data Scientist where his efforts led to the establishment of nearly 40 Chief Data Officer roles across the Federal government. He also established new health care programs including the Precision Medicine Initiative and the Cancer Moonshot, new criminal justice reforms including the Data-Driven Justice and Police Data Initiatives that cover more than 94 million Americans, as well as leading the national data efforts. He also has been active in national security and for his efforts was awarded by Secretary Carter the Department of Defense Medal for Distinguished Public Service which the highest honor the department bestows on a civilian. In industry, he led the product teams at RelateIQ which was acquired by Salesforce, was founding board member for Crisis Text Line which works to use new technologies to provide on demand mental and crisis support, and was a member of the venture firm Greylock Partners. He has also was Chief Scientist, Chief Security Officer and Head of Analytics and Data Product Teams at the LinkedIn Corporation where he co-coined the term Data Scientist. He has also held a number of roles at Skype, PayPal, and eBay. As a member of the faculty at the University of Maryland, his research focused on nonlinear dynamics and chaos theory and he helped start a major research initiative on numerical weather prediction. As an AAAS Science & Technology Policy Fellow for the Department of Defense, Dr. Patil directed new efforts to leverage social network analysis and the melding of computational and social sciences to anticipate emerging threats to the US. He has also co-chaired a major review of US efforts to prevent bioweapons proliferation in Central Asia and co-founded the Iraqi Virtual Science Library (IVSL). In 2104 he was selected by the World Economic Forum as a Young Global Leader and is also a Member of the Council of Foreign Relations. More details can be found on his LinkedIn profile: http://www.linkedin.com/in/dpatil and can be followed on twitter @dpatil

Abstract: Mapmakers, like historians, face a tension in their work between the ideal of objectivity and the recognition that their work will inevitably be political. Even when factually accurate, maps, like history, are shaped by their perspective, as well as countless decisions about what to include and what to leave out. This means that all maps will be biased. But it also means that simply pointing out this bias is only the first step toward trying to understand how and why they were created. In this talk, I draw on my experience with history, and the politics which surround the writing of it, to look at how maps can be used to understand the worldviews of the people who created them.

Abstract: Developing intervention strategies that can reduce transmission and alleviate the impacts of social distancing is an essential goal of long-term public health responses to the ongoing COVID-19 pandemic. In this talk I will describe our efforts to develop models of ‘shield immunity’, i.e., leveraging serological testing and the increased interaction rates of recovered individuals to reduce new chains of transmission and foster safer economic re-engagement. I will then discuss initial steps to translate these principles into actions, including connecting serological test analytics and developing feasibility models for implementation in pilot settings.