Deep stupidity: what neural networks can and cannot do .. Prof J. - PowerPoint PPT Presentation

Deep stupidity: what neural networks can and cannot do .. Prof J. Mark Bishop (PhD) Director TCIDA (Tungsten Centre for Intelligent Data Analytics) & Professor of Cognitive Computing, Goldsmiths, University of London. What force directs these actions? Where lies the deep body-magic that brings forth my world? Where sits this mind? If science can fully respond to these questions then it might be possible to one day simulate the human ability to act mindfully, with intelligence, with computation ..

Computation and Intelligence For much of the twentieth century the dominant cognitive paradigm seated the mind in the brain - if computers can model the brain then, theory goes, it ought to be possible to program them to act like minds - and in the latter part of the twentieth century this “astonishing hypothesis * ” fuelled an explosion of interest in ANNs (Artificial Neural Networks): both high-fidelity simulations of the brain (cf. computational neuroscience; theoretical neurobiology) and looser – merely ‘neural inspired’ - analogues (cf. computational connectionism; engineering ‘intelligent’ systems). * “ You, your joys and your sorrows, your memories and your ambitions, your sense of personal identity and free will, are in fact no more than the behaviour of a vast assembly of nerve cells ”, Francis Crick, The Astonishing hypothesis: the scientific search for the soul, (1994).

Deep magic ..

Researchers at Sony have developed FlowMachines , a system that learns music styles from a huge database of songs. Exploiting unique combinations of style transfer, optimization and interaction techniques, FlowMachines composes novel songs in many styles. " Daddy's Car " is composed in the style of The Beatles. French composer Benoît Carré arranged and produced the songs, and wrote the lyrics. Sporting events are a major source for companies looking for marketing opportunities. Global sporting brands, reach audiences on every continent. This is worth millions to the brands and they are in constant pressure to prove the value and exposure of sponsorship to their clients. Multiple logos fight for space therefore one needs to ensure that you can track instances of the brands being photographed and published. IDS ’s proprietary image matching technology is perfectly suited to this challenge.

Deep Associations: “ A maps to B ”

The technological singularity ..

The impenetrable barriers between Deep Minds and Real Minds • Well known examples illustrating this barrier include: 1. Computers don’t understand the meaning of the bits they so adroitly manipulate (Searle: Chinese Room Argument; Szegedy: Classification Fallacy); 2. Computers lack mathematical insight (Penrose: Computability); 3. Computers cannot experience phenomenal sensation (Bishop: Dancing with Pixies).

Computers lack understanding - Searle; Szegredy From Szegredy et al., (2013), Intriguing properties of neural networks . Proc. Inc. Conf. International Conference on Learning Representations. Searle, Minds, Brains Programs , Behavioral and Brain Sciences 3(3): 417-457. “Syntax is not sufficient for semantics”, computers do not understand the bits they manipulate.

Computers lack insight - Penrose Penrose, (1989), The Emperor’s New Mind , OUP.

Computers lack feeling - Bishop Nancy is a complex 19th century stage automaton moved by a hand crank hidden under the stage In 1950 Alan Turing described a ‘wheel machine’ … … a device that sequentially clicks through one of three possible machine positions – machine states – at 120 degree intervals; like a clock with a Implementing Turing’s wheel-machine ‘computation’ hand that can only rest at the 12 (A) , 8 (B) and 4 (C) o’clock positions with a car’s digital milometer

Conclusion • The view that intelligence is rooted solely in ratiocination and computation is a relatively modern one, and one that has not gone unchallenged .. .. taken together, I suggest the arguments I have outlined today undermine the notion that the human mind can be completely instantiated by mere computations; if correct, although computers will undoubtedly get better and better at many particular tasks - say playing chess, driving a car, predicting the weather etc. - there will always remain broader aspects of human mentality that future AI systems will not match. Under this conception there is a ` humanity-gap ' between the human mind and mere `digital computations'; although raw computer power - and concomitant AI software - will continue to improve, the combination of a human mind working alongside a future AI will continue to be more powerful than that future AI system operating on its own; contra Professor Stephen Hawking, to paraphrase Gil Scott-Heron, ‘ the singularity will not be computerised ..’

Bishop, J.M. (2009), A Cognitive Computation fallacy? Cognition, computations and panpsychism , Cognitive Computation: 1(3), pp. 221-233. Bishop, J.M. (2009), Why robots can’t feel pain , Mind and Machines: 19(4), pp. 507-516. Bishop J.M, (2016), Trouble with Computation: refuting digital ontology in Cooper B.S. & Soskova, M.I. (eds), The incomputable: journeys beyond the Turing barrier, (Springer).