The main message that we can extract from Hello World. How to remain human in the era of the algorithms is that the algorithms are not the holy grail, nor should we demonize them. Hannah fry He has tried to present these sets of computer instructions in the most equitable way: they turn out to be much more competent than human brains in innumerable tasks, but not in all.
The corollary would be the following: there are aspects in which we already need algorithms, and others in which human-algorithm collaboration will be much more fruitful than simply algorithmic or human. Just for that, reading this book is already worth it. However, there are many more reasons, such as taking a look not so much at what comes our way, but what is already happening in the field of justice, crime, medicine or art and we probably still did not know. That they, They, the Machines, are better than us in things we thought were unique to humans.
Something, not magic
Fry insists on putting the algorithm's skills in context against human abilities. Thus, for example, he admits that humans are excellent at interpreting subtleties, analyzing contexts, applying experience and differentiating patterns. On the contrary, humans are not good at paying attention, being precise, consistent and fully aware of our surroundings. In those weaknesses is where the algorithms can complement us. And, in fact, they do it in increasingly effective ways.
Thus, Fry puts amazing examples, such as that of an algorithm that determined in which county a serial killer had to live just by analyzing the geographic patterns of his victims, something that no human policeman had managed to deduce. However, we cannot blindly trust the algorithms. Fry sets an example for the diagnosis of cancer by examining breast x-ray patterns:
The problem is that refining an algorithm often means having to choose between sensitivity and specificity. If we focus on improving one of the two aspects, that will often mean losing out on the other. If, for example, we decided to give priority to completely eliminating false negatives, the algorithm could mark all the breasts that it considered suspicious. That would mean 100% sensitivity, which would certainly satisfy our goal; but it would also imply that a large number of perfectly healthy women would be subjected to unnecessary treatment. Let's say that, on the contrary, we decided to prioritize the absolute elimination of false positives. The algorithm would consider everyone healthy, thus obtaining 100% specificity. Fantastic!… As long as you are not one of the women with a tumor that the algorithm has overlooked.
Judges are often wrong, they even disagree with other judges (and even disagree with themselves because the human being is not consistent). Do we have to place our trust in algorithms that calculate, for example, the probability of recidivism of a defendant to determine a penalty? The answer is not black or white. Algorithms are also victims of bias. They are also wrong. Betting on "warmth" or human "intuition" is not the solution either. We must intervene in the weaknesses of human judges, not replace them completely. And the same is extrapolated to doctors. Or to autonomous driving.
Fry's book is accessible, fresh, full of studies and amazing examples. Like that a group of people believed that a musical composition belonged to a classical composer more times if the composition was actually conceived by a machine than by the classical composer. It also clarifies confusing or difficult concepts to define without a certain background in computer science or mathematics in a very illustrative way, explaining unequivocally what an algorithm is, or artificial neural networks.
Therefore, in addition to being a book that must be a must read to be minimally informed about what is coming and what is happening already, it has inspired us for articles in Xataka Science how:
- In this simple way, Google could incline you to vote for certain political training.
- If pigeons are as good as humans diagnosing cancer, we need algorithms.
- A dog or a wheel? Only one pixel decides everything for this algorithm.