AI : between Science and Fiction - Introduction

Myths mentioning human-like creatures built from inanimate material date back to the antiquity with Pygmalion, a sculptor who fell in love with a statue he had carved which turned into a woman, or with the clay golems of Jewish mythology. Such a fascination for creation or replication of life, in particular human life, is universal.  Multiple other examples can be found in Chinese, Nordic or Arabic cultures. Today, these myths have evolved in tight correlation with technical and scientific progress and creatures are often restricted to their intelligence, which seems to be the most challenging issue since the brain has not yet yielded up all its secrets. Artificial Intelligence has thus been one of the main themes of Science Fiction in the past century and the latest scenarios, mainly dystopian, raise awareness regarding the recent surge of scientific progress in that field. People like Elon Musk or Stephen Hawking shared their concern about AI leading to destruction and no more than the end of humanity. Nevertheless, distinction must be made between Science and Fiction to apprehend this technical revolution and envision how AI can help create a better world.

Artificial Intelligence is the field of computer science created in the 1950’s studying agents able to perceive their environment and perform actions to maximize the chance of success to reach a goal. Paradoxically enough, the first successes of AI were in tasks difficult for humans such as playing checkers or chess while functions as natural as vision or language comprehension long remained hard to mimic. Indeed, the first intelligent programs relied on hard-coded rules which are suited for small, perfectly defined environments such as games where the computing power of a machine is no match for a human brain. Hence the iconic victory of IBM’s Deep Blue over chess world champion Garry Kasparov in 1997. 20 years later, AlphaGo beat Go world champion in a game enabling more possible configurations than particles in the universe and where brute force approach is ineffective. This has only been possible with Machine Learning, a field where rules are not written by a human being anymore but learned through examples and training. More particularly, Deep Learning is at the root of the recent renewed interest in AI and while theorized in the early 1960’s, it only soared in 2012 when the combination of labeled data quantity and computational power enabled to prove its efficiency. The exact same algorithms are used in computer vision, speech recognition and language understanding, powering technologies such as autonomous cars, virtual assistant or automatic translation. This history of AI led to an evolution of the very definition of artificial intelligence since games such as checkers or chess have been proven solvable only with predefined routines and computational power which is not what is commonly accepted as intelligence. The leading current definition of Artificial Intelligence is the ability to perform high-level tasks such as perception or planning and above all to learn.

On the other hand, fictions have mentioned a vision of AI way before these scientific breakthroughs, mainly using humanoïd robots often viewed as mechanical copies of humans, gifted with feelings and consciousness. This description is very distant from the current advances in the technology which is only qualified as weak AI. Indeed, even though programs outperform human beings at recognizing every breed of dog from a picture or at Go, they are specifically designed and trained for a very narrow task and cannot even compare with the abilities of a mouse to perceive its environment, learn and perform the multitude of tasks necessary to survive. Such an ability in a machine is designated as strong AI and is still currently fictional. The research field most related to achieving strong AI is Artificial General Intelligence (AGI) and is for now limited to training one system for multiple tasks. Projecting human behaviours and feelings to robots is thus currently not relevant and there is no proof it will ever be, but it raises awareness. As soon as 1942, Asimov perceived the potential threat embodied by these fictional creatures and defined the three famous laws of robotics that should be hardwired into their brain :

1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
2. A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws.

Nevertheless, these laws have always been challenged by Science-Fiction authors who have imagined hundreds of worst-case scenario stories and questioned risks of AI for safety and ethics. Among those stories, the ones visually told in movies or series come first to mind and both were inspired by and inspire scientific progress. From 2001 : A space odyssey to the latest Westworld series, my goal is to study how SF visual creations forged a dystopian vision of AI and distinguish fiction from reality on some topics they depict such as threat to humanity, ethics issues or feelings from or for a machine.