Teužgimsta dirbtinis intelektas

ENThe foundations of artificial intelligence (AI) were laid as early as the second half of the twentieth century, but as a discipline impacting real-world tasks it began flourishing only recently. To date, AI systems have achieved human-level or superhuman performance on a variety of tasks that rely on large datasets of labeled examples to learn from. This technique, known as supervised learning, has enabled machines to recognize objects and faces from photographs, diagnose malicious conditions in patient data, perform machine translation and synthesize natural-sounding speech from text. Machines have also shown impressive abilities when trained to achieve goals by trial and error, using a technique known as reinforcement learning. By being encouraged to remember successful action patterns that lead to reaching a particular goal, machines are now mastering many classical computer games and winning against the world's Go champions, for example. These skills may seem purely mechanical and quite remote from true intelligence demonstrated by humans. But AI has recently begun to master even uniquely human skills. For instance, machines create compelling paintings and musical works by attempting to produce pieces indistinguishable from real art. While one might still argue that this does not equate to true curiosity or true art, AI systems are undoubtedly moving forward at a rapid pace.What do these trends imply for our society? It is worth noting that humans have long been contemplating the thought of a creation overtaking its creator. Superintelligence may determine that it makes sense to get rid of the incompetent human species standing in the way of more ambitious goals that superintelligence may hope to achieve. Shutting the human species into a Matrix-like imaginary world, for example, may not necessarily be seen as enslaving us, but rather as an act of kindness where loving machines provide us with an eternal heaven. And indeed, why not allow more competent machines to assume control of the social and political decision-making that we seem incapable of performing competently? Ideas such as these sound an awful like a new wave of eugenics, aspiring to correct man's corrupt nature. The danger here is that, due to biases inherent in any training datasets, machines are just as imperfect and biased as we are. Without explicitly working towards aligning machines with societal goals, unacceptable machine behaviors may emerge. At the same time, intelligent machines will present a unique turning point in human history: for the first time, we will no longer be the sole intelligent species in the world. Living together with such novel forms of life will require humans to find new ways to deal with our own existence. [From the publication]