THING 19 – ARTIFICIAL INTELLIGENCE
What is AI?
Artificial intelligence (“AI” for short) is a very broad term to describe any man-made system that can make decisions based on inputs. In its most broad definition, even a pinball machine is an AI because it “decides” when you get points based on the position of metal balls. In the context of emerging technologies however, AI describes systems that can make very complex decisions based on many inputs. AI programs have already been built to do many tasks, everything from flight traffic control to playing chess. In fact, the most publicised of these is a system called “alpha-go”, designed to play the board game Go. It defeated the world champion handily, and the only players to beat it since have been newer, better AI systems.
What is ML?
Machine Learning (“ML” for short) has been under development for quite a while. Much like AI, the definition can be a bit fuzzy. Most commonly, Machine learning techniques, are mathematical models run on computers (usually very powerful computers) that output very complex algorithms to solve problems. This is a bit unintuitive: Most computer programs are made by people, however this is a new class of techniques that use programs to make programs. Some of the more well known ML techniques include evolutionary algorithms, artificial neural networks (think artificial brains), and reinforcement learning. The result of this has been that problems once considered too complicated even for computers to solve—because a human still had to write the programs for the computers to use in solving them—can now be solved by the computers themselves.
Where is this going/What’s all the fuss about?
As with any technology, these new techniques can be used for good or ill. On the good side, ML / AI systems have proven to be much better than human doctors at diagnosing illness. Doctors are expected to memorize dozens of textbooks worth of complex information, which is really only reasonable if that doctor is a computer. On the darker side, ML systems can reinforce algorithmic bias (see Thing 6: Diversity), or facilitate highly realistic fake videos (for example, the series of ‘Nicholas Cage Deepfake’ video clips). In the latter case, ML algorithms have been made to edit video with much more efficiency and skill than a human for the purpose of editing a person’s face into footage they were not originally in. There are myriad possible negative uses for this technology.
New uses for this technology are being thought of every day, and it is doubtful that any aspect of life will not be affected even 5 years from now.