January 2021 – SE metrics (Software Engineering)

War and algorithm

Amazon.com: War and Algorithm (9781786613646): Liljefors, Max, Noll, Gregor, Steuer, Daniel: Books

Understanding legal aspects of modern autonomous systems requires a philosophical and practical discourse. On the one hand, we need to understand what legal responsbility means in the context of autonomous systems. We need to understand who is responsible for the actions of the system, what the actions are and whether the system actually reacted as designed vs. whether new behaviour occurred.

On the other hand, we also need to understand that the introduction of the autonomous systems changes the legal systems. Autonomous systems do not require operators and therefore they are capable of interacting with each other. The notion of conflicts, damage and collateral damage get completely new dimensions.

I’ve picked up this book because I’ve had the possibility to work with one of the authors and met one more at a dinner a while back. They got me interested in the legal aspects of autonomous systems. In their book, the authors discuss various aspects of such systems. They start from the foundation of the legality of conflicts and then they move over to modern warfare. They provide historical examples of how the legal systems were (and are) shaped by the so-called LAWS (Lethal Autonomous Warfare Systems).

I particularly like the aspects related to the design of the systems and the fact that in chapter 4, the authors discuss the process of learning from the machine, or the algorithm. They call it the process of debugging, which is a new way of looking at the concept of understanding algorithms.

What I miss in the book, however, is the discussion on the quality of the AI systems. Although it is not explicit, it seems to me that the authors assume that an AI system is perfect, makes no mistakes dues to design defects (bugs). If this assumption is true, then it the discussion about the responsibility is a bit simpler, because we do not recognize the problems where an individual (a programmer) gives his/her best, but the testers or others in his team make mistakes. So, the responsibility is not on an individual (programmer, tester, architect), but on the entire company.

Either way, I’m happy that I had the possibility to listen to some of the authors and to work with them.

Law for Computer Scientists and other folks (review)

Law for Computer Scientists (pubpub.org)

Recently, a colleague of mine has recommended me this book. At first, I thought it would be a bit like “Law for dummies”, but it turned out to be much better than I actually thought.

The book is about how we, as software engineers, should look at the legal systems. It poses more questions than it actually answers, but it provides a number of great examples.

I sincerely recommend this book. The following parts have captured my attention:

Existence of different types of law and jurisdictions: national, international and supernational. Data and computer programs are perfect examples of different jurisdictions and the fact that different types of laws apply.
What constitutes data, meta-data and sensitive data. In Chapter 5, the authors mention that we cannot process sensitive data very easily, e.g. data about religion, gender, etc. Then, how can we make the systems fair and unbiased if we cannot process this kind of data?
Cybercrimes and how to deal with them. The author provides great examples of legislation that is supposed to help to fight cybercrime.

However, the best is always left for last and this book is no exception. The author provides a great discussion on the future of our legal systems. She does that by discussing the concept of personhood for AI or any other complex system. Although it sounds like a distant future, it is closer than we think. EU has already started to work on this kind of legislation.

Finally, I love the fact that the author brings in the three laws of robotics by Asimov – a real connection to computer science and software engineering.

Understanding what’s going on helps you become a better software developer…

10.1109/MS.2020.3014223

I’m a big fan of the Matrix movies, but well, to be honest, who isn’t:) I like the scene where Morpheus gives Neo the choice of two pills – one to know the truth and the other one to go on living his life as previously.

Well, sometimes I feel the same when I do my programming tasks – do I really want to know what the code does, or just make a quick fix and move on? I would say that it’s 50-50 for me – sometimes I feel like contributing and sometimes I just fix the problem and move on.

In this paper, the authors conduct an experiment to understand how and when software developers make mistakes. They find that “[the] study suggests that a relatively high number of mistakes are related to communicating with stakeholders outside of the development team. “

Having worked with metrics teams all over the globe, I’ve noticed that the communication with the stakeholders is often the largest problem that you can have. The stakeholders don’t speak “requirements” and we do not understand “wants” of the stakeholders. But, well, it’s not what the paper is about.

What I like about the paper is the systematic approach to the study – using experiments and a technique for teaching the developers how to work with their limitations. This is what the authors recommend as remedies (quoted directly from the paper):

Know your own weaknesses. Every developer is different and struggles with different concepts. Our analysis shows a variety of types of errors that developers make. Developers becoming more conscious of the human errors they commonly make and actively checking for these can help reduce errors.
Use cognitive training. We have shown that using cognitive training, like the OODA loop, seems to help decision making and can reduce the human errors a developer makes.
Simplify your workload. One of the biggest causes of human error reported by the developers in our study was the complexity of the development environment. Reducing the cognitive load by simplifying the complexity of the development environment could reduce human errors. Actions such as minimizing the number of simultaneous development tasks and closing down unnecessary tools and windows can help reduce the cognitive load.
Communicate carefully with stakeholders outside your team. Our study suggests that a relatively high number of mistakes are related to communicating with stakeholders outside of the development team. Ensuring that communication is clearly understood seems important to reducing mistakes.

Consistency in code reviews (article review)

tse2020_hirao.pdf (uwaterloo.ca)

In the last year, I’ve written a lot about code reviews, mostly because this is where I put my effort now and where I see that software engineers could improve.

Although there is a lot of studies about how good code reviews are and what kind of benefits they bring, there is no doubt that code reviews are a tiresome task. You read software code and try to improve it, but, let’s be honest, if it works don’t break it – right?

In this paper, the authors study open source communities and check how often the reviewers actually agree upon the code review score. They find that it’s not that often – 37% disagree. From the paper: “How often do patches receive divergent scores? Results: Divergent review scores are not rare. Indeed, 15%–37% of the studied patch revisions that receive review scores of opposing polarity“

They also study how the divergence actually influences the patches – are they integrated or not: “Patches are integrated more often than they are abandoned. For example, patches that elicit positive and negative scores of equal strength are eventually integrated on average 71% of the time. The order in which review scores appear correlates with the integration rate, which tends to increase if negative scores precede positive ones. “

Finally, they study when the discussions/disagreements happen and how many reviewers there actually are: “Patches that are eventually integrated involve one or two more reviewers than patches without divergent scores on average. Moreover, positive scores appear before negative scores in 70% of patches with divergent scores. Reviewers may feel pressured to critique such patches before integration (e.g., due to lazy consensus).2 Finally, divergence tends to arise early, with 75% of them occurring by the third (QT) or fourth (OPENSTACK) revision. “

I think that these results say something about our community – that we tend to disagree, but do integrate the code anyways. What does that mean?

It could mean two things, which IMHO are equally valid:

The review comments do not really touch upon crucial aspects and therefore are deemed not so important (e.g. whether we call something weatherType or typeOfWeather as a variable…)
The reviewers’ reputation makes it difficult to get some of the comments through, e.g. when a junior reviewer is calling for a complete overhaul of the architecture.

Either way – I think that the modern code review field is quite active these days and I hope that we can get something done about the speed and quality of these long and tiresome code review processes.