Last Updated on
Creating true artificial intelligence is the Holy Grail of the programming world. Although many artificial intelligence algorithms can mimic human responses, or learn from experience, they have not yet been able to pass the Turing Test, which is the accepted benchmark for artificial intelligence.
AI programming code is typically approached from one of two perspectives. Either the coder is working toward a programming goal, such as passing the Turing Test, or toward a functional goal, such as improving security systems.
Copying Human Responses
Early AI programs produced randomly selected responses based on keywords in the input received. Much current AI software used for entertainment works on the same principle. Many websites offer the chance to chat with an AI program, for example, so that people can experience AI firsthand, but these chatbots are typically quite elementary. Interactions with these programs can take place via text, or for more advanced programs, via verbal conversation. Both use natural language processing to interpret the input from a human.
Passing the Turing Test
The Turing Test gets its name from Alan Turing, who devised a theoretical test to measure a computer’s responses against human behavior. During the test, judge have the opportunity to ask questions without knowing whether the answers are computer generated or are from a person. After the conversation, the judges attempt to identify the other “person” as a computer or a human. When judges are less than 50% correct, the computer is judged to be indistinguishable from a human being. No AI program has ever passed this test, despite yearly attempts for the Loebner Prize, which awards cash and prizes to programmers who come close to the goal.
The Turing Test, however, is not the only test of AI accomplishments. Although no computer is currently able to pass this test, the other feats of logic and learning that computers are successful with allow them to be called, “Artificial Intelligence”.
AI Programmers Create Programs that Learn From Experience
Artificially intelligent computers can, in some cases, use input in order to learn, in a process called machine learning. The combination of supervised and unsupervised learning allows these programs to provide increasingly appropriate responses. Artificial neural nets mimic the processes in the human brain that are accomplished by biological neural networks. Neural networks connect a number of smaller nodes, which work both together and side by side, to process information. These networks are extremely complex, and are the closest reflection of human-like processing that is currently available.
Decoding Science. One article at a time.