Matter and Memory and Deep Learning
Download as PPTX, PDF1 like603 views
The document discusses the intersections of artificial intelligence, memory, and human learning through the lens of philosophical and sociological perspectives. It explores the historical development of AI, particularly deep learning, and contrasts algorithmic capabilities with human consciousness and memory, drawing on the works of figures like Bergson and Libet. The conclusion suggests that while machines may exhibit forms of memory and learning, they significantly lack the depth and experiential quality of human cognition.
![Human Learning
Libet, Bergson, and the brain
✤ Benjamin Libet Mind Time - experiments directly stimulating areas of the
brain during neurosurgery
✤ Awareness takes place 0.5secs after stimulation/physical sensation
✤ Micro-differences in time taken from feet and hands to brain also erased
✤ Awareness is a composite of different moments all somewhere around
0.5secs after the event : “awareness itself is a mental phenomenon
separate from the content of a mental event” (Libet 2005:56).
✤ Often interpreted to support supervenience of physics over conscious will
✤ On the contrary - Libet is clear there is a “conscious veto” - Bergson is
clear that the ‘now’ we experience is all too often in the past and that free
will is precisely that veto - the ability to intervene in automaticity and
choose
✤ Plus - deliberative consciousness occurs over much longer time periods
than the perception-action flow of physical stimuli measured by Libet
✤ “Practically we
perceive only the past,
the pure present being
the invisible progress
of the past gnawing
into the future.
Consciousness, then,
illumines, at each
moment of time, [an]
immediate part of the
past” (Bergson
1920:194).
“Memory… imports the
past into the present,
contracts into a single
intuition many
moments of duration”
(Bergson 1920:80)
https://thehumanevolutionblog.com/2015/01/07/is-violence-what-made-humans-smarter-than-other-animals/](https://image.slidesharecdn.com/bergsonjapan-161111145121/85/Matter-and-Memory-and-Deep-Learning-8-320.jpg)
![Artificial Intelligence
Turing and intelligent machines
✤ Alan Turing (1950) “Computing Machinery and Intelligence”
Mind LIX (236): 433–460 : ‘Can machines think?’
✤ Imitation game: A(male) B(female) / C must determine which
is which; A must try to deceive C.
✤ Dismisses ‘mind’ and ‘think’ as undefinable, replaces
question with ‘Are there imaginable digital computers which
would do well [as A] in the imitation game?’
✤ Sleight of hand creates ‘human computers’ that digital
computers may mimic
✤ But - as he says - digital computers are ‘discrete state
machines’ defined by programmable Laplacian rules: initial
states determine outcomes
• Hayes and Ford (1995) - Turing
Test is harmful
• Dennett (1985) and Floridi et.
al. (2008) - it is simply too hard.
• Block (1981) - intensely
programmed ‘Blockhead’
capable of a myriad special
responses but still neither
intelligent nor possessing ‘mind’
• Searle (1981) questions whole
notion that machines could
think
http://www.rutherfordjournal.org/article040101.html](https://image.slidesharecdn.com/bergsonjapan-161111145121/85/Matter-and-Memory-and-Deep-Learning-9-320.jpg)
Matter and Memory and Deep Learning
- 1. Diagnoses of Matter & Memory: Brain, Time and Memory Tokyo & Osaka, Japan November 2016 Matter and Memory and Deep Learning David Kreps
- 2. Agenda ✤ Preface ✤ Introduction ✤ Deep Learning ✤ Human Learning : Language; the brain ✤ Artificial Intelligence ✤ The future
- 3. Preface ✤ Who am I? - ‘a sociologist of technology and a philosopher of science’ - a ‘new materialist’? ✤ Theatre -> Arts Management -> Cultural Theory -> Sociology -> Information Systems -> Business School -> communications? philosophy? ✤ Book: Bergson and Systems Theory : ✤ Philosophy need neither be the ‘handmaiden of science’ nor divorced from instrumental reality ✤ Strong emergence through self-organisation but with ‘teleological’ trends: active consciousness both end-point and inherent in both matter and evolution
- 4. Introduction A little background ✤ Artificial Intelligence - 1950s - General A.I. - C3PO / Terminator ✤ Narrow A.I. - 1960s - real practical baby steps… ✤ Machine Learning - 1970s onwards, including ‘neural nets’ ✤ Deep Learning - 2010s ✤ 1940s/50s computer engineering (mis)use of language ✤ Algorithm: a flow diagram that represents the bare essentials of what an engineer can understand and reproduce of a human activity ✤ not the human activity itself
- 5. Deep Learning ✤ Artificial Neural Networks ✤ 1970s machine learning ✤ Not like biological neurons - c1011 flexible interconnections ✤ Algorithms failing multiple times in order to isolate ‘correct’ answer ✤ 2000s: GPUs & social media ✤ Andrew Ng - Google ✤ Layering of artificial neural networks - ‘deep’ ✤ Photo-‘recognition’ - cats in YouTube videos / self-driving cars? http://www.nytimes.com/2012/06/26/technology/in-a-big-network-of-computers-evidence-of-machine-learning.html
- 6. Human Learning Language ✤ “the unfortunate anthropomorphic language used about computers” (Checkland 1988:239) ✤ Information theory : signal processing theory V X I C M IV + VIII ©MMXIV ✤ Memory : storage ✤ Learning : pattern-matching https://sites.google.com/a/rbbschools.net/mrs-stone-s-garden/home/essential-words-for-third-grade
- 7. Human Learning Bergson, Memory and Learning ✤ Two kinds of memory, two kinds of learning ✤ Human memory is so much more than storage: it is a conscious experience of time ✤ Algorithms can only match what is in front of them with what is stored within them ✤ The logfile of an algorithm exists continuously, stored as tokens in the ‘now’ of the computer storage ✤ Memory recalls each repetition of learning a lesson as an individual moment in personal history ✤ Algorithmic ‘memory’ is Bergsonian ‘habit’ - an action that is a lesson learnt ✤ Humans are aware in time: computers are not ✤ “I study a lesson… and in order to learn it by heart I read it a first- time, accentuating every line … it is said that I know my lesson by heart, that it is imprinted on my memory” : “Each several reading then recurs to me with its own individuality… each reading stands out before my mind as a definite event in my history” (Bergson 1908:89).
- 8. Human Learning Libet, Bergson, and the brain ✤ Benjamin Libet Mind Time - experiments directly stimulating areas of the brain during neurosurgery ✤ Awareness takes place 0.5secs after stimulation/physical sensation ✤ Micro-differences in time taken from feet and hands to brain also erased ✤ Awareness is a composite of different moments all somewhere around 0.5secs after the event : “awareness itself is a mental phenomenon separate from the content of a mental event” (Libet 2005:56). ✤ Often interpreted to support supervenience of physics over conscious will ✤ On the contrary - Libet is clear there is a “conscious veto” - Bergson is clear that the ‘now’ we experience is all too often in the past and that free will is precisely that veto - the ability to intervene in automaticity and choose ✤ Plus - deliberative consciousness occurs over much longer time periods than the perception-action flow of physical stimuli measured by Libet ✤ “Practically we perceive only the past, the pure present being the invisible progress of the past gnawing into the future. Consciousness, then, illumines, at each moment of time, [an] immediate part of the past” (Bergson 1920:194). “Memory… imports the past into the present, contracts into a single intuition many moments of duration” (Bergson 1920:80) https://thehumanevolutionblog.com/2015/01/07/is-violence-what-made-humans-smarter-than-other-animals/
- 9. Artificial Intelligence Turing and intelligent machines ✤ Alan Turing (1950) “Computing Machinery and Intelligence” Mind LIX (236): 433–460 : ‘Can machines think?’ ✤ Imitation game: A(male) B(female) / C must determine which is which; A must try to deceive C. ✤ Dismisses ‘mind’ and ‘think’ as undefinable, replaces question with ‘Are there imaginable digital computers which would do well [as A] in the imitation game?’ ✤ Sleight of hand creates ‘human computers’ that digital computers may mimic ✤ But - as he says - digital computers are ‘discrete state machines’ defined by programmable Laplacian rules: initial states determine outcomes • Hayes and Ford (1995) - Turing Test is harmful • Dennett (1985) and Floridi et. al. (2008) - it is simply too hard. • Block (1981) - intensely programmed ‘Blockhead’ capable of a myriad special responses but still neither intelligent nor possessing ‘mind’ • Searle (1981) questions whole notion that machines could think http://www.rutherfordjournal.org/article040101.html
- 10. Artificial Intelligence Complexity, novelty and consciousness ✤ Contemporary complexity theory discounts Laplacian model : 50/50 bifurcations/self-organisation/quantum model - means probabilities is all there are; strong emergence: no ‘discrete state machines’ except computers (Prigogine and Stengers 1985) ✤ Origination, creativity, innovation - Turing describes as mere ‘surprise’ : the Laplacian view ✤ For Bergson novelty is key : humans as ‘centres of action’ able to make choices - the ‘conscious veto’ in duration ✤ Turing’s only answer to this seems to be that “there is nothing new under the sun” and that originality does not exist (Turing 1950:450) • Addressing ‘objections’ Turing is dismissive of ‘consciousness’ as solipsism • “I think that most of those who support the argument from consciousness could be persuaded to abandon it rather than be forced into the solipsist position.” (Turing 1950:447). • Strong emergence: conscious beings arising from self- organisation who can make changes to the conditions from which they arose http://www.rutherfordjournal.org/article040101.html
- 11. ✤ Dissipative structures : e.g. tap running into a bath; water forms a spiral at open plughole. Could be oil or wine - the dissipative structure would be the same (Goodwin 1994:9-10) ✤ Structure independent of contents: exists only as they pass through it ✤ Could quantum computing incorporate the complexity of bifurcation ✤ Complex ‘open state’ computing generate dissipative structures? ✤ Could one/combination of dissipative structures constitute/generate ‘awareness?’ ✤ Awareness, or consciousness required if, beyond data, storage, and pattern matching, we are to consider information, memory, recognition or learning. ✤ Conclusion: We cannot rule out the possibility of machines having memory, information, learning, or intelligence. But we can be confident that those digital computer engineering tools we already do have fall far short of deserving the use of such words. The future Artificial intelligence beyond discrete digital state machines? • ‘Deep learning’ fad may already be over : eCommerce giants rush to hire curators • Amazon bought Goodreads, a website based around personal book reviews • Spotify have expanded their range of human playlist makers. • Netflix actively train their users to tag their movies • Facebook in trouble because of perceived bias amongst their human news curators, but falling back upon algorithms for their news feed has by contrast brought numerous examples of ‘fake news’ to the ‘Trending’ feature – in trouble now as perhaps contributing to Trump’s success • Samsung’s news app divides into what you want to know and need to know; the former chosen by algorithm, the latter by editors
- 12. Contact ✤ Dr David Kreps ✤ http://david.kreps.org ✤ www.creativeemergence.info ✤ d.g.kreps@salford.ac.uk