CS 105P

Modern Computer Trends

Modern Computer Trends

Three computer trends likely to have the most impact over the next few years are the Internet, speech recognition and portable computers. Half (1.3 billion dollars, 1997) of Microsoft's R-and-D money is being spent working on speech recognition.

Internet

• Faster Internet connections will enable richer media experiences.
• Positive feedback loop will drive pervasive electronic commerce
• Inexpensive (relative to the cost a few years ago) PC's will increase reach
• Current investment activity will help generate new products and ideas

Speech Recognition

• Discrete speech recognition systems - user must separate each word with a short pause. Makes it easy to find the beginning and ending of words. Forces user to use consistent pronunciation.
• Continuous speech - User speaks naturally. Much harder.
• Speaker-independent - No prior training required. A group of people can use the same system for speech recognition.
• Speaker-dependent - User trains the system by speaking vocabulary words.
• Vocabulary Size - It's much easier to recognize a limited vocabulary.
• Example consumer products: DragonDictate, IBM VoiceType.
• Ideal system:
  1. Allows continuous speech
  2. speaker-independent
  3. Large vocabulary

Portable Computers

• Classes of systems (Lap Top, Hand Held PC, Palm Top, wearable computers)
• Embedded systems (Special purpose computers. For example, cars have many embedded computers. One to adjust air fuel mixture. One to provide cruise control.)
• Types of input (character recognition, small keyboard, voice input)
• Weakness fond in most PDA's: don't connect well to central computers.
• Web phones.

Artificial Intelligence

Long term, the technologies likely to have the most impact are those related to Artificial Intelligence. The problems are harder, but the rewards are greater.

By now you should be convinced that computers are great number crunchers. You may also have the impression they can be stubbornly consistent and precise. This chapter is about using computers for something more than number crunching. It's about using computers for tasks that on the surface seem to require human characteristics such as intelligence and intuition.

Several years ago there were high expectations and hopes for Expert Systems and Artificial Intelligence. Today these technologies are still used but they have never really brought the changes that were forecasted.

Some key terms and concepts you should understand:

• What is Artificial Intelligence (AI)? Research area. Study of how to mechanize human intelligence.
• When does a problem fall into the AI category? (When we can't explain how we as humans do it. If we can explain how we do it a conventional algorithm is used to solve the problem.)
• http://www.mcs.net/~jorn/html/ai/cyc.html- cyc project = started in the early 1980's. Contains about 400,000 logical assertions.
• Difficulty of automating something we don't understand.
• Knowledge base = set of facts and rules about facts
• Inference engine - manipulates existing facts to come up with new facts.

Example: Given the rule "If x is y's wife then y is x's husband" and the fact "Amy is Ken's wife" you can conclude that "Ken is Amy's husband".

• Neural Networks - Computers programmed to mimic the human brain.
• Turing Test
• Why is it hard for a computer to exhibit intelligence? We don't really understand what "causes" intelligence. Or put another way, we don't understand the process of intelligent action. Think back to the last time you had a good idea. How did it occur? Usually you were thinking of related things and then it just "clicked".
• Given a list of list of tasks which ones are easy/hard/impossible for a computer to do. Why are some problems impossible?
• Data mining. - Turning data into information. Looking for patterns in information.
• Expert system. Limited domain. Facts and rules. Emulates a human expert. Not very adaptable. Works poorly at the perimeter of its knowledge.
• Robots = Very successful in limited domains. Still no general purpose robot.
• Virtual Reality. Not very developed. Requires very fast computers. Many future applications..