mersenneforum.org - Official "Science News" Thread

[size=1]Wow, I just realized how long this post is now that it's out of the "Quick Reply" box...please pardon my verbosity, I definitely don't mean to scare you off with a wall of text! :smile:[/size]
[QUOTE=Dubslow;286232]I wouldn't be surprised if what they call 'sensors' in Star Trek is just an umbrella term for a wide variety of techniques and instruments that everyone is familiar with, so that in universe, saying 'instrument A using method B detected C' would be common knowledge, so that if someone said instead 'sensors detected C' everyone would still know that it was instrument A and method B without saying so. Does that make sense? Is there any evidence to the contrary that I am unaware of?[/QUOTE]
Indeed, that does seem to be a pretty accurate description of how the "sensors" seemed to operate. I guess what I was thinking, though, is that present-day detection methods seem to give rather more "shot in the dark" answers than those of science fiction--by observing side effects from which we can make inferences as to the causes, rather than observing the underlying fundamentals directly. For instance, when we want to scan for water below the surface of Mars, we have to analyze all sorts of geological samples in order to "guess" a location where the water might be; whereas the [I]Enterprise[/I] sends out a sensor beam of some sort that, when reflected back to the ship, can be analyzed to determine "hey, this thing made contact with water x meters below the surface".

Probably the biggest hurdle to getting to that sort of ability is not so much in designing a radical new method of "sensing" (we do, after all, use basically the same concept as the basis for radar, sonar, and such), but in finding ways to increase the accuracy of the methods we already have. Once you're able to determine what information is "bounced back" from the sensor beam with mcuh greater granularity, it's a lot easier to, say, tell the difference between rock and water on the other end of the beam, rather than just seeing that there's "something" on the other end of the beam (as is often the limit of present-day technologies). Then it's just a matter of empirically determining what figures correspond to what substances (and so forth), and programming that into the computer controlling the sensors.

On this tack, another thing that would help a lot is having a computer that can perform basic logical analysis of the information avaialble to it. Our computers today are very strictly deterministic; they can [I]only[/I] handle situations that the programmer anticipated and accounted for in very fine detail. A good example of this is the "Siri" automated assistant feature built into the latest iPhone: it analyzes queries based on keywords and format, and can do that very well because it's programmed for a huge range of possible scenarios, but it will still fall flat if presented with a truly unanticipated situation. This isn't a problem for, say, asking directions to the nearest sushi restaurant (since that's a relatively common scenario), but it wouldn't do very well at determining the composition of a completely unknown chemical compound. The computer needs to be able to "think" in a manner somewhat like we do in order to do such a task.

In Star Trek, the computers work on a completely different principle than those in real life: instead of following granular concrete steps programmed in by the operator in a (relatively) low-level language, the operator tells the computer what he wants it to do--often in spoken English, though it is often done more efficiently by using an interface panel (the keyboard-meets-iPad touch interfaces all over the ship). Basically, instead of telling the computer "read an integer x off sensor data line 0, add it to variable y, then search array z[] for a possible match with that integer" the operator just says "Computer, analyze data from planetary sensors and catalogue all compounds detected." The fine details of implementation are come up with by the computer "on the fly", and can thus accommodate a large range of unexpected scenarios. It's like the difference between doing plug-and-chug arithmetic, versus being able to take a 3D surface and apply integration methods to determine its volume. Both are well-defined methods for which the "hard work" of creativity has been already done by humans, but the latter is a dynamic problem which requires some actual reasoning to determine limits of integration and so forth before the "plug and chug" can begin.

Interestingly enough, we do already have a basic idea of how to construct such a computer in real life; in fact, an early prototype of such a thing already exists. You may recall IBM's "Watson" supercomputer that played an exhibition game of Jeopardy against two human champions--it was based on a simulated neural net which could "learn" in a manner rather similar to a human. Not having actual sentience, it wouldn't be able to do things like proving a new mathematical theorem on its own, but it didn't need to: all it needed was to interpret questions in often-trickily-worded English to determine what kind of scenario was being analyzed, then "connect the dots" to information in its database that it had "learned" earlier in its programming process. (It "learned" in much the same way a human would: by "reading" in the digitized text of books and internet sources on a wide variety of subjects.) When it went to the contest, it did poorly at first, since it had a hard time with plays on words and other idiomatic forms of speech; but as it "observed" the other contestants' responses to questions, it "learned" empirically how to interpret them itself, and by the end of the contest it was answering every question flawlessly. Now that the contest is over, the computer is slated to be put to work assisting doctors with medical diagnoses--they'll submit queries and pertinent information (patient records, test results, etc.) to the computer from a "typical" computer over a network, and it will respond with its diagnosis and--get this--the [I]reasons[/i] to back it up, so the human doctor can verify it based on his own expertise.

At any rate, there's still a lot of work to be done in this field, but it is particularly neat to see that it is indeed possible to build a computer in real life that works, really, exactly the same way the computer in Star Trek were portrayed to work. (The writers of Star Trek definitely did an excellent job of "guessing" how such a computer would theoretically work even before it actually existed--the in-show references to how the computers worked described a neural-net-based system basically spot-on to the real thing.) There were a few articles on the Watson supercomputer linked earlier in this thread (and perhaps also the "Uninteresting Stuff" thread in the Lounge), which

Anyway, I could go on about this forever...as you can probably tell, I'm a rather serious fan of Star Trek and find these developments in computer technology in a remarkably similar direction to be quite fascinating. :wink: Personally, I think once we can perfect the analytical computer and bring it to more general availability, we'll find that a lot of "far-off" technological advances suddenly become a lot more realistic when you have such a computer driving them--the tricorder being one of them. :smile: The tricorders in Star Trek actually seem to have full-fledged analytical computers built into them (definitely not as powerful as the ship's computer, but pretty amazing nonetheless)--but even though such a level of miniaturization is likely quite far off in real life, the idea of building a sensing device that relies on a remote mainframe computer for analysis would seem quite feasible.