This is a piece I wrote in June, 2015 for “Stony Brook Frontiers Magazine“, a graduate-student run STEM review magazine for middle school students. I had the idea to do an article on AI, as it is a technology that will become pervasive in the next few decades. AI/ machine learning is a good career path for young people, especially as many jobs are susceptible to replacement by AI.
Robots and Mechanical Men
The science fiction author Isaac Asimov published a famous book in 1950 called I, Robot. This book presents a vision of how robots could be used for the benefit of all mankind. Asimov envisioned robots not just as servants to man but also as companions, private detectives, outer space explorers, and politicians. Today, 65 years later, we have many robots that help us manufacture things like cars, computers, and electronic gadgets. We also have robots that mimic the human form and are remarkably dextrous. For instance, today’s most advanced robotic arms can move in approximately the same ways as a real human arm, such as the “Luke” robotic arm developed by Deka Research and funded by the Defense Advanced Research Projects Agency. The robots we have today, however, cannot think for themselves. Robots must be programmed by humans and given specific instructions about what to do. Because they cannot think for themselves, they cannot adapt to new situations or learn things on their own. In other words, today’s robots lack intelligence. There is evidence, however, that we are close to the technological breakthroughs necessary to provide robots with real intelligence. The next few decades are likely going to be an exciting period of history. All aspects of society would be affected by the development of intelligent robots.
The field which seeks to create intelligent machines is called Artificial Intelligence, or AI. Traditionally, AI has been a branch of computer science. To create intelligent machines, we must first have some understanding of human intelligence. Thus, findings from neuroscience, psychology, and philosophy are all useful to AI researchers. There are two approaches to AI – “top down” and “bottom up”:
“Top down” researchers are not very concerned with details of how the human brain works. From their point of view, the process of thinking can be broken down into a series of abstract logical procedures. The success of this approach has lead to computers that can play chess, computer programs that can assist mathematicians in finding proofs, and programs that can read written text and make logical deductions. It has also lead to computer programs that can make predictions about the future – for instance, how the price of stocks on the stock market will change. Many programs can also learn from the past and become better at what they do, at least for a particular task. We also have software (such as Apple’s “Siri”) that can understand spoken work communications and process basic commands. While the successes of “top down” AI are impressive and numerous, the software programs created by ‘top down’ researchers usually focus on a particular task. This is in contrast to a machine with general intelligence, which would be capable of learning any task a human can.
Unlike those using the “top down” approach, researchers using the “bottom up” approach are interested in the details of how the human brain works. The human brain is made of billions of neurons (tiny cells that send electrical signals to each other). These neurons are connected in extremely complicated networks called neural networks. “Bottom up” researchers believe that intelligence is directly tied to these networks. Many “bottom up” researchers believe that the right hardware is important, and that our current computer hardware is not suited for general artificial intelligence. Today’s computer hardware is very good at performing mathematical calculations, but cannot be adapted easily to things like identifying objects in photographs.
What does the future hold?
The “top down” and “bottom up” approaches complement each other and insights from both will be important in the coming years. Right now the “bottom up” view of artificial intelligence is gaining a lot of attention. New “neuromorphic” hardware is being developed, which creates artificial neural networks on a silicon chip. This hardware could be the basis for future robots’ brains. Neuromorphic hardware is in the early stages of development but is already being applied to things like facial and object recognition. This new hardware is the first technology that is theoretically capable of reproducing the number and density of neurons in the human brain at reasonable cost. Therefore it is time to start thinking about our future with intelligent robots!
Questions to ponder –
Which approach appeals to you more – “bottom up” or “top down”?
How should intelligent robots be regulated by the government?
Should intelligent robots have the same legal rights as humans?
Would you want a robotic companion?