Intelligence – Where Next?
Artificial Intelligence development may be hitting a wall. Where next? 3-minute analysis.
Cutting edge AI – Deep Learning – is inspired by natural intelligence.
Neurons and neural networks are key to how the brain works and have been recreated in code.
Deep Learning recognises patterns by using layers of neurons in a network.
Data is passed through the layers, and neurons fire if the data resembles what it has ‘seen’ before.
The result is a classification which can represent virtually anything.
Deep Learning can surpass humans at some pattern recognition tasks such as analysis of CT scans, translation and fraud detection.
But there are limits to what Deep Learning can achieve on its own.
For example, Deep Learning can beat a human at Chess or Go but can’t decide whether to play the game or not.
Deep Learning systems have to be retrained to perform even slight variations to what they were originally trained for.
So far, Deep Learning:
· Cannot deal with abstraction or hierarchical structures
· Cannot identify causation
· It struggles with finding out what is not specifically defined at the beginning
On its own, Deep Learning will never produce human intelligence because human intelligence is more than just pattern recognition.
So the field of AI is engaged in problem solving on multiple fronts:
· Modelling the brain’s ‘wiring’(2)
· Reverse engineering mammalian brains(3)
· Integrating Machine Learning model types into a “Master Algorithm”(4)
· Building a “common sense” knowledge base(5)
Human intelligence has evolved over 7 million years, whereas AI has been developing for no more than 70 years.
We have evolved 9 distinct types of intelligences(6).
In 1950, Alan Turing proposed a test of a machine’s ability to exhibit intelligent behaviour, indistinguishable from that of a human(7), which launched the quest for AI.
So far, nothing has passed any reasonable definition of a Turing test8.
Until software, data and hardware is sufficiently developed, natural intelligence will remain superior to artificial intelligence.
Until then, AI’s main role will continue to be to augment natural intelligence.
- https://www.wired.com/story/how-to-teach-artificial-intelligence-common-sense/ And www.bloomberg.com/opinion/articles/2018-04-04/artificial-intelligence-research-might-have-hit-a-wall
- Blue Brain Project, EPFL – École polytechnique fédérale de Lausanne
- Human Connectome Project, Mark and Mary Stevens Neuroimaging and Informatics Institute, USC
- The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World, Pedro Domingos, 2015
- CYC – the world’s longest-lived artificial intelligence project attempting to assemble a comprehensive ontology and knowledge base that spans the basic concepts and “rules of thumb” about how the world works (Wiki)
- Howard Gardner, John H. and Elisabeth A. Hobbs Professor of Cognition and Education at the Harvard Graduate School of Education
- Alan Turing (1950) Computing Machinery and Intelligence. Mind 49: 433-460 , 1 . The Imitation Game
- In the BBC reported that the test has been passed 2014, https://www.bbc.co.uk/news/technology-27762088 – but there has been much criticism of the idea that a chatbot could exhibit behaviour that is comprehensively indistinguishable from a human – https://www.theguardian.com/technology/2014/jun/09/scientists-disagree-over-whether-turing-test-has-been-passed