The idea of artificial superintelligence (ASI) has long tantalized and taunted the human imagination, but only in recent years have we begun to analyze in depth the technical, strategic, and ethical problems of creating as well as managing advanced artificial intelligence.
Artificially intelligent agents are already replacing human jobs at the factories (for example, it is estimated that about 15% of American manufacturing jobs were lost to other countries, the remaining 85% was due to automation). It is replacing doctors in the diagnosis of illness. It is replacing taxi drivers. It is composing music.
For instance, there is a hotel and a restaurant in Japan that is staffed almost entirely by robots. And, this is just the beginning. Nick Bostrom in his book approximates and states that it seems entirely feasible that we will have a more than human AI – a super intelligent AI – by the end of the century. However, scientists might be one of the few groups to actively suppress that desire to make predictions. Conservative and data-driven by nature, they might be uncomfortable making guesses about the future because that requires a leap of faith. Even if there is a lot of data to support a prediction, there are also infinite variables that can change the ultimate outcome in the interim. Trying to predict what the world will be like in a century does not do much to improve it today; if scientists are going to be wrong, they’d rather do it constructively.
Indeed, the world has changed a lot in the past 100 years. In 1918, much of the world was embroiled in the first World War. 1918 was also the year the influenza pandemic began to rage, ultimately claiming somewhere between 20-40 million lives — more than the war during which it took place. Congress established time zones, including Daylight Saving Time, and the first stamp for U.S.
airmail was issued. Looking back, it is clear that we have made remarkable strides. These days, scientists do their best to achieve new results, in particular – teaching machines to think like humans.
Recently, a new type of neuronal network has been built, which can dramatically improve the efficiency of teaching robots to think like we. The network, called a reservoir computing system, is made with memristors and could predict words before they are said during conversation, and help predict future outcomes based on the present. In addition, it can process a photo and correctly identify a human face, because it has learned the features of human faces from other photos in its training set. Keeping this progress