I’m not a huge fan of the “Daisy AI Podcast” but he often rattles off a lot of interesting information quickly, like in his 2013 podcast.

A definition he quoted (but I can’t find the source) really struck home for me:

Branch of Computer Sciences that solves problems for which there are no known efficient solutions but which we know are efficiently solvable because [something] with intelligence can solve them.

Doing the impossible is always fun!

I like this definition because it bypasses a lot of controversy about various tests for AI and debates regarding the nature of consciousness or sentience. And it makes the AI singularity a fuzzy boundary rather than a breakthrough occurring at a moment in time when we throw the switch on a particularly advanced computer system.

If AI is just another Computer Science technique for solving problems, then measuring the ability of a computer system to mimic the behavior of something intelligent (usually a human) is not the best metric. A better metric might be the number and breadth of these problems a system can solve. And that seems to me to be more like what progress in Machine Intelligence looks like these days, an exponential growth in the types of problems computers can solve. AI is just an advanced form of Machine Learning. And contributing to this field only required developing algorithms to solve hard problems, which basically all developers do every day.

Another way I might state this definition, though I may be reaching a bit, is:

AI/MI/CI solves NP-hard problems approximately and quickly, like humans and other complex systems do.

This allows for the intelligence of plants in their solving of the respiration planning and regulation problem of distributed stomata opening and closing in complex patterns to optimize the flow of water and CO2 into their “bodies” and the exhale of water vapor and O2 out to maximize photosynthesis efficiency.

In the same episode, “Daisy” listed all these applications of AI:

• study of artificial life
  • if you can simulate something (behavior) you understand it
• imitate intelligent human behavior
• speech recognition
• computer vision
  • image recognition
  • facial recognition
  • image classification
  • self driving cars
• study the human mind
• operations research
• control theory
  • drones
• optimization
• robotics
  • robot control systems
    • Boston Dynamics (dog, cheetah, mule, atlas humanoid)
    • iRobot
  • behavioral robotics
  • evolutionary robotics
• scheduling
• planning and motion control
  • self driving car
  • geographic directions and navigation instructions
  • motion planning and execution/tracking for Amazon warehouse robots
  • municipal government transit departments
    • traffic light optimization
    • road, exchange, bridge, train, bus route planning and optimization
• artificial life
• automated reasoning
• concept mining
  • social media
• data mining
• spam filtering
• intelligent agents or expert systems
• intelligent control
• legal litigation
  • search documents
• NLP
  • chatter bots
  • translation
  • requires “understanding” of texts in both languages
    • may be AGI already
  • machine learning
  • dialog systems
    • airline, rental car, hotel reservation
    • Uber?
• pattern recognition
  • OCR
  • handwriting recognition
  • postal address recognition
  • speech recognition
    • Siri
    • Google Now
  • license plate recognition
  • infrared image recognition of walking people (drone targets)
• indexing
• artificial creativity
  • art
    • create “paintings”
    • compose music
  • write stories
    • sports reporting
    • financial reporting
• game agents and players
  • interesting opponents
  • can download game AIs open source
  • often beat human
    • Deep Blue (chess)
    • Watson (jeopardy)
• medicine
  • diagnostics
  • drug design
• game theory
  • corporate business logistics planning (Amazon)
  • linear programming
• AGI - general intelligence
  • understanding natural language
• knowledge representation
  • logic
  • data/fact storage
• machine learning
• cybernetics
  • brain simulation
  • brain signal interpretation
  • prosthetics
  • study of C. Elegans
• symbolic manipulation
  • SOAR (state operator and result) or scruffy AI (not hand-designed math or logic)
  • nonscruffy AI (machine learning)
• web search
  • interpret queries optimally
  • neural nets
• factory automation
• banking system credit card fraud
• telephone dialog systems
• problem solving expert systems
• industrial robot control
• embedded in medical testing machines
• military
  • killing machines
  • image interpretation
  • self-driving vehicles (DARPA Grand Challenge)
    • CMU van drove from
• mathematical proofs
  • solutions oftern appear more creative/elegant than human-derived
• spacecraft independent planning/control
  • NASA’s Eclipse system
• finance
  • high frequency trading
  • portfolio optimization
  • comodity trading
  • banking
  • hedge funds reading twitter
    • Bingo