The AI Fundamentalists

Baseline modeling is a necessary part of model validation. In our expert opinion, it should be required before model deployment. There are many baseline modeling types and in this episode, we're discussing their use cases, strengths, and weaknesses. We're sure you'll appreciate a fresh take on how to improve your modeling practices.Show notesIntroductions and news: why reporting and visibility is a good thing for AI 0:03Spoiler alert: Providing visibility to AI bias audits does NOT mean exposing trade secrets. Some reports claim otherwise.Discussion about AI regulation in the context of current events and how regulation is playing out between Boeing and the FAA (tbc)Understanding baseline modeling for machine learning 7:41Establishing baselines allows us to understand how models perform relative to simple rules-based models, aka heuristics.Reporting results without baselines to compare against is like giving a movie a rating of 5 without telling the listener that you were using a 10-point scale.Baseline modeling comparisons are part of rigorous model validations and should always be conducted during early model development and final production deployment.Pairs with analyses of theoretical upper bounds for modeling performance to show how your technique scores between acceptable worst and best case performance.We often find complex models being deployed in the real world that haven’t proven their value over simpler and explainable baseline modelsClassification baselines and model performance comparison 19:40Uniform Random Selection - simulate how your model does against a baseline model that guesses classes randomly like a dice.Most Frequent Class (MFC) - the most telling test and often the most telling test in the case of highly skewed data with inappropriate metrics.Single-feature modeling - Validates how much the complex signal from your data and model improves over a bare minimum explainable model.And more…Exploring regression and more advanced baselines for modeling 24:11Regression baselines: mean, median mode, Single-variable linear regression, Lag 1, and Least 5% re-interpretationAdvanced baselines in language and visionConclusions 35:39Baseline modeling is a necessary part of model validationThere are differing flavors of baselines that are appropriate for all types of modelingBaselines are needed to establish fair and realistic lower bounds for performanceIf your model can’t perform significantly better than a baseline consider scrapping the model and trying a new approachWhat did you think? Let us know.Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics: LinkedIn - Episode summaries, shares of cited articles, and more. YouTube - Was it something that we said? Good. Share your favorite quotes. Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

In this episode, we explore information theory and the not-so-obvious shortcomings of its popular metrics for model monitoring; and where non-parametric statistical methods can serve as the better option. Introduction and latest news 0:03Gary Marcus has written an article questioning the hype around generative AI, suggesting it may not be as transformative as previously thought.This in contrast to announcements out of the NVIDIA conference during the same week.Information theory and its applications in AI. 3:45The importance of information theory in computer science, citing its applications in cryptography and communication.The basics of information theory, including the concept of entropy, which measures the uncertainty of a random variable.Information theory as a fundamental discipline in computer science, and how it has been applied in recent years, particularly in the field of machine learning.The speakers clarify the difference between a metric and a divergence, which is crucial to understanding how information theory is being misapplied in some casesInformation theory metrics and their limitations. 7:05Divergences are a type of measurement that don't follow simple rules like distance, and they have some nice properties but can be troublesome in certain use cases.KL Divergence is a popular test for monitoring changes in data distributions, but it's not symmetric and can lead to incorrect comparisons.Sid explains that KL divergence measures the slight surprisal or entropy difference between moving from one data distribution to another, and is not the same as KS test.Metrics for monitoring AI model changes. 10:41The limitations of KL divergence and its alternatives, including Jenson Shannon divergence and population stability index.They highlight the issues with KL divergence, such as asymmetry and handling of zeros, and the advantages of Jenson Shannon divergence, which can handle both issues, and population stability index, which provides a quantitative measure of changes in model distributions.The popularity of information theory metrics in AI and ML is largely due to legacy and a lack of understanding of the underlying concepts. Information theory metrics may not be the best choice for quantifying change in risk in the AI and ML space, but they are the ones that are commonly used due to familiarity and ease of use.Using nonparametric statistics in modeling systems. 15:09Information theory divergences are not useful for monitoring production model performance, according to the speakers.Andrew Clark highlights the advantages of using nonparametric statistics in machine learning, including distribution agnosticism and the ability to test for significance without knowing the underlying distribution.Sid MangaliWhat did you think? Let us know.Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics: LinkedIn - Episode summaries, shares of cited articles, and more. YouTube - Was it something that we said? Good. Share your favorite quotes. Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

In this episode, the hosts focus on the basics of anomaly detection in machine learning and AI systems, including its importance, and how it is implemented. They also touch on the topic of large language models, the (in)accuracy of data scraping, and the importance of high-quality data when employing various detection methods. You'll even gain some techniques you can use right away to improve your training data and your models.Intro and discussion (0:03)Questions about Information Theory from our non-parametric statistics episode.Google CEO calls out chatbots (WSJ)A statement about anomaly detection as it was regarded in 2020 (Forbes)In the year 2024, are we using AI to detect anomalies, or are we detecting anomalies in AI? Both? Understanding anomalies and outliers in data (6:34)Anomalies or outliers are data that are so unexpected that their inclusion raises warning flags about inauthentic or misrepresented data collection. The detection of these anomalies is present in many fields of study but canonically in: finance, sales, networking, security, machine learning, and systems monitoringA well-controlled modeling system should have few outliersWhere anomalies come from,  including data entry mistakes, data scraping errors, and adversarial agents Biggest dinosaur example: https://fivethirtyeight.com/features/the-biggest-dinosaur-in-history-may-never-have-existed/Detecting outliers in data analysis (15:02)High-quality, highly curated data is crucial for effective anomaly detection. Domain expertise plays a significant role in anomaly detection, particularly in determining what makes up an anomaly.Anomaly detection methods (19:57)Discussion and examples of various methods used for anomaly detection Supervised methodsUnsupervised methodsSemi-supervised methodsStatistical methodsAnomaly detection challenges and limitations (23:24)Anomaly detection is a complex process that requires careful consideration of various factors, including the distribution of the data, the context in which the data is used, and the potential for errors in data entryPerhaps we're detecting anomalies in human research design, not AI itself?A simple first step to anomaly detection is to visually plot numerical fields. "Just look at your data, don't take it at face value and really examine if it does what you think it does and it has what you think it has in it." This basic practice, devoid of any complex AI methods, can be an effective starting point in identifying potential anomalies.What did you think? Let us know.Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics: LinkedIn - Episode summaries, shares of cited articles, and more. YouTube - Was it something that we said? Good. Share your favorite quotes. Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

We're taking a slight detour from modeling best practices to explore questions about AI and consciousness. With special guest Michael Herman, co-founder of Monitaur and TestDriven.io, the team discusses different philosophical perspectives on consciousness and how these apply to AI. They also discuss the potential dangers of AI in its current state and why starting fresh instead of iterating can make all the difference in achieving characteristics of AI that might resemble consciousness. Show notesWhy consciousness for this episode?Enough listeners have randomly asked the hosts if Skynet is on the horizonDoes modern or future AI have the wherewithal to take over the world, and is it even conscious or intelligent? Do we even have a good definition of consciousness?Introducing Michael Herman as guest speakerCo-founder of Monitaur, Engineer extraordinaire, and creator of TestDriven.io, a training company that focuses on educating and upskilling mid-level senior-level web developers.Degree and studies in philosophy and technologyEstablishing the philosophical foundation of consciousnessConsciousness is around us everywhere. It can mean different things to different people.Most discussion about the subject bypasses the Mind-Body Problem and a few key theories:Dualism - the mind and body are distinctMaterialism - matter is king and consciousness arises in complex material systemsPanpsychism - consciousness is king. It underlies everything at the quantum levelThe potential dangers of achieving consciousness in AIWhile there is potential for AI to reach consciousness, we're far from that point. Dangers are more related to manipulation and misinformation, rather than the risk of conscious machines turning against humanity.The need for a new approach to developing AI systemsThere's a need to start from scratch if the goal is to achieve consciousness in AI systems.Current modeling techniques might not lead to AI achieving consciousness. A new paradigm might be required.There's a need to define what consciousness in AI means and to develop a test for it. Final thoughts and wrap-upIf consciousness is truly the goal, the case for starting from scratch allows for fairness and ethics to be established foundationallyAI systems should be built with human values in mindWhat did you think? Let us know.Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics: LinkedIn - Episode summaries, shares of cited articles, and more. YouTube - Was it something that we said? Good. Share your favorite quotes. Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

Data scientists, researchers, engineers, marketers, and risk leaders find themselves at a crossroads to expand their skills or risk obsolescence. The hosts discuss how a growth mindset and "the fundamentals" of AI can help.Our episode shines a light on this vital shift, equipping listeners with strategies to elevate their skills and integrate multidisciplinary knowledge. We share stories from the trenches on how each role affects robust AI solutions that adhere to ethical standards, and how embracing a T-shaped model of expertise can empower data scientists to lead the charge in industry-specific innovations.Zooming out to the executive suite, we dissect the complex dance of aligning AI innovation with core business strategies. Business leaders take note as we debunk the myth of AI as a panacea and advocate for a measured, customer-centric approach to technology adoption. We emphasize the decisive role executives play in steering their companies through the AI terrain, ensuring that every technological choice propels the business forward, overcoming the ephemeral allure of AI trends. Suggested courses, public offerings:Undergrad level Stanford course (Coursera):  Machine Learning SpecializationGraduate-level MIT Open Courseware: Machine LearningWe hope you enjoy this candid conversation that could reshape your outlook on the future of AI and the roles and responsibilities that support it.Resources mentioned in this episodeLinkedIn's jobs on the rise 20243 questions to separate AI from marketing hypeDisruption or distortion? The impact of AI on future operating modelsThe Obstacle is the Way by Ryan HolidayWhat did you think? Let us know.Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics: LinkedIn - Episode summaries, shares of cited articles, and more. YouTube - Was it something that we said? Good. Share your favorite quotes. Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

Get ready for 2024 and a brand new episode! We discuss non-parametric statistics in data analysis and AI modeling. Learn more about applications in user research methods, as well as the importance of key assumptions in statistics and data modeling that must not be overlooked, After you listen to the episode, be sure to check out the supplement material in Exploring non-parametric statistics.Welcome to 2024  (0:03)AI, privacy, and marketing in the tech industryOpenAI's GPT store launch. (The Verge)Google's changes to third-party cookies. (Gizmodo)Non-parametric statistics and its applications (6:49)A solution for modeling in environments where data knowledge is limited.Contrast non-parametric statistics with parametric statistics, plus their respective strengths and weaknesses.Assumptions in statistics and data modeling (9:48)The importance of understanding statistics in data science, particularly in modeling and machine learning. (Probability distributions, Wikipedia)Discussion about a common assumption of representing data with a normal distribution; oversimplifies complex real-world phenomena.The importance of understanding data assumptions when using statistical modelsStatistical distributions and their importance in data analysis (15:08)Discuss the importance of subject matter experts in evaluating data distributions, as assumptions about data shape can lead to missed power and incorrect modeling. Examples of different distributions used in various situations, such as Poisson for wait times and counts, and discrete distributions like uniform and Gaussian normal for continuous events.Consider the complexity of selecting the appropriate distribution for statistical analysis; understand the specific distribution and its properties.Non-parametric statistics and its applications in data analysis (19:31)Non-parametric statistics are more robust to outliers and can generalize across different datasets without requiring domain expertise or data massaging.Methods rely on rank ordering and have less statistical power compared to parametric methods, but are more flexible and can handle complex data sets better.Discussion about the usefulness and limitations, which require more data to detect meaningful changes compared to parametric tests.Non-parametric tests for comparing data sets (24:15)Non-parametric tests, including the K-S test and chi-square test, which can compare two sets of data without assuming a specific distribution.Can also be used for machine learning, classification, and regression tasks, even when the underlying datWhat did you think? Let us know.Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics: LinkedIn - Episode summaries, shares of cited articles, and more. YouTube - Was it something that we said? Good. Share your favorite quotes. Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

AI regulation, data privacy, and ethics in 2023 are discussed, including their impact on innovation. The tension in NLP research between AI safety and OpenAI's approach is explored. The importance of ethics in AI and the challenges it presents are highlighted.

Joshua Pyle joins us in a discussion about managing bias in the actuarial sciences. Together with Andrew's and Sid's perspectives from  both the economic and data science fields, they deliver an interdisciplinary conversation about bias that you'll only find here.OpenAI news plus new developments in language models. 0:03The hosts get to discuss the aftermath of OpenAI and Sam Altman's return as CEOTension between OpenAI's board and researchers on the push for slow, responsible AI development vs fast, breakthrough model-making.Microsoft researchers find that smaller, high-quality data sets can be more effective for training language models than larger, lower-quality sets (Orca 2).Google announces Gemini, a trio of models with varying parameters, including an ultra-light version for phones Bias in actuarial sciences with Joshua Pyle, FCAS. 9:29Josh shares insights on managing bias in Actuarial Sciences, drawing on his 20 years of experience in the field.Bias in actuarial work defined as differential treatment leading to unfavorable outcomes, with protected classes including race, religion, and more.Actuarial bias and model validation in ratemaking. 15:48The importance of analyzing the impact of pricing changes on protected classes, and the potential for unintended consequences when using proxies in actuarial ratemaking.Three major causes of unfair bias in ratemaking (Contingencies, Nov 2023)Gaps in the actuarial process that could lead to bias, including a lack of a standardized governance framework for model validation and calibration.Actuarial standards, bias, and credibility. 20:45Complex state-level regulations and limited data pose challenges for predictive modeling in insurance.Actuaries debate definition and mitigation of bias in continuing education.Bias analysis in actuarial modeling. 27:16The importance of identifying dislocation analysis in bias analysis.Analyze two versions of a model to compare predictive power of including vs. excluding protected class (race).Bias in AI models in actuarial field. 33:56Actuaries can learn from data scientists' tendency to over-engineer models.Actuaries may feel excluded from the Big Data era due to their need to explain their methodsStandardization is needed to help actuaries identify and mitigate bias.Interdisciplinary approaches to AI modeling and governance. 42:11Sid hopes to see more systematic and published approaches to addressing bias in the data science field.Andrew emphasizes the importance of interdisciplinary collaboration between actuaries, data scientists, and economists to create more accurate and fair modeling systems.Josh agrees and highlights the need for better governance structures to support this collaboration, citing the lack of good journals and academic silos as a chaWhat did you think? Let us know.Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics: LinkedIn - Episode summaries, shares of cited articles, and more. YouTube - Was it something that we said? Good. Share your favorite quotes. Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

Episode 9. Continuing our series run about model validation. In this episode, the hosts focus on aspects of performance, why we need to do statistics correctly, and not use metrics without understanding how they work, to ensure that models are evaluated in a meaningful way.AI regulations, red team testing, and physics-based modeling. 0:03The hosts discuss the Biden administration's executive order on AI and its implications for model validation and performance.Evaluating machine learning models using accuracy, recall, and precision. 6:52The four types of results in classification: true positive, false positive, true negative, and false negative.The three standard metrics are composed of these elements: accuracy, recall, and precision.Accuracy metrics for classification models. 12:36Precision and recall are interrelated aspects of accuracy in machine learning.Using F1 score and F beta score in classification models, particularly when dealing with imbalanced data.Performance metrics for regression tasks. 17:08Handling imbalanced outcomes in machine learning, particularly in regression tasks.The different metrics used to evaluate regression models, including mean squared error.Performance metrics for machine learning models. 19:56Mean squared error (MSE) as a metric for evaluating the accuracy of machine learning models, using the example of predicting house prices.Mean absolute error (MAE) as an alternative metric, which penalizes large errors less heavily and is more straightforward to compute.Graph theory and operations research applications. 25:48Graph theory in machine learning, including the shortest path problem and clustering. Euclidean distance is a popular benchmark for measuring distances between data points. Machine learning metrics and evaluation methods. 33:06Model validation using statistics and information theory. 37:08Entropy, its roots in classical mechanics and thermodynamics, and its application in information theory, particularly Shannon entropy calculation. The importance the use case and validation metrics for machine learning models.What did you think? Let us know.Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics: LinkedIn - Episode summaries, shares of cited articles, and more. YouTube - Was it something that we said? Good. Share your favorite quotes. Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

Episode 8. This is the first in a series of episodes dedicated to model validation. Today, we focus on model robustness and resilience. From complex financial systems to why your gym might be overcrowded at New Year's, you've been directly affected by these aspects of model validation.AI hype and consumer trust (0:03) FTC article highlights consumer concerns about AI's impact on lives and businesses (Oct 3, FTC)Increased public awareness of AI and the masses of data needed to train it led to increased awareness of potential implications for misuse.Need for transparency and trust in AI's development and deployment.Model validation and its importance in AI development (3:42)Importance of model validation in AI development, ensuring models are doing what they're supposed to do.FTC's heightened awareness of responsibility and the need for fair and unbiased AI practices.Model validation (targeted, specific) vs model evaluation (general, open-ended).Model validation and resilience in machine learning (8:26)Collaboration between engineers and businesses to validate models for resilience and robustness.Resilience: how well a model handles adverse data scenarios.Robustness: model's ability to generalize to unforeseen data.Aerospace Engineering: models must be resilient and robust to perform well in real-world environments.Statistical evaluation and modeling in machine learning (14:09)Statistical evaluation involves modeling distribution without knowing everything, using methods like Monte Carlo sampling.Monte Carlo simulations originated in physics for assessing risk and uncertainty in decision-making.Monte Carlo methods for analyzing model robustness and resilience (17:24)Monte Carlo simulations allow exploration of potential input spaces and estimation of underlying distribution.Useful when analytical solutions are unavailable.Sensitivity analysis and uncertainty analysis as major flavors of analyses.Monte Carlo techniques and model validation (21:31)Versatility of Monte Carlo simulations in various fields.Using Monte Carlo experiments to explore semantic space vectors of language models like GPT.Importance of validating machine learning models through negative scenario analysis.Stress testing and resiliency in finance and engineering (25:48)Importance of stress testing in finance, combining traditional methods with Monte Carlo techniques.Synthetic data's potential in modeling critical systems.Identifying potential gaps and vulnerabilities in critical systems.Using operations research and model validation in AI development (30:13)Operations research can help find an equilibrium in overcrowding in gyms.Robust methods for solving complex problems in logistics and hWhat did you think? Let us know.Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics: LinkedIn - Episode summaries, shares of cited articles, and more. YouTube - Was it something that we said? Good. Share your favorite quotes. Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.