Artificial (lack of) intelligence

We have discussed several times in the past already how master craftspeople might have a hard time to explain their trade, let alone teach it. The "curse of expertise" arises from the fact that they have learnt by building their knowledge over many years and numerous attempts, putting all this information together, discarding some, abstracting other, until they distill a set of heuristics that actually make them the expert they have become. That is why a carpenter can choose or reject a piece of wood within a second, or a sculptor remembers an old piece of marble back in the storage room that would be perfect for their next work. The most rigorous among them might have taken careful notes and recorded in detail their decision processes over the years so that they can review it later or transmit it to the next generation, but the majority of craftsman would be happy with just incorporating the new knowledge as it comes without any registry of how it came about. But this is not a strictly human problem.

One frequent criticism in the current application of artificial intelligence (AI) systems in everyday life is precisely their lack of explanatory capabilities: even the designers of the AI cannot explain how the system has reached a certain conclusion. Taking into account that the AI only does what it is programmed to do it might seem paradoxical that we are not able to reconstruct its reasoning, but if we look at the way they are trained, the perspective might change, because their education is, in many ways similar to the one of a master artisan: they get millions upon millions of examples to examine and, like the human experts, their algorithms tell them to adjust, little by little, the relevance of every piece of information in each example, until some of them become irrelevant while others turn into true compasses for their task. However, this lack of explanation poses a significant problem for the use of AI systems in real life, because it make corrections very hard to introduce, and the errors can have an impact in the lives of actual people.

Photo: MaxPixel

Some weeks ago Karen and I watched the 2020 American documentary "Coded bias" on Netflix, where the authors explore the racial bias that has been built into several algorithms which are currently in use. Among the examples cited in the documentary, it is remarkable the case of a HR algorithm in Amazon that started to reject all female applicants: oblivious to the complex realities of life, family and motherhood, the algorithm figured out that women earned less than men with similar education and concluded that they must be performing worse and therefore it preferred men. However, this explanation did not come straight out of the algorithm; instead it took weeks of meticulous back-tracking to figure out why women were being rejected. Unsurprisingly, the same effect can be seen in terms of racial bias for AIs trying to establish the risk for criminal behavior: since African Americans are a significant majority in the recluse population, they were deemed twice as likely to re-offend that white criminals with similar records. As in the previous case, the AI was trained on biased data and, not having (or not being able) to justify its decision, it just perpetuated the injustice.

I address this matter today because suddenly Facebook's algorithm has decided that my blog does not meet the community standards and has now banned me from linking to it from my Facebook page. If only the algorithm were just as expeditious at banishing certain websites that are an endless source of blatant misinformation, we would all live happier (or at least more relaxed) lives. Apparently, there is a method for contesting the decision of the algorithm, but I have made use of it and the website has already warned me that, due to the pandemic, they are understaffed so it might take some time to have my complaint reviewed. I have no clue of how many new sites they add daily to their black lists, but I think they should certainly adjust the sensitivity of the algorithm to their own ability to process users' claims.

One way or another, I will continue to write my daily articles and hope that a solution eventually comes around. I will not deny that I was somewhat excited to see that my occasional link on Facebook actually sparked the curiosity of some of my friends and drove traffic into the blog, but it is still evident that readership was never the main driver for my writing. Hopefully, Blogger will continue to run for many years and will give me, at some point in the future, the opportunity to come back and read again my barrages. Have a nice evening.