A Computer (or, perhaps, a computer program) that can "learn" by itself. Often implemented using NeuralNetworks
Does a NTM have to "learn" toward an objective? How do you measure what it has learned? What are good learning programs?
One of the foundations of ArtificialIntelligence
Along with ...?
What qualifies a computer as a machine that can "learn" by itself?
- Prioritization of seeking knowledge toward a specific objective
- Flexibility to utilize ancillary knowledge when objectives change
- The ability to juxtapose tidbits of knowledge to raise curiosity for exploring new avenues
Well, at the moment machines can't really "learn" by themselves. There are three general approaches to machine learning:
And each of these has various algorithms for accomplishing the task. But this is more in the nature of PatternRecognition
than what you would call a parallel to HumanLearning?
. There are also learning methods that learn concepts instead of numbers; e.g., InductiveLogicProgramming is a sort of concept-based learning method.
NTMs are more commonly called people
. Educational psychology offers a vast literature on NTM learning, especially assessment, and instructional design (e.g. setting learning objectives).
The only other mention I see of this phrase is from an article in Europhysics News talking about artificial intelligence. I suspect they are referring to MachineLearning
I read that article too... rather interesting. It prompted me to write an essay about Machine Curriculum (on my web site). -- JeffChapman
What is the difference between "teaching" a machine that can learn and "programming" that machine? Would we eventually hire "computer teachers" with a different skill set than programmers to sit down with the ThinkingMachine
and give it lessons?
I think one important difference between programming and learning is that a teacher makes the assumption that the student is an intelligent agent: the student can generalize from examples, occasionally ask questions, and can tolerate some noise in the instruction. A prime example is natural language: people can learn it, but so far no-one has programmed (or taught) a computer enough natural language to speak it fluently.
Let us here make a some quickly composed points about some combinations of the three words. Aside from the fact that the title is probably an oxymoron, it is worthwhile to examine what is implied by the usage of these three words. Taken two at a time:
- To think naturally, one must exist naturally.
- Since machines are manufactured, not born, they are unnatural.
- To think naturally, one must have internal pattern matching and processing capabilities which are unrestrained.
- Since machines are designed, not internally evolutionary, they have workings that are limited and restricted.
- To think as a machine, one must be a machine.
- Machines have no being apart from their manufactured content.
- To think as a machine, on must be capable of thinking
- Since thought in a machine must be defined and enabled from without, thinking capabilities in a machine do not exist.
- Machines receive, manipulate, transfer, store, transmit.
- Such transactions cannot be construed as thinking.
- Machines perform or manipulate bits based on programmed instructions whose existence and operations have been designed.
- Thought is adaptive, generative, free, and not subject to designed restrictions.
- There are some natural machines, which are intricate and exhibit complicated operations.
- Such machines are not operationally controlled by thought, but by the rules of natural chemistry and physics.
In a sense the universe is a machine which contains both objects which are animated and inanimated, which can be said to have life, or not to have life. The animated or moving non-alive parts of the machine are not in motion by thought, but by initialized and physically deterministic movements.
Now taking all three together NaturalThinkingMachine
-- It does not and cannot exist unless it has a life, and no such machines are presently in production, and if they were, they would no longer be machines.
- The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.
See also David Aha's Machine Learning page at: