A brief history of AI and machine learning
AI is transforming the way we live and interact at an accelerating rate. Robotic companions help us as we age, the cities we live in are becoming smarter, and machines are enabling us to better manage chronic diseases.
Our four-part video series “The End of the Beginning” highlights some of the technology that is becoming part of our everyday lives. Part one, What’s Next for AI, looks at the evolution of the science behind artificial intelligence that builds on a history of slow advances that began in the middle of the 20th century.
The origins of artificial intelligence
Following developments of machine-based computation during World War II, some early computer scientists decided to take on a challenge much greater than calculating ballistics tables or tabulating census results. They had a vision for creating machine intelligence.
Computer science as a field may well have started during in 1956 at a summer workshop at Dartmouth College in the United States, where the participants undertook a study “to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.”
The focus of the first decades of AI was on intelligent tasks, such as playing chess and translating languages. Many of these challenges turned out to be more difficult than anticipated. For example, it wasn’t until 1997 that IBM’s Deep Blue chess-playing program could beat Gary Kasparov, the top-ranked chess player in the world. But in another area, translation, human translators still outperform machine translation programs to this day.
AI researchers have made slow but steady advances over the decades. During the 1970s and 1980s computer scientists collaborated with peers from emerging fields, such as cognitive science and knowledge engineering, to create expert systems. The goal of these applications was to capture and algorithmically apply expert knowledge. An early AI system, MYCIN, diagnosed infectious diseases and recommended treatments.
Understanding problem-solving as a general skill was also widely studied. For example, programs were developed to solve toy problems, like the Monkey and Banana problem, in which a monkey is in a room with bananas hanging from the ceiling and out of reach. The room also contains a chair and a stick. Scientists needed to program a machine so that it could create a plan of action, such as moving the chair closer to the bananas and using the stick to reach them. This is a trivial example but exemplifies the focus on formal reasoning and logic in early AI.
Some AI researchers tried to build intelligence based on formal logic, while others modeled intelligence on biological intelligence by creating neural networks that mirror the inner workings of the human brain. These researchers started from the assumption that the real world is not well-structured, and many problems cannot be solved using logic.
For example, logical predicates are not helpful for distinguishing a benign from a malignant tumor in a medical image. Neural networks could solve some problems that logic-based systems could not, but they were difficult to train. Despite advances in neural networks, then-called connectionist networks, there was no effective way to train larger networks that were needed for more complex tasks.
A failure to deliver
While progress was being made, there was still a fundamental problem impeding the development of AI at scale. In the video, Prof. Thomas Malone, founding director of the MIT Center for Collective Intelligence, says that AI practitioners and their marketing colleagues set unrealistic expectations that could not be met — at least not until the next significant advance in AI algorithms.
“There was a lot of excitement and enthusiasm about the possibilities for that in those days,” he recalls. “People actually greatly overestimated how quickly progress would be made.”
The major breakthrough came in 2006 when researchers developed algorithms that cloud train neural networks with many layers. These were called deep learning networks. This approach to AI and machine learning had a significant advantage over earlier approaches.
First, there was no need to identify rules or other forms of structured knowledge. This avoided the time-consuming and often incomplete task of defining a comprehensive body of knowledge. With deep learning, machines could be programmed to learn from data without significant human effort to process or structure that data. This type of AI scaled beyond anything that could have been done when humans tried to hand code rules and procedures of intelligence.
Two factors were crucial to the success of deep learning — machine learning algorithms capable of learning from exposure to large amounts of data and computing devices with resources to meet the demanding load.
What’s next for machine learning
Yosi Taguri, co-founder MissingLink.ai, which is a part of the Samsung NEXT product team, points out machine learning and analytics have crept into our daily lives and brought us services that otherwise wouldn’t exist without AI. For example, using only a phone and an Uber or Lyft app, someone can have a personal driver at their location in a few minutes. While these apps may seem simple, they complex coordination and decision-making is needed in order to assign a task to a driver and ensure he or she is in the right place at the right time.
“Given those conditions, it’s insane,” says Taguri. “Humans cannot be that effective at scale.”
As highlighted in the End of the Beginning videos, AI has come a long way from trying to help monkeys reach bananas or play chess. It is now part of our everyday lives and has become woven into our lives faster than history would have predicted. In the future, AI and machine learning promise to bring data-driven decision making, “smart” devices that can interact in real-time, increased productivity and efficiency, and the democratization of specialized medical knowledge and advice.
As Dor Skuler, co-founder and CEO of Intuition Robotics, points out, “we’re going to see, not just robots doing things for us, like vacuuming the floor, but we’re going to have relationships with machines, someone or something that’s very loyal, always has your back, and sometimes makes you laugh.”
It’s been a long and winding road. But now, looking ahead, it is difficult to find an area of work and life that will not be affected by the latest generation of AI.
Learn more about the cutting edge of AI entering our daily lives by watching the End of the Beginning.