Evolution of AI

Artificial Intelligence

Bombe machine

• The concept of ‘Artificial Intelligence’ first emerged in 1950 when an English Mathematician called Alan Turing (1912-1954) grappled with the question: "Can machines think?".
• A team led by Alan Turing developed an AI machine called ‘Bombe’ to crack ‘Enigma code’ used by Germans in World War II that were used to send encrypted messages.
• As an honour, the Turing Test was named after Alan Turing, that determined whether or not a computer is capable of thinking like a human.
• The term "Artificial intelligence" was then first coined by John McCarthy during the first conference on the subject in 1955.
• John McCarthy is known as the ‘Father of AI’.

Alan Turing

John McCarthy

Unimate - 1961

• Unimate, the 'grand father of industrial robots', was the first industrial robot.
• It was founded by George Devol.
• It was used to automate metal working and welding processes in General Motors assembly line in NJ, USA.

Photos from Wikimedia

ELIZA - 1966

• A computer program that demonstrates superficiality of communication between a human and a computer therapist using pattern matching techniques.
• Even though ELIZA has no intelligence, it inspired the development of many AI chat bots.
• ELIZA was developed by the MIT AI Lab, led by Joseph Weizenbaum.

Photo from Wikimedia

Try ELIZA - http://psych.fullerton.edu/mbirnbaum/psych101/Eliza.htm

Deep Blue - 1997

IBM Chess computer 'Deep Blue'

World chess champion 'Garry Kasparov'

• Deep Blue is a chess playing computer developed by IBM.
• It defeated world chess champion Garry Kasporav in 1997 in a game that lasted several days.
• Deep Blue uses AI technology to play chess.

Kismet - 1998

• Kismet was the first emotionally intelligent robot.
• It is capable of detecting and responding to human feelings.
• Kismet was developed by Dr. Cynthia Breazeal at MIT

Photo from Wikimedia

IBM Watson - 2011

• IBM Watson is a question-answering computer system that uses Natural Language Processing.
• It won 'Jeopardy', an American-based television game show, beating former champions Brad Rutter and Ken Jennings.

Siri - 2011

• Apple Inc. introduced Siri in 2011.
• Siri was the first AI-based virtual assistant on a smart phone (iphone 4s model).
• Siri uses natural language processing technology.

Roomba - 2012

• Roomba was the first mass-produced autonomous robotic vacuum cleaner.
• It was developed by a company called iRobot.
• It uses range of sensors to recognise objects such as chairs, walls and tables. It determines the cleaning path using AI algorithms to navigate and clean homes on its own.

Photo from Wikimedia

Eugene Goostman - 2014

• Eugene Goostman was the first AI chatbot to pass the Turing score threshold of 30% by convincing 33% of event judges that it is a human.
• Even though event's organiser (Warwick) considers Eugene Goostman has passed the Turing's test, the validity of claim is questioned by critics.
• Eugene Goostman was developed by three computer programmers - Vladimir Veselov,

Eugene Demchenko and Sergey Ulasen.

• It uses Natural Language Processing technology.

Self-driving cars - 2015

• Google's World’s first fully driverless car, nicknamed 'Firefly' was on public roads in 2015.
• This ride was not accompanied by a test driver or police escort and the car had no steering wheel or floor pedals.
• Self-driving cars use a combinations of sensors (radar, GPS, sonar) to perceive the environment.
• They use AI technologies such as computer vision and deep learning.

Computers outperform humans in identifying images

Another milestone in 2015 was at the Annual ImageNet Challenge. At this event, algorithms compete to show their proficiency in recognising and describing a library of 1,000 images. Researchers achieved 97.3% accuracy in 2015 resulting in them claiming computers could identify objects in visual data more accurately than humans.

AlphaGo - 2017

• Google’s AI AlphaGo system beat world champion, Ke Jie, in the complex board game of Go.
• Go is considered significantly difficult for computers to win because it can include more than 30 million moves on the board. The AlphaGo is trained using examples of these moves.
• The AlphaGo system uses complex algorithms including machine learning and deep learning techniques.
• The image to the left demonstrates how quickly the Alpha-Go AI system is learning to play Go over a number of days.

See the interactive demo: https://deepmind.com/blog/alphago-zero-learning-scratch/#gif-120

Now & into the future...

• Can you find a more recent example of AI advancement?

• What do you imagine the future evolution of AI to become?