The AI Hype cycle
@h6165
Abhishek Yadav
ரூபீ ப்ரோக்ராமர்
Who is this ?
By Unknown - http://www.turingarchive.org/viewer/?id=521&title=4, Public Domain, https://commons.wikimedia.org/w/index.php?curid=22828488
The imitation game
By Juan Alberto Sánchez Margallo - https://commons.wikimedia.org/wiki/File:Test_de_Turing.jpg, CC BY 2.5, https://commons.wikimedia.org/w/index.php?curid=57298943
- A human, a computer, and a (human) interrogator are in three different rooms.
- The interrogator’s goal is to distinguish the human from the computer
- The interrogator can ask a series of questions and read their typewritten responses
- The computer’s goal is to convince the interrogator that it is the human
The imitation game - prediction
By 2000, machines with 100MB of storage, can fool 30% judges in 5 mins - Turing (1952)
Whats this ?
- A vacuum tube
- A neural network
- A neuron
- SNARC
Image courtesy Gregory Loan (https://cyberneticzoo.com/mazesolvers/1951-maze-solver-minsky-edmonds-american/)
SNARC
- Stochastic neural analog reinforcement calculator
- Built by Marvin Lee Minsky, Dean Edmonds
- In 1951
- Models a rat searching food in maze
- The first neural network
Marvin Minsky
- Created 7,4 Turing machine
- Invented head-mounted graphical display
- Wrote Perceptrons
- Advised 2001: A Space Odyssey
Image courtesy: Wikipedia - https://en.wikipedia.org/wiki/2001%3A_A_Space_Odyssey_(film)#/media/File:HAL9000_Case.svg
Who's this ?
Image courtesy: By Jeff Kubina from Columbia, Maryland - John McCarthy, CC BY-SA 2.0, https://commons.wikimedia.org/w/index.php?curid=65871744
John McCarthy
Image courtesy: By Jeff Kubina from Columbia, Maryland - John McCarthy, CC BY-SA 2.0, https://commons.wikimedia.org/w/index.php?curid=65871744
- Coined the term Artificial Intelligence
- Created LISP
- Invented garbage collector
- (Many more contributions)
1956, Dartmouth Conference, started AI as a field
More predictions
Within ten years a digital computer will be the world's chess champion - Simon and Navell, 1958
Within our lifetime machines may surpass us in general intelligence - Minsky, 1961
Within a generation, I am convinced, few compartments of intellect will remain outside the machine's realm – the problem of creating ‘artificial intelligence’ will be substantially solved - Minsky, 1967
It was here
Winter has come
- By mid 70s, predictions remained unfulfilled
- US and British governments withdrew funding
- Japanese govt stayed put, but withdrew funding by 1980
- 1974-82 AI Winter
Winter has come
Reasons:
- Insufficient computing power
- Scientists underestimated how human brain functions
The predictions ?
- Minsky published Perceptrons, in 1969, that detailed the limitations (of hidden layer perceptron) - deterred people from research for many years
- John McCarthy - because what you want is 1.7 Einsteins and 0.3 of the Manhattan Project, and you want the Einsteins first. I believe it’ll take five to 500 years
AI winter Other impacts
- Backpropagation remained un-noticed
Several research works remained ignored till 1982 - RNN (Recurrent Neural Networks) remained un-noticed
- People started avoiding the term Artificial Intelligence
1983-2010
- Mixed progress, but no hype
- Research progressed.
Hype cycles covered by PCs, Internet, supercomputing, Web2.0 etc
2012 onwards
- 2012: University of Toronto research paper reduces error rates by 50% [1] using Deep Convolutional Network
- 2015: Google projects using Deep Learning - 2700
- 2016: Project Oxford by Microsoft
1- https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf
2012 onwards
Hype again ?
- Big companies investing heavily
- Startups and VCs investing
- Governments investing (China, France)
- Open source getting involved
- Cheaper computing power, cheap provisioning (cloud)
- Big data - mature tools and tech, more available data
2012 onwards
Predictions
- 2029 - Long Bet Project - lengthy Turing test - Mitch Kapoor and Ray Kurzweil
- 2029 - Turing-test capable computers manufacturing - Ray Kurzweil (revised from 2020 in 2005)