FUTURE A.I.

Books of Interest
 Website: chetyarbrough.blog

Human Compatible (Artificial Intelligence and the Problem of Control)

By: Stuart Russell

Narrated By: Raphael Corkhill

Stuart Russell has written an insightful book about A.I. as it currently exists with speculation about its future. Russell in one sense agrees with Marcus’s and Davis’s assessment of today’s A.I. He explains A.I. is presently not intelligent but argues it could be in the future. The only difference between the assessments in Marcus’s and Davis’s “Rebooting AI” and “Human Compatible” is that Russell believes there is a reasonable avenue for A.I. to have real and beneficial intelligence. Marcus and Davis are considerably more skeptical than Russell about A.I. ever having the equivalent of human intelligence.

Russell argues A.I. information gathering is still too inefficient to give the world safe driverless cars but believes it will happen. There will be a point where fewer deaths on the highway will come from driverless cars than those that are under the control of their drivers. The point is that A.I. will reach a point of information accumulation that will reduce traffic deaths.

After listening to Russell’s observation, one conceives of something like a pair of glasses on the face of a person being used to gather information. That information could be automatically transferred by improvements in Wi-Fi to a computing device that would collate what a person sees to become a database for individual human thought and action. The glasses will become a window of recallable knowledge to its wearer. A.I. becomes a tool of the human mind which uses real world data to choose what a human brain comprehends from his/her experience in the world. This is not exactly what Russell envisions but the idea is born from a combination of what he argues is the potential of A.I. information accumulation. The human mind remains the seat of thought and action with the help of A.I., not the direction or control by A.I.

Russell agrees with Marcus, and Davis–that growth of A.I. does have risk. However, Russell goes beyond Marcus and Davis by suggesting the risk is manageable. Risk management is based on understanding human action is based on knowledge organized to achieve objectives. If one’s knowledge is more comprehensive, thought and action is better informed. Objectives can be more precisely and clearly formed. Of course, there remains the danger of bad actors with the advance of A.I., but that has always been the risk of one who has knowledge and power. The minds of a Mao, Hitler, Beria, Stalin, and other dictators and murderers of humankind will still be among us.

The competition and atrocities of humanity will not disappear with A.I. Sadly, A.I. will sharpen the dangers to humanity but with an equal resistance by others that are equally well informed. Humanity has managed to survive with less recallable knowledge so why would humanity be lost with more recallable knowledge? As has been noted many times in former book reviews, A.I. is, and always will be, a tool of human beings, not a controller.

Knowledge will increase the power and influence of world leaders to do both good and bad in the world. Human nature will not change but A.I. will not destroy humanity. Artificial Intelligence will insure human survival and improvement. History shows humanity has survived famine, pestilence, and war with most cultures better off than when human societies came into existence.