Artificial intelligence. Guide for thinking beings

In the midst of the exploration and exploitation of AI, several voices have warned of the dangers it entails if its use is not regulated at a legislative and governmental level. Because, as often happens, technological advancement and progress are faster than the analysis and awareness we have about its possible future uses. It has happened with mobile phones and game addiction, but it has also happened with something that is much more specific but equally harmful and that has even led its creator to ask for forgiveness: infinite scrolling. Thus, tools that a priori were thought, designed and implemented to make people’s lives easier have become dangers and threats, creating addictions from which it is not easy to evade or escape.

AI is not far from falling into that same infinite black hole that absorbs everything and adds those who fall into it into an eternal spiral in which the time dedicated to stopping and reflecting seems to have no place. But AI carries other dangers beyond possible addiction and this book published in 2020 and fully valid alerts us to this while reassuring us because, although the dangers exist and must be taken into account, science is not so advanced so that right now they can be a problem or, at least, not at the moment, and perhaps precisely for this reason it is time to stop and analyze what may come, before the snowball of technological advance drags us all towards a destination chosen by a few.

Melanie Mitchel, expert in Complexity, cognitive architecture and Artificial Intelligence, alerts us at the outset by stating that «Google has stopped being an internet search portal. It is rapidly becoming an applied AI company (…) Parent company Alphabet’s ultimate aspiration is reflected in its Deep Mind group’s original mission statement: “to solve intelligence and use it to solve everything.” the rest””. With this adventurous, ambitious and disturbing statement, and given the rise of AI and its increasingly important impact on our lives, the author explains that “this book was born from the attempt to understand the true situation of artificial intelligence: what they can do computers today and what we can expect from them in the coming decades.

As in every elaborate essay, the author traces a time arc and places us in the beginnings of artificial intelligence in 1956 in a seminar at Dartmouth College organized by a young mathematician named John McCarthy and from there until the mid-2000s she breaks down its Fast forward to the most recent qualitative leap produced in AI with the creation of systems such as Google Translate, self-driving cars, assistants such as Siri, Alexa, automatic subtitles on YouTube and simultaneous translation in calls via Skype as well as facial identification. upload photos on Facebook. But, despite the great advances in AI, the predictions made about what would be achieved have always been excessively optimistic, reality has always curbed expectations because it seems that emulating human intelligence is not as easy as it seemed because, despite Although Zuckerberg predicted that by 2025 AI would reach the level of humanity (“general AI”), the truth is that current models continue to be an example of “narrow AI” since they can only carry out a strictly defined task ( The best program for playing Go can only do that, it can’t even play checkers or tic-tac-toe). Therefore, «although general AI was the objective of this field of research, materializing it has proven much more difficult than expected. AI work has ended up focusing on specific and very defined tasks.

Analyzing the weaknesses of the system, the author exposes one of the main challenges of AI: visual narrative. When we see an image, our brain is capable of creating a story around it: what could have happened before the photograph was taken, what after, what emotions did its characters feel, what is the atmosphere it conveys, etc. This is something very difficult to emulate computationally as well as object recognition (something easy for humans, but not so easy for computers). Thus, the author describes the enormous difficulties of AI in interpreting objects, something that, on the contrary, a small child can do through other examples. This is a field in which artificial intelligence encounters more difficulties beyond the fact that, in any case, networks cannot learn on their own, but rather a team of programming engineers is necessary to train them with a training process that requires millions of images obtained thanks to all the publications of users on networks such as Facebook, etc. but also the identification of images in the captcha verification that Google uses to tag images from its image bank. In any case, “it is unrealistic to think that we can label everything in the world and meticulously explain every last detail to the computer.”

Likewise, it is also far from achieving one of the attractions of technological advances in driving and the long-awaited self-driving cars, since “it is impossible to train or code a system for all possible situations” because even though humans make mistakes We have “a great fundamental competence that all current AI systems lack: common sense. We have extensive background knowledge of the world, both physically and socially, not to mention the problems that must be resolved in ethical aspects that the author exemplifies in the tram dilemma and whether buyers of autonomous vehicles would buy them. taking into account the moral principles under which their driving and choice in decision-making are governed because, after all, who would want to buy a car that, for the sake of the “common good”, chooses to crash oneself in a fatal accident before running over a group of people? Statistics show that an immense minority. Therefore, beyond the advances in some specific aspects that AI has achieved quite successfully (automatic translators, voice transcriptions and subtitles in real time…) there are others in which it is not yet so advanced because “the “The truly difficult goal is to create machines capable of truly understanding the situations they face.”

The conclusion of all this is that the great distance that separates us from any AI is “transfer learning”, something that is common for humans but totally incipient in AI. We are capable of learning things based on knowledge acquired in similar tasks, but AI is so specific that it must always start from scratch for each new task. Thus, what for a person who knows how to play tennis (for example) implies a certain advantage in learning paddle (for example) in a machine they are two totally different processes. And not only in terms of executing mechanical tasks but something much more complex: understanding the situation. Thus, “while the most advanced AI systems have nearly equaled (and in some cases surpassed) humans in some very specific tasks, none have the understanding of the rich meanings that humans bring to perception, language and the reasoning”.

For all these reasons, this book is a very interesting read as it deals with a topic that affects us all and presents us not only with the advances of AI but also with its pros and cons, its weaknesses and its strengths. In this in-depth analysis of AI, the author shows us her extensive knowledge on the subject, although sometimes she goes too far into technical details and this means that someone without knowledge of programming or technology may have some difficulty in following the development of some parts of the book as well as the presentation of the author’s ideas due to the density of her analysis. The author herself recognizes this by stating in some section “we are going to get a little technical, so prepare yourselves (or skip this part).”

The author states in the final pages that the main concern is “that algorithms and data are used in dangerous and unethical ways.” That is the main challenge, and it is not a technological challenge, but a social one. And I fear that this makes it even more complex (and I would dare say more dangerous), since we always already know which side of the scale progress leans towards and what its results are when there are economic interests at stake. Let’s not let the artificial overcome intelligence… or ethics.