Adam Gleave’s Post

Loved the talk. I know category theory but never saw a software implementation side-by-side comparison as he does here using Python and Topos Institute's AlgebraicJulia. Highly recommend it to those that want to think about software engineering in a different way. His use of the triangle graph definitely reminds me of my current implementation of algebraic evaluation in the ntqr Python package. I know the implementation is brittle (coded for exactly three noisy classifiers) and getting the pieces to work has had its issues. I'm definitely going to think about how to do a general n-ensemble implementation using AlgebraicJulia.

⚙️ "Category Theory for Legible Software and Corrigible World Modeling" by Kristopher Brown presented at FAR Labs Seminar. Brown discusses the role of category theory in creating interpretable and adaptable AI systems, aiming for enhanced safety and stakeholder alignment. 🌟 🔍 Model Legibility: Emphasizes the need for AI models that are both interpretable and economically viable. 🔧 Mathematical Principles: Advocates for software engineering based on principled mathematical models for abstraction reuse. 🔄 Stakeholder Collaboration: Highlights the importance of diverse stakeholder collaboration to align AI systems with broad human values. 🧩 Examples of Abstractions: Showcases how databases, Petri nets, and wiring diagrams exemplify effective abstraction in software engineering. 📺 Watch the full recording:  https://lnkd.in/ghT4irux  -- and subscribe to our YouTube channel for future research presentations!

Is causality necessary for trustworthy and explainable systems? From fair bandits to counterfactual explanations and back. In her doctoral thesis, Limor Gultchin presents a comprehensive overview of the field of trustworthy & explainable machine learning. The paper introduces a bunch of novel theoretical and practical contributions: a methodology for identifying mediation paths, feature-set-independent measure of necessity and sufficiency (I think this one's really exciting!) and more. IMHO, a must read for anyone interested in explainability, fairness, and trustworthy system design. #causality #causalinference #fairness #machinelearning #explainability #trustworthyAI 💡 Paper: https://vist.ly/vhbu

Matt Beane and I sit down to discuss AI, learning, mentorship and the vital role that experimentation plays in the area of improved performance. His book, "The Skill Code" is a must read for anyone interested in the intersection of education and technology. Enjoy the pod!

Want to learn about boosting from a statistical standpoint: Check out our new article on Ensemble learning. Thanks Shantanu Awasthi, PhD for collaborating.

An interesting and non-obvious insight from learning science: Forgetting something is different from not understanding something. Here is a narrative from Andy Matuschak that beautifully illustrates the difference between forgetting and not understanding.

Mindsmith Thinking support by using our expertise to provide "insight' into interpretation. Developing interpretation is progress, and progress, is the path to better performance. Here we proved a practical cognitive assessment of politics and how opinion becomes truth: Politics is about messages based on information and opinion. Messages become narrative and influence. Influence becomes thinking and behaviour. The "political definition of truth", is better described by the use of he word narrative. Narrative, often supported by data, is convincing, but may not amount to truth. The progress we make is not confusing this definition as "truth" and seeing the use of data in this context. We remove "the veil, understand what we see, and act and direct our resources accordingly.

Mindsmith Thinking support by using our expertise to provide "insight' into interpretation. Developing interpretation is progress, and progress, is the path to better performance. Here we proved a practical cognitive assessment of politics and how opinion becomes truth: Politics is about messages based on information and opinion. Messages become narrative and influence. Influence becomes thinking and behaviour. The "political definition of truth", is better described by the use of he word narrative. Narrative, often supported by data, is convincing, but may not amount to truth. The progress we make is not confusing this definition as "truth" and seeing the use of data in this context. We remove "the veil, understand what we see, and act and direct our resources accordingly.

The initial introduction of AI will be similar to the “accept all cookies, manage cookies” prompts, which were initially met with skepticism but gradually became commonplace. As AI becomes more integrated into our lives, others will start knowing more about us than we do ourselves. This normalization of AI will lead to its widespread adoption. Eventually, companies, governments, and various sectors like defense, transportation, food, legal services, and machinery will fiercely compete to incorporate AI. Over time, we will not only get used to AI but also become dependent on it. just like we are now on internet. This dependency will bring about changes that we may not like but will ultimately accept and adapt to. We will change. now did we change in future or Ai changed us in the past ? 🤔

Some thought-provoking insights here from former Open AI board members Helen Toner and Tasha McCauley on the challenge of ensuring AI safety while maximising AI's unfolding benefits for humanity. Governance is always needed. #Governing AI