It seems to me…
“I’m not dumb. I just have a command of thoroughly useless information.” ~ Bill Watterson, Calvin and Hobbes.
Quite a few years ago, with a freshly-minted bachelor’s degree (Math/Physics), I accepted a position at Boeing in Seattle as a computer programmer in a management control unit. At that time (mid-1960s), Boeing was able to only estimate the cost for about every one hundred commercial aircraft. Cost estimation required inverting a matrix containing cost data but computers at that time had neither the speed nor storage capacity to perform the necessary calculations which required repeatedly partitioning the matrix and merging results.
The unit manager asked if I could handle the calculations to enable them to improve their cost estimates. I very egotistically told him it would not be any problem – Just let me know what he wanted – If he wanted the computer to talk, I would do that. Boeing was in the process of updating to a new generation of computing equipment (from IBM 1400s, 7080/7090s to IBM System 360s). We were able to improve estimate results to about every ten planes; better estimates would have required over 24 hours of continuous computation – something which even the new generation of computers were incapable.
Fortunately, he never took me up on my offer to make a computer speak. I returned to graduate school the following year and then began to realize just how difficult a task computer generated speech actually was given the limitations computers had at that time. The basic speech algorithms had not yet been developed and capabilities available today did not exist.
I’ve previously stated my belief that human speech eventually will be the preferred human/computer interface (https://lewbornmann.wordpress.com/2011/11/03/speech-interface/) and while limited speech recognition currently is available, speech understanding remains quite a few years in the future. It is relatively easy to parse text or voice input into basic tokens and based on those tokens, create a response but this does not constitute speech understanding.
Everyone has experienced the frustration associated with attempting to find an answer to some question on the Internet. It might be necessary to repeatedly re-write a search-engine query to obtain the desired answer. Speech is natural to humans and we fail to appreciate the actual difficulty in understanding the meaning of what is intended to be said. The classic example is the sentence “Time flies swiftly like an arrow” which supposedly can be interpreted in well over a dozen different ways. A computer hasn’t any way to determine what is meant; it doesn’t know there isn’t any insect called a “Time Fly”. Speech understanding is dependent upon world knowledge. Meaning can not be determined from a single sentence – it is dependent upon the overall context in which it is said. Intelligence is required to fully understand meaning and intent.
A somewhat related problem is machine translation – translating one language into another. Many years ago, a now infamous Congressionally funded project to translate between English and Russian was cancelled when visiting Congressmen requested the sentence “The spirit is willing but the flesh is weak” be translated into Russian and back to English. The result of the double translation was “The wine is good but the meat is rotten”. Another example of word ambiguity would be the sentences “The pen is in the box” and “The box is in the pen”. The same words used in a different context results in a completely different meaning.
I’ve already expressed my opinion regarding the probability of Artificial Intelligence (https://lewbornmann.wordpress.com/2011/05/12/intelligences-future/) – there isn’t any question of “if”, only how soon it can be achieved.
That’s what I think, what about you?