Upgrade to Pro — share decks privately, control downloads, hide ads and more …

AI in 2014: Progress and Problems

89289602684a545b286cf4937f13f8fc?s=47 Beau Cronin
October 17, 2014

AI in 2014: Progress and Problems

From Strata + Hadoop World NYC 2014

89289602684a545b286cf4937f13f8fc?s=128

Beau Cronin

October 17, 2014
Tweet

Transcript

  1. AI in 2014 Beau Cronin @beaucronin

  2. Covering AI is hard https://flic.kr/p/oC9NCe Hype Technical Evaluation Complexity Definitions

    Crackpots
  3. AI is a moving target • What is it [not]?

    • Who’s doing it, where, and what progress is being made? • Where is it going in the next few years?
  4. What is AI [not]?

  5. ENVIRONMENT METHODS PROJECTS FICTION 2001 Dartmouth Summer ARPA Started LISP

    Vision Project ELIZA Shakey SHRDLU Lighthill Report Winter #1 Winter #2 Expert Systems Deep NNs Connectionism PDP Turing Test “Golden Years” Symbolic “GOFAI” General Problem Solver A.I. The Matrix Blade Runner The Terminator I, Robot Alien Star Trek: TNG Deep Blue Watson Siri Colossus: Forbin Proj Her Wargames Jetsons DARPA Grand Challenges Google Car XCON 5th-Gen Computer Strategic Computing Initiative ALPAC Report Stats & Machine Learning MIT AI Lab Prolog BKG Connection Machine Society of Mind Cog DART Kismet Roomba Mars Rovers 1950 ‘60 ‘70 ‘80 ‘90 2000 ‘10 ‘20 Cloud Computing Excession Diamond Age Daemon
  6. ELIZA Perceptron SHRDLU Shakey

  7. Interlocutor Superoptimizer Android AI Archetypes & Definitions Big Data Learner

  8. “AI has failed.”* “AI is all around us. It’s in

    Excel, FFS”** * Paraphrased ** Heavily paraphrased
  9. None
  10. What progress is being made? Context Where? How? How much?

  11. When we find a new [AI] paradigm or a new

    set of techniques, it feels like we are driving a car on a highway and nothing can stop us until we reach the destination. But the reality is that we are really driving in a thick fog and we don't realize that our highway is really a parking lot with a brick wall at the far end. (…perceptrons, rule-based systems, neural nets, graphical models, SVMs) Yann LeCun, Facebook AI Research https://flic.kr/p/364e61
  12. Summer Vision Project: Off by 100x …Moravec’s Paradox

  13. AI Progress Is Relocating Increasingly: • Applied • Commercial •

    Integrated • Scaled • Data-Centric • Small/Large Universities Research Labs Independent Cos. Tech Giants Research Labs? Startups (small)
  14. Impact requires methods and integration Integration New Methods DL Rsch

    DL image search Watson Siri IMPACT Prob. Prog Sibyl • New methods is a moving target • Integration includes scale, robustness, testing, deployment, …
  15. “Self-driving cars are not an anomaly; they’re part of a

    broad, fascinating pattern. Progress on some of the oldest and toughest challenges associated with computers, robots, and other digital gear was gradual for a long time. Then in the past few years it became sudden… “Our digital machines have escaped their narrow confines and started to demonstrate broad abilities in pattern recognition, complex communication, and other domains that used to be exclusively human" The Bulls
  16. http://imgs.xkcd.com/comics/tasks.png https://github.com/karpathy/convnetjs

  17. The enterprise of achieving [human-level intelligence] artificially — the field

    of ‘artificial general intelligence’ or AGI — has made no progress whatever during the entire six decades of its existence http://www.kurzweilai.net/why-artificial-general-intelligence-has-failed-and-how-to-fix-it The Bears
  18. or

  19. Where is it going? • Richer models • Is winter

    coming? • Privacy, intrusiveness, control • The AIs (almost) no one expects
  20. Richer Models • The path from narrow to general •

    Possible approaches: • DL++ • Probabilistic programs • Deeper connections with the brain sciences "Planetarium in Putnam Gallery 2, 2009-11-24" by Sage Ross - Own work. Licensed under Creative Commons Attribution-Share Alike 3.0-2.5-2.0-1.0
  21. None
  22. The Unreasonable Effectiveness of The Unreasonable Effective of Data More/Better

    Data Richer Models
  23. • AI is here to stay this time - there’s

    too much money to be made • But localized retrenchments are still possible, maybe even inevitable NOT
  24. Privacy, intrusiveness, control • Social, cultural, political issues just as

    important as technical ones • This interplay is a huge source of uncertainty and complexity in forecasting
  25. ?

  26. Climate system Traffic Logistics Agriculture Factories Supply Chain Climate Power

    Plants
  27. Thank you Beau Cronin @beaucronin