Scientific Research and Writing 4 16 Good Science Practice by Florian Michahelles

What is science? In this video, we will find out how scientific paradigms evolve. Kun shows us that science isn't just about fence. It's also about frameworks that change over time. Then we learn about cold firearm. who argued against the idea of fixed scientific methods. His famous claim anything goes challenged us to think more openly about how knowledge is created. Finally, we will review core scientific concepts like validity, reliability, repusibility, stringency, conclusiveness, and how science sometimes violates its own ideals and yet still makes progress. Whether you're building systems or testing algorithms, understanding how science actually operates helps you become more critical thinker and better computer scientist. Thomas spent much of his career studying how science actually develops over time. He passed away in 1996, but his ideas still shape how we think about science today. He challenged the traditional view that science progresses in straight cumulative line simply stacking up facts and discoveries. Instead, argued that science moves in cycles. illustrated this with examples like the shift from the pitilomeic the geocentric earth in the center to the copernican heliocentric where the sun is in the center astronomy or the transition from the Newtonian physics to Einstein's relativity theory or the emergence of quantum mechanics which broke classical assumptions in all these revolutions entire worldviews were overturned what made Coon's view radical was his claim that these shifts aren't driven by neat logical rules like falsification. Instead, they involve deep changes in perspective, often messy or contested and without clear path. So, science in Coon's view isn't just methodical. It's also historical, social, and sometimes even dramatic. So, let's have close look at his cycle. suggested science progresses within paradigm and shifts revolutions. How does science change over time? According to philosopher Thomas science doesn't just move forward step by step, but evolves through cycles. It starts with what he called normal science. This is when scientists work with an accepted framework or paradigm, solving problems, building on shared assumptions, and refining what's already known. for example, if then else rules. But eventually, science begin to notice anomalies, results that don't fit the paradigm. At first, they try to fix or explain them. But when too many contradictions pile up and confidence in the existing paradigm starts to weaken, the model drifts. Example, natural language processing cannot be solved by rules. This leads to model crisis. During this phase, alternative ideas emerge. These may seem radical or even strange at first. Example realizing that AI symbolic rules no longer work. But if one of them explains the anomalies better, it can trigger model revolution. For example, machine learn based approaches proved to solve natural language processing. Eventually, new paradigm takes over. The rules, assumptions, and questions all shift, and new phase of normal science begins. Now under the new framework, for example, neural networks, transformers, reinforcement learning are taking over. So science, argues, doesn't progress in straight line. It moves in cycles, stable periods, crisis, and revolutions as our understanding of the world deepens and evolves. Understanding this cycle reminds us even in computer science major breakthroughs often come not from small fixes but from challenging the assumption we take for granted. Paul fighter said even coy is too rich science doesn't always follow coherent structure. In real practice scientists often break rules mix methods or borrow ideas across paradigms. The only principle that does not inhibit progress is anything goes. Paul Faden says, "No paradigm is ever completely irrationally superior to another." Shifts often involve non-scientific elements like persuasion, politics, or personal taste. He blurs the line between science, art, and politics. Some say he undermines the authority or credibility of science, potentially fueling pseudocience or anti-science sentiment. Others argue that without some method or standard, science loses its coherence and reliability. To understand how good science works, we need to get familiar with some core concepts that help us judge the quality of research and experiments. By the way, this will resurface also when we talk about peer review much later. First, there's validity. This is about whether an experiment actually measures what it claims to measure. For example, if we isolate the variable of mass and measure the time it takes for objects to fall, are we really testing the effect of mass on fall time or something else? Next is reliability. That's about consistency. If you repeat the experiment using different objects of various masses and get the same result, the experiment is considered reliable. Closely related is reproducibility. This means that other scientists should be able to perform the same experiment under the same conditions and get the same outcome. Science gets credibility when results can be reproduced independently. Then we have stringency. The level of control and precision applied in the experiment. Take Galileo's experiment. He carefully controlled the mass of falling spheres but measured time by eye and didn't account for air resistance. It was insightful but not highly stringent by today's standards. And finally, there's conclusiveness. How strongly the results support conclusion. Despite its limitation, Galileo's experiment was conclusive in overturning Aristotle's flawed belief that heavier objects fall faster. These five concepts, validity, reliability, reproducibility, stringency, and conclusiveness, helped us assess how trustworthy and meaningful scientific findings really are. These criteria are used when reviewing and assessing the quality of scientific reports. While science is built on trust, transparency, and rigor, it doesn't always live up to those ideals. Let's take look at some common violations of good scientific practice and why they matter, especially in fields like machine learning and AI. First, data fabrication. This is the outright invention of data. In AI, this can happen subtly. For example, when deep learning demos only showcase cherrypicked outputs like perfect GAN generated images, it misleads others by hiding the true behavior or limitations of the model. Next, hacking and harking that stands for hypothesizing after the results are known. common form in machine learning is tuning hyperparameters directly on the test set. This inflates performance and leads to biased results, quiet but widespread issue, especially when benchmarks drive publications. Then there's plagiarism in computer science. This often shows up as copied code reused from academic papers or open source projects. sometimes found in GitHub AI bots without proper credit. It undermines both trust and originality. Selective reporting is another problem. Researchers may only report results or well balanced or data sets, leaving out poor performance on harder real world data. This distorts the scientific record and skews the fairness and robustness debate. Gold authorship is more subtle. Some white papers, especially from big tech, are written by in-house engineers but published under well-known academic names. This blurs lines of responsibility and heights who related the work. Finally, reproducibility or the lack of it. Too many AI papers can't be independently verified due to missing code, undocumented hyperparameters, or restricted data sets. This has led to stronger code and data submission policy at top machine learning conferences. Violations like these don't just break rules, they damage trust and slow progress. For you, understanding and avoiding these pitfalls is part of doing good, responsible research. You should not forget to acknowledge that science is powerful tool, but it's not all powerful. They're question it simply isn't designed to answer. Science explains what it can observe, test, and measure. Science can analyze the physical properties of art, but it cannot tell us whether something is beautiful or not. Aesthetics are subjective and fall outside the realm of scientific inquiry. Science can explore the physical processes of life, but it cannot definitely answer the questions about the meaning of life or the purpose of the universe. Science can inform discussions about the consequences of actions, but it cannot dictate what is morally right or wrong. It doesn't provide basis for making ethical judgments. These are deep human questions and they belong to philosophy, religion or personal reflection rather than science. As novel laurate Sir Peter Meadow once said, "There is no limit upon the power of science to answer questions of the kind science can answer. The key is knowing which kinds of questions those are. In Douglas Adams Hitchhiker's Guide to the Galaxy 42 is the answer to the ultimate question of life, the universe, and everything. But the actual challenge is to produce the ultimate question. And that's also what science is about. Asking good questions and following good practices. So let's wrap up with few key takeaways. Thomas Kon taught us that science doesn't just grow steadily. It shifts through paradigm changes where old models are replaced by new ones. Paul Firearm went further arguing that sometimes in science anything goes. Real progress may come from breaking the rules. This reminds us that science needs to build structure and logic and requires review processes to evaluate potential violations when turning them into remissions of the work. And finally, science has its limits. It's powerful in explaining the observable world, but it can't fully answer deeper philosophical or existential questions. Understanding both the power and boundaries of science helps us use it more wisely.