main page

 

INTRODUCTION TO A.I.

 

The Next 50 Years

 

 

 

 

 

 

 

 

1
Artificial Intelligence - Main Challenges
  The challenge of independence from nature

Summarized by the statement "Nature has invented one type of intelligence - through engineering we can explore dozens/hundres/thousands/millions others".

Bottom line : So far no evidence that the level of intelligence exhibited by higher organisms has thousands of different ways of instantiation.

  The challenge of computing power

Summarized by: "Today's computing power is not sufficient for us to develop really smart AI similar to higher-level animals or humans".

The computing power of a human mind has been estimated to be somehwere on the order of 10^15 operations/instructions per second (petaops). Today's desktop machines (used by most AI researchers) can muster 32 x 10^9 operations per second (32 gigaops) on a good day.

Bottom line: We may not really be able to do our work until 2025, when a desktop computer costing $1000 is predicted to have the computing power of a single human mind.

Reference: http://www.transhumanist.com/volume1/moravec.htm

  The challenge of real-time

Summarized by: "No natural intelligence is non-realtime; it is unlikely that we will understand the principles of intelligence when studied in the form of non-realtime systems".

Classical AI did not study temporal knowledge or behavior much; behavior-based did away with deliberation and higher-level knowledge and learning.

Bottom line: The representation and effect of time on cognition remains a largely unexplored subject; all evidence supports the notion that

  The challenge of embodiment and grounding

Summarized by: "AI systems cannot understand the world properly unless they have a body in a real world".

Bottom line: If this is true then we should be studying robots more and (isolated) algorithms less. This is happening to some extent. Both physical and simulated robots can be helpful in understanding issues related to embodiment and grounding.

  The challenge of computability

Summarized by: "Human-level intelligence relies on principles that cannot be computed - that are not computable - hence AI is in principle not possible".

Bottom line : Most evidence so far suggests that intelligence indeed relies extensively on principles that are computable, although it is quite likely that some computable principles necessary for real AI remain undiscovered, and possible that some uncomputable principles necessary for real intelligence are undiscovered.

  AI requires enormous amounts of information

Summarized by: "Real animals take a long time to learn about the world, taking in enormous amounts of data via the eyes, ears etc. Why should artificial creatures be any different?".

Bottom line: Video cameras and processing power to handle their information flow has recently become very affordable, leaving only the problem of clever architectures for processing the data.

  AI is in principle not possible

Summarized by: "Intelligence relies on principles that are an integral part of the universe and simulating those principles is impossible" or "intelligence relies on emotions, and emotions are feelings and feelings are not algorithmic."

Bottom line: No conclusive evidence has been presented either against or for this claim.

  AI is unethical

Summarized by: "Artificial creatures may one day also have emotions and feelings, hence we should stop AI research" or "Autonomous machines may accidentally cause harm to humans, hence we should tread carefully" or "Ultimately AI will be used for warfare to kill humans, hence AI is unethical".

Bottom line : Successful AI systems will be so extremely useful that it is difficult to argue against improved progress towards smarter machines.

     

 

 

 

 

 

 

 

 

 

 

 

 

2
Artificial Intelligence - The Future
  AI will become increasingly important for a number of applications

In industry: Smarter robots in factories means more programmable factories, hence cheaper and more general-purpose; in photographic devices (object recognition, automatic tagging, etc.); in the coming semantic web (growing intelligence of meta-data), etc.

In medicine: Robot-assisted surgery, intelligent diagnosis, etc.

In society: controling traffic via intelligent traffic lights; in cars (autmated everything), etc.

  Computing power will continue to increase per Euro Powering ever-more-powerful applications
  Things will become more interconnected Bringing more information to intelligent devices; calling for ever-smarter devices and systems for helping us manage the information
  Robotics is reaching maturity Making it possible to create smarter and smarter mobile objects that can do tasks in dangerous and inaccessible locations such as outer space and underground.
  EU prediction Robotics industry (not counting toys) reached $21B in 2007; expected to increase exponentially. Japan Robot Association predictions are even more optimistic. Expect AI to grow even more!