Categories
Reading List EN

RL #044: A Reading List to Demystify Machine Learning

Machine learning shapes much of our world, yet many of us understand little about how it works. We remain in the dark about its mechanics. This reading list helps demystify the algorithms that are transforming our future.

In this Oikoplus Reading List, we take a look at machine learning. Of course, we also use various AI tools to support us in our day-to-day work. But to be honest, we have little understanding of how their algorithms work. And since we’re not alone in this, here are a few good reads on the subject. 

Let’s start with a look at the history of software. The article “A Little History of Software Development” on the Ferrovial blog provides a concise history of the evolution of software development. It explores key phases from the early years when programming was informal and experimental, through the industrialization process in the 1980s with structured methods, to the Agile revolution in the 2000s that restored a focus on creativity and collaboration. The piece concludes by discussing modern trends, like AI and low-code platforms, that could transform the industry.

The blog of the MIT Business School has an introduction to machine learning that is well worth reading. The title says it all: The article “Machine Learning, Explained” provides a comprehensive overview of machine learning (ML), its role as a critical subset of artificial intelligence (AI), and its growing impact across industries. It explains how ML enables computers to learn from data without being explicitly programmed and highlights different types of ML (supervised, unsupervised, and reinforcement learning). The article also addresses key issues such as explainability, bias, and the importance of ethical AI.

Quanta Magazine has published an article that is really worth reading. It shows how abstract concepts can influence the way we think about computers, and that computer science is not just about programming languages and technical solutions. The article discusses Lenka Zdeborová, a computer scientist who applies concepts from statistical physics to better understand machine learning and algorithm behavior. She explores how the physics of phase transitions, like water freezing, can model algorithmic changes, particularly in neural networks. By bridging physics and computer science, Zdeborová aims to decode the fundamental properties of large machine learning systems, aspiring to develop a framework similar to thermodynamics for machine learning.

Of course, machine learning is not merely about curiosity-driven innovation, but also a multi-billion dollar business. Matthew Ball provides a good insight into the questions this raises for tech companies. His article “Parallel Bets, Microsoft, and AI Strategies” explores Microsoft’s strategic approach to AI investments, particularly its partnerships with OpenAI, while simultaneously developing its own models. Microsoft has placed parallel bets on different AI technologies, from large acquisitions like Nuance to smaller investments in startups. The piece outlines the competitive landscape between Microsoft and OpenAI and examines the company’s historical strategy of making diversified bets to secure its position in emerging technologies.

Now we’ve looked into machine learning and AI. But we have hardly really understood how the self-learning algorithms work. But wouldn’t that be important? Won’t we all become mere users if we have no idea how the technologies we use in our everyday lives work? This question was asked almost a decade ago. This Brookings article from 2016 discusses the importance of educating the public about machine learning (ML) to increase understanding and trust in AI technologies. It emphasizes that while machine learning algorithms are increasingly influencing various sectors, including autonomous vehicles and facial recognition, the public often lacks knowledge about how these systems make decisions. The article advocates for more transparency and better communication strategies to demystify ML and its impacts on everyday life.

At the end of our little exploration of machine learning, it’s clear that whether it’s corporate strategies like those of tech companies or the need for public AI education, understanding these algorithms is essential to navigating the digital world that is shaping our future. We will continue reading.

Categories
Reading List EN Uncategorized

RL #031: Artificial Intelligence in Science

Many tips have been shared over the past weeks and months. This one is the perfect AI for research, and the other is the perfect AI for editing texts. Ideas for the best prompts for semantics-based generative AIs are flooding Twitter, Reddit, and the like. In this Reading List, we don’t want to give tips on which AI can be used for what. For a reading list, that doesn’t make much sense at the moment, not least because of the fast pace of technological development. We also don’t want to report on how AI can have an impact on science communication. We already did that last summer in Reading List #021. Rather, we have collected a few texts on thoughts about how AI could change science in the coming years. Enjoy reading!

Artificial intelligence with an overview

One of the biggest challenges of science, regardless of discipline, is keeping up with the flood of articles. 70,000 publications deal with the protein p53, according to the think tank Enago. This is the first I’ve heard of it today. Apparently, it is relevant for the early detection of tumors. In 1993, it was voted “Molecule of the Year”. On the occasion of this anniversary, an AI of my choice finds the following review: “The first 30 years of p53: growing ever more complex” by Arnold J Levine and Moshe Oren (paywall). In fact, there are now a number of tools that claim to find articles and present them in their respective publication context. The start has been made.

Disruptive Artificial Intelligence

With the newly gained overview, the quality of results and outcomes can also be reclassified. And this also applies outside of science. In an interview with Digitale Welt, Prof. Mario Trapp, director of the Fraunhofer Institute for Cognitive Systems IKS, remarks: “Even if you can still have the results of AI checked for plausibility by doctors today, this will hardly be possible in the future because of the increasing complexity.” The choice of words is exciting: Trained people can still check the plausibility of results. This will probably no longer be possible for a long time.

As a new key technology with a broad spectrum of applications (even if all references and points of reference so far point to medicine), universities are now facing investment hype for the third time since the 1950s and 1970s. This time, multidisciplinary research in step with action (i.e. industry) and politics is particularly in demand. At least that is the argument of Y. K Dwivedi et al. in an opinion paper published in the International Journal of Information Management. More applied, and with a focus on the extent to which the greatly altered interests brought about by AI interact with media, industry, and research, G. Berman, K. Williams, and S. Michalska argue in their study that research in the field of artificial intelligence functions differently than in other fields.

Proactive Artificial Objectivity

AIs help to keep track of things, they flush new money into the universities’ coffers. Overwhelmed, I return to medicine and to an article from 2018. On the Science Blog – Kaleidoscope for Science, Norbert Bischofberger wrote a fascinating article entitled “With artificial intelligence to a proactive medicine? A question that applies in a modified form to all disciplines today?

At that time, Bischofberger concluded that we might soon no longer “react” but proactively take care of ourselves. Five years later, knowledge production could soon be taken proactively into the hands of AIs. The question is whether an objective understanding of science will play into our hands. We will see.