Slashdot Mirror
AI Pushing the Boundaries of Space Exploration
An anonymous reader writes "An interesting look at how artificial intelligence will help probes to undertake more complex missions in deep space, aid robot rovers in exploring other planets and improve satellites' ability to monitor activities on earth."
Just make sure the 'do not kill' weighting is greater than any possible combination of other directive weightings, rather than merely greater than the second-highest weighting.
I don't care if the A2's were a bit twitchy.
My AI prof said that the term "AI" refers to software systems which address the class of problems which are easy for biological brains but difficult for computers. For example: summing a thousand numbers is superior intelligence, but it isn't AI. Recognizing a face, on the other hand, is AI.
But every AI problem which is solved shrinks the definition of AI. Now that facial recognition software works, it isn't AI anymore. Because the definition itself changes, the term itself seems somewhat meaningless. Yesterday's AI is today's mundane consumer electronics feature. For this reason, the use of the word AI makes me feel the same way as the word "nano." It isn't really very meaningful.
A slashdotter who didn't build his own computer is like a Jedi who didn't build his own lightsaber.
Who is this Al?
fuel chemistry is pushing the bounds of space exploration. And steel engineering. And antenna design. And numerous other fields.
Which is exactly why you will never see anything more than an expert system in space. There is no way any space agency is going to punt hundreds of millions or euros/dollars/pounds into space without a full understanding of the decision tree in the spacecraft control loop. It is hard enough at the moment without introducing outliers into the system.
Please remain calm, there is no reason to pani... wait, where are you all going?
I read TFA. It's kind of funny this type of story is posted as news at all. The types of things that NASA and the ESA are describing in their interviews are more complex flight control software algorithms. It used to be that very simple feedback loops were used in combination with various controller chips (like PIDs) in order to give a spacecraft a few modes of operation. Activation and deactivation of these modes of operation were performed manually by ground controllers. However, as tech has progressed and onboard computing power has gotten cheaper, engineers have been able to design control software that activates and deactivates various modes of operation itself. In other words, it forms the same basic feedback loops that you might find on a Roomba, or some other terrestrial robot. It reads some input from a set of sensors. It uses those inputs to formulate a series of commands, be it rates and velocities, or mode-change commands. It then performs the commands in an expected manner.
What I find funny is that this is being touted as some sort of new AI revolution in space. Since our very first probes into LEO, we have been upgrading and complicating the controller software on every mission, be it Hubble, an Atlas V rocket, the mars rovers, or anything. Each new generation of spacecraft tends to have more complex, more robust flight software as the natural evolution of technology progresses. That said, I am not really sure why the ESA or NASA are talking about AI control software. This software isn't anymore AI oriented than the typical control software of any autonomous or semi-autonomous robot on the ground. It's only AI in the most liberal sense of the word. All that is happening is that, as missions and technology progress in maturity, engineers are capable of designing more robust control techniques using methods like Kalman filtering, direct adaptive control algorithms, state estimators, and so on. The only news here is that today missions are being designed with the capability to process more complex instructions set than they could 10, 20, or 30 years ago. That doesn't strike me as very newsish...but hey, I guess something had to fill the columns today.
Motorcycles, Robots, Space Gossip and More!
WTH? 29 posts and no one posted this: http://xkcd.com/695/
John Connor: You just can't go around killing people.
The Terminator: Why?
John Connor: What do you mean why? 'Cause you can't.
The Terminator: Why?
John Connor: Because you just can't, OK? Trust me on this.
Have gnu, will travel.
...as a remote agent, and actually controlled the craft for a few days.
There's no reason automated systems couldn't eventually earn the same level of trust that we place in humans.
Just like they will not punt a multi-million dollar telescope into orbit without testing the primary mirror? And they'd never shoot a $327 million Orbiter to mars without checking the math to make sure the units add up, right?
Face it, space agencies are run by people and governments. They are at least as prone to mistakes and financially driven shortcuts as any other element of human society.
Common Sense isn't as Common as people think...
Agreed, they look mostly like sophisticated expert systems.
Expert systems are a type of AI. In fact a highly successful type of AI. Aside from being highly successful, they also have the useful property of being predictable. Which is beneficial when the idea is to have a probe operating autonomously and without human supervision/observation for periods of time. An expert system may not be programmed to give optimal output in all situations, but unlike some other kinds of AI (neural nets for example) it is unlikely to go completely bonkers when given inputs outside of its training set.
AI simply means a program that tries to select an optimal behavior based on environmental input (and the environment doesn't even have to be the real world). AIs don't necessarily need to learn, self-modify, or do anything more sophisticated than take the set of inputs from sensors, and look up the proper response in a static table (your car probably contains such an AI).
Existing probes contain a small amount of AI, but they are still almost entirely dependent on human operators and this is about expanding their capabilities.
In popular culture, AI means Hal 9000 or Skynet, but in the context of NASA or just about any real, technical application (Expert Systems for example) that is not the definition being used.
The enemies of Democracy are
And that is just it... somehow we have this image that once we impart logic and "artificial intelligence" into a machine that it will somehow trascend into a god-like being. It will be just as limited and flawed as we are, but in different ways that we won't expect and may not be able to correct for.
And my other problem... what is the end goal of an intelligent artificial entity? Humans are driven by biological urges that have been ingrained into us over billions of years. What if the intelligence realized there is no real point to "life" and just chooses to end it all?
See my journal for slashdot ID's by year. Mine created in 2005. http://slashdot.org/journal/289875/slashdot-ids-by-year