io9 
Enter your username and password.
San Francisco, 7:49 AM
Sun Nov 29
10 posts in the last 24 hours
Tip your editors: tips@io9.com
Editor-in-Chief:
Annalee Newitz | Email
News Editor:
Charlie Jane Anders | Email
Associate Editor:
Meredith Woerner | Email
Assistant Editor:
Lauren Davis | Email
Weekend Editor:
Graeme McMillan | Email
Contributors:
Joshua Glenn
Stephen Goldmeier | Email
Ed Grabianowski | Email
Austin Grossman
Paul Hogan | Email
Lauren Davis | Email
Chris Hsiang | Email
Lynn Peril | Email
Ann VanderMeer
Alasdair Wilkins | Email
Graphic Designer:
Stephanie Fox | Email
Interns:
Tim Barribeau | Email
Julia Carusillo | Email
Alex Eichler | Email
Cyriaque Lamar | Email
Caitlin Petrakovitz | Email
Mary Ratliff | Email
Josh Snyder | Email
Original material is licensed under a Creative Commons License permitting non-commercial sharing with attribution.
08/07/09
08/07/09
08/06/09
08/06/09
Please teach the AI the ethical implications of using the warps.
Signed
John Conner.
08/06/09
What's the frequency of the human mind for awareness? I bet reaction times cant get too far below 100mS. To paraphrase Data, to an android, that's an eternity.
08/06/09
08/06/09
08/06/09
08/06/09
Wish I could program like this.
08/06/09
08/06/09
08/06/09
But it takes a neural network to learn how do play the game well. Like, as phoenix says below, when to use the pow block. It requires intelligence to learn how to balance priorities (points, coins, power ups) and adapt to changes (abilities, new bad guys).
Like chess, there is a perfect solution to every Mario level, given a victory criteria, but it's so subtle and complex that the only way to find it is billions of learning iterations.
08/07/09
08/07/09
Very good point on the swimming since movement isn't straightforward. But I think this would just amount to an annoying interface to program (much like running jumps) since the algorithm can determine exactly where Mario will be after each press.
08/06/09
Kill it or enslave it.
08/06/09
08/06/09
08/06/09
08/06/09
08/06/09
For example, if this is the original Mario, we all had to plan when to use that POW block. Do I think I can last a bit longer without it? When should I move to another level? I have a fireball coming at me at one end and an enemy in my path...how do I avoid? I have a new enemy...and it's a crab. Wait, one hit didn't do anything!
I see this as being a true "learning algorithm" versus chess is which based more on probabilities than "learning" though it does have a significant amount of learning as well. If they pull this off, this would be very impressive.
08/06/09
08/06/09
The sorts of value judgments that @mr_dude mentions above (prioritizing coins, power-ups, etc.) are much more complicated than the value judgments in chess once we start to formalize them. The human sensory-perceptive experience is a crucial part of SMB, but not chess, which can be easily internalized as a set of formally-defined states. To me, this is interesting in that it's not only an AI test, but it also examines the participants' ability to process some sorts of information in an external environment.
@MonkeyT: As valuable as the Turing Test is, I think that it's implicitly a very anthropocentric test. I'd argue that passing the Turing test is a sufficient, but not necessary, requirement for the possession of intelligence. A great test, but it's like trying to run before we can walk: it's requiring the automaton to be intelligent in a specifically-human sort of way.
It's the reliance on language that creates this situation; you can think of other intelligent creatures that don't have linguistic capabilities. If we look to the animal behaviorists, who love this sort of thing, we find that they frequently look for behavioral flexibility in making ascriptions of intelligence.
So, with this in mind, I ran an experiment that was designed to be a sort of preliminary "behavioral" Turing Test. We wrote a program for a Sony AIBO that consisted of a complete behavioral repertoire: it would respond to visual/vocal/tactile stimulation, engage in behavior on its own, and could chase and kick a ball around. We also had a WiFi card installed in the little mutt that let me see through its camera, hear through its microphones, and see where it was being touched while controlling it with a PS2 controller.
We let participants play with the AIBO for a few minutes in each of "program-controlled" and "Feroh-controlled" modes, and then quizzed them on how closely different components of the AIBO's behavior seemed "human" or computer-controlled, ultimately asking them to determine which mode was which.
That sort of pitch worked well to draw participants, at least: play with a robotic dog for a few minutes and we'll pay you.
08/07/09
The trouble with contests like these is that some people see AI as a question of intelligence and others see it as a matter of sentience. I fall in the sentience camp.
My own definition of AI will be when there is no living human human who is capable of predicting with accuracy how the AI will behave - when the complexity of its code outstrips our comprehension of its processes, yet it still functions in a rational (yet possibly not entirely logical) manner. This is inevitable. We will reach a point when code will write coding software, first in programming languages we use, then they will develop their own, more concise programming languages, where there is absolutely no requirement for human interpretation. They will make procedural leaps we will not understand, and the first app that works that way will be 'alive'. Of course, apps will outsmart me long before it reaches that point.
08/07/09
You realize that, by your definition, most humans are not sentient, don't you?
08/07/09
I wonder if it matters that an artificial intelligence will fail most tests that have been presented. I would say an important feature of intelligence (but not definition) is adaptability and survival. A well-developed ai would have the cognizance to intentionally fail the Turing test or predictability test.
Speaking purely, this shouldn't matter since we'd like to think of a scientific/honest experiment for intelligence. And maybe all intelligence has a phase of 'innocence'. But the nature of any intelligence is that you can't induce honesty.
08/08/09
:->
08/10/09
08/06/09
08/06/09
08/06/09
08/06/09
Computer: "Impossible. All my circuits are functioning perfectly. DOES NOT COMPUTE"
08/06/09