Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Predictions and Hard Problems With AI


Published on

General Artificial Intelligence or Super Intelligence are often predicted for the next few decades. These assumptions are criticized as part of the scientific community but they are more than ever conveyed to the general public. The current capabilities of Artificial Intelligence are both amazingly powerful in many (narrow) domains and surprisingly weak in others. While we hear crazy scenarios for the future, AI faces some theoretical limitations and a few difficult problems, including the famous "Hard Problem" which creates an endless debate among both scientists and philosophers.

Published in: Technology
  • Login to see the comments

  • Be the first to like this

Predictions and Hard Problems With AI

  1. 1. Rakuten Technology Conference, Tokyo, October 28, 2017 Laurent Ach Manager of Rakuten Institute of Technology Paris CTO PriceMinister - Rakuten
  2. 2. 2 • GIZMODO, article by George Dvorsky, Oct. 18, 2017, • VentureBeat, article by John Brandon, Oct. 2, 2017, • The Sun, article by James Beal and Andy Jehring, Aug. 1, 2017, • Mirror, article by Louise Sassoon, Aug. 1, 2017, • TECH TIMES, article by Aaron Mamiit, Jul. 30, 2017,
  3. 3. 3 Ray Kurzweil Nick Bostrom Intelligence explosion (I.J. Good, 1965) Consulted by S. Kubrick for 2001: A Space Odyssey (1968) Science fiction movies become reality… at least in predictions! 2006: The Singularity Is Near (Ray Kurzweil) 2014: SuperIntelligence (Nick Bostrom) Stanley Kubrick • Ray Kurzweil picture by Ed Schipul [CC BY-SA 2.0], via Wikimedia Commons • Nick Bostrom picture by Future of Humanity Institute [CC BY-SA 4.0], via Wikimedia Commons
  4. 4. 4 Intelligence(wait,what?) Time Human Intelligence Artificial Intelligence Artificial General Intelligence Artificial Super Intelligence
  5. 5. 5 Brain as Hardware? Mind as Software? • Digicomp picture by Pterre [CC BY-SA 3.0 or GFDL], via Wikimedia Commons • Punched card picture by Mutatis mutandis [GFDL, CC-BY-SA-3.0 or CC BY 2.5], via Wikimedia Commons
  6. 6. 6 Confirm action Are you sure you want to upload your mind to this computer? ? Warn me when I attempt to upload my mind Cancel Upload
  7. 7. 7
  8. 8. 8 Super Intelligent machines in conflict with Humans or Stupid machines, with too much decision power “[…] we fight to make the machine slightly more intelligent, but they are still so stupid. […] The thing I’m more worried about, in a foreseeable future, is not computers taking over the world. I’m more worried about misuse of AI” Yoshua Bengio, in MIT Technology Review, January 29, 2016
  9. 9. 9 Features Label (what it is, semantics) TrainingSupervised learning Unsupervised learning Features Training Clustering (need a human to add semantics) apple / pear / banana attributes (size, color, weight, …) picture (raw pixels) text description Objects Objects
  10. 10. 10 Using deep learning, computing semantic distances Similar meanings Transformation into vectors, using a very big neural network Pictures Pictures Pictures Picture or Text “low” dimension vectormillion dimensions vector
  11. 11. 11 1 + 1 = 2 Bits00000010 Computer Memory (hardware) Data 2 This is “two”, (useful to count things!) This does not mean anything for a computer Human interpretation
  12. 12. 12 1. Deep Learning and Reinforcement Learning need to train on millions of examples (see AlphaGoZero) 2. Computers don't know how the world works, have no “common sense” 3. No generalization capability: AI today is only narrow intelligence 4. Without human interpretation, there is no intelligence in Artificial Intelligence “The definition of today’s AI is a machine that can make a perfect chess move while the room is on fire.” - A sentence from the ’70s quoted by Fei-Fei Li.
  13. 13. 13 John Searle: “A program merely manipulates symbols, whereas a brain attaches meaning to them” (1990) David Chalmer, distinguishes the easy problem and the hard problem of consciousness (1994)
  14. 14. Reductive Materialism (mind explained by brain events) Eliminativism (consciousness does not exist) Panpsychism (consciousness is everywhere) Integrated Information Theory (everything is information) 14 Dualism (mind + body) centuries of fight against dualism subjective experience remains a mystery for objective sciences