What is AI and What can we do with it today?

With this first post in the series on AI, I thought I’d start with what AI is and what we can do with it today. There are many schools of thought on how to define AI and they broadly fall into one of four buckets:

Definition of AI

AI is the science of designing agents that

Let’s expand on each bucket a little bit.

  1. Thinking humanly: According to this school of thought, we should be building agents that follow the thought procedures of humans. This is a field of science in itself (cognitive science) but it is not AI anymore.
  2. Acting humanly: According to this school of thought, we should be building agents that act like humans. The Turing Test, proposed by Alan Turing, was designed to provide a satisfactory operational definition of intelligence. However, it turns out that it is also not a very useful definition because the focus turns on capturing the traits we didn’t value in humans in the first place (e.g., not spelling too correctly and not solving math problems too accurately).
  3. Thinking rationally: According to this school of thought (with roots from all the way back to Aristotle), we should be building agents that think rationally and have correct reasoning processes. There are some problems to this approach: It’s not easy to state informal probabilistic knowledge in the formal logical notation and any non-trivial problem we care about in AI is computationally intractable without any guidance.
  4. Acting rationally: The most accepted definition of AI today is the science of building agents that act rationally. A rational agent is one that achieves the best possible outcome or, when there is uncertainty the best expected outcome. Being rational means maximizing your expected utility. Rationality only concerns what decisions are made and not the thought processes behind them.

Artificial Intelligence started out as a field in 1940s, with McCulloch-Pitts boolean circuit model of brain, and had a period of great enthusiasm and expectations in 1950-70, with Samuel’s checkers program and Robinson’s algorithm for logical reasoning. Everyone was predicting that the future of thinking machines was finally here but it never came. This period was termed as AI Winter (1970-90), where funding and interest in AI waned. Then, in the 90s, AI had a resurgence (AI Spring) with increased focus on probability, uncertainty and statistics.

With this short history of AI out of the way, let’s talk about what we can do with it today:

  1. Natural Language
    • Speech technologies (e.g. Cortana, Siri, Skype)
      • Automatic speech recognition
      • Text-to-Speech Synthesis
      • Dialog systems
    • Language processing technologies
      • Question Answering (Watson on Jeopardy)
      • Machine translation (Google translator, Bing translator, Skype real-time language translator)
      • Web search
      • Spam filtering
  2. Vision
    • Object and face recognition (Kinect)
    • Image classification (Imagenet challenge)
  3. Robotics
    • Google self driving cars
    • Table tennis playing robot (TOPIO)
  4. Logic
    • Theorem provers
    • Fault diagnosis
    • Satisfiability solvers
  5. Game playing
    • Checkers is solved
    • Computer are much better at chess than humans
    • Huge advances in Go
  6. Decision making
    • Scheduling
    • Route planning
    • Medical diagnosis
    • Product recommendations
    • Fraud detection


2 thoughts on “What is AI and What can we do with it today?

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s