2024-05-11 04:00:10
By David Monniaux – Computer science researcher, National Center for Scientific Research (CNRS), University of Grenoble Alpes (UGA)
The word “algorithm”, once used exclusively in certain scientific disciplines, is now everywhere. Sometimes adorned with almost magical virtues, sometimes vilified, algorithms are mentioned in social and political debates. However, they are generally poorly known, and the word has ended up having connotations far removed from its scientific meaning.
An algorithm is a calculation method. The word “algorithm” was already used in this sense at the beginning of the 17th century.e century, or more generally in the sense of calculative art. An algorithm has a goal: for example, the algorithm of multiplication that we learn in primary school aims to… multiply two numbers. An algorithm implements mechanical operations, which do not require the intuition or intelligence of the person doing the calculation to be involved. Several algorithms can achieve the same goal, but requiring more or fewer operations and more or less memory space. Obviously, in the absence of other factors to take into account, we try to minimize this necessary calculation time and memory space.
For example, there is much faster than the algorithm learned in primary school if you want to multiply numbers with a lot of digits. This may seem surprising, as this algorithm is so familiar to us and does not seem to do unnecessary work! The great Soviet mathematician Andrei Kolmogorov conjectured in 1956 that there was no fundamentally faster multiplication algorithm, and was very surprised when, in 1960, a young student attending his seminar, Anatoly Karatsouba, offered him one. Karatsouba’s ideas opened a line of fruitful research (Toom-Cook algorithm, etc.), which notably enabled the use of effective cryptographic encryption, used daily today for access to websites, banking transactions, etc.
The science that designs and analyzes algorithms is called algorithmic ; it is a component of science computer science. A question as seemingly simple as sorting data or searching for data in a table motivated the writing of long scholarly works !
How did the term “black box” flourish?
How did we start from this scientific and precise meaning to arrive at that of a “black box” with unpredictable and poorly understood results? I see several factors.
The first is that computer science is relatively poorly understood. Certainly, we are no longer in the time when people were surprised that we might obtain doctorates in this discipline, but it is less identified than, for example, history or physics. In fact, when certain computer science researchers are presented in the media, they are described as working “in computer science and mathematics”, perhaps because the word “computer scientist” evokes other images (technician who installs equipment, hacker in a hooded sweatshirt in the dark in front of a green screen where cabalistic signs scroll…).
In 2021, a Liberation article declared: “For some researchers, there would be a mistake here: by looking too much at the effects of algorithms, we forget to study them themselves. As if they were black boxes impossible to open, almost autonomous – the proof , there are still debates to define what a “is.”
Thus, a major newspaper seems to deny the existence of a scientific field, algorithmics, which designs and analyzes rather precisely defined objects, algorithms…
The second factor is that a human does not interact with an algorithm, but with software, or even an even more complex system. When you search for your way in a GPS guidance application, it is certainly an algorithm for finding the shortest path in a graph that responds, but this algorithm is based on a cartographic model, which may be incorrect or inaccurate… The map is not the territory! As in physics, the devil hides in the way we have modeled the world using mathematical quantities, the only ones susceptible to an algorithmic process. Furthermore, the algorithm will try to optimize a certain quantity, which can be an estimate of time (assuming what driving style?), an estimate of gasoline consumption, according to parameters not necessarily within reach of the ‘user.
Likewise, the ParcourSup systemwhich is used in France to manage admissions to theeducation superior, does not only consist of the few algorithms for calling candidates, very simple and directly derived from regulatory texts, but also of online applications with perfectible ergonomics, and of ranking commissions, human and which each have different criteria. The so-feared “algorithm” is then a metonymy for this set whose decisions we sometimes do not understand.
A “black box” can be partly analyzed
Additionally, for most consumer software, their code source, that is to say the description of their internal functioning, is not public. It is therefore impossible, even for a person of art, to know what they really do – although it is often possible to suspect it: even a “black box” can be partly analyzed. It is therefore healthy to be wary!
Finally, the last factor: the massive deployment, in recent years, of applications involving automatic learning, a form of “artificial intelligence“We have developed this type of approach for problems that we did not know how to attack using classical algorithms, because this requires modeling of the world until a well-defined problem is obtained.
However, we cannot easily produce, for example, a definition mathematical how to tell a cat from a dog using the pixel values of an animal photo. We therefore proceed by “learning” correct answers from databases, which in concrete terms amounts to an algorithm adjusting parameters, sometimes billions of them, so that the system answers correctly not only on the examples that we provided it, but also on others, by a form of analogy.
The quality of such a system therefore depends on the database used for it.learning, which may have errors and coverage biases, and also the care that the designers took to identify various possible biases. Above all, such a system produces results without justifying them – explainable artificial intelligence is a subject of research. We can therefore speak here of a black box, even if we can understand certain things.
The confusion between algorithms, decision support, software, and artificial intelligence is in full swing. For example, the MonMaster site for registering for a master’s degree in French universities is presented as “without algorithm”whereas the simple fact of displaying a sorted list of proposals or being able to search for a submitted application involves algorithms!
What to do ? In a good number of public debates or broadcasts dealing with computing, only promoters or detractors of this or that technology are invited. However, States, and in particular France, pay people whose job is precisely to analyze algorithms and their limits: computer science (teacher-)researchers. Perhaps we should invite them more and listen to them?
1715487886
#Algorithms #black #boxes
