(Phys.org)—Decision-making—the ability to choose one path out of several options—is generally considered a cognitive ability possessed by biological systems, but not by physical objects. Now in a new study, researchers have shown that any rigid physical (i.e., non-living) object, such as an iron bar, is capable of decision-making by gaining information from its surroundings accompanied by physical fluctuations.
The researchers, Song-Ju Kim, Masashi Aono, and Etsushi Nameda, from institutions in Japan, have published their paper on decision-making by physical objects in a recent issue of the New Journal of Physics.
"The most important implication that we wish to claim is that the proposed scheme will provide a new perspective for understanding the information-processing principles of certain lower forms of life," Kim, from the International Center for Materials Nanoarchitectonics' National Institute for Materials Science in Tsukuba, Ibaraki, Japan, told Phys.org. "These lower lifeforms exploit their underlying physics without needing any sophisticated neural systems."
As the researchers explain in their study, the only requirement for a physical object to exhibit an efficient decision-making ability is that the object must be "volume-conserving." Any rigid object, such as an iron bar, meets this requirement and therefore is subject to a volume conservation law. This means that, when exposed to fluctuations, the object may move slightly to the right or left, but its total volume is always conserved. Because this displacement resembles a tug-of-war game with a rigid object, the researchers call the method "tug-of-war (TOW) dynamics."
Here's an example of how the idea works: Say there are two slot machines A and B with different winning probabilities, and the goal is to decide which machine offers the better winning probability, and to do so as quickly as possible based on past experiences.
The researchers explain that an ordinary iron bar can make this decision. Every time the outcome of a play of machine A ends in a reward, the bar moves to the left a specific distance, and every time the outcome ends in no reward, the bar moves to the right a specific distance. The same goes for a play of machine B, but the directions of the bar movements are reversed. After enough trials, the bar's total displacement reveals which slot machine offers the better winning probability.
The researchers explain that the bar's movements occur due to physical fluctuations.
"The behavior of the physical object caused by operations in the TOW can be interpreted as a fluctuation," Kim said. "Other than this fluctuation, we added another fluctuation to our model. The important point is that fluctuations, which always exist in real physical systems, can be used to solve decision-making problems."
The researchers also showed that the TOW method implemented by physical objects can solve problems faster than other decision-making algorithms that solve similar problems. The scientists attribute the superior performance to the fact that the new method can update the probabilities on both slot machines even though it plays just one of them. This feature stems from the fact that the system knows the sum of the two reward probabilities in advance, unlike the other decision-making algorithms.
The researchers have already experimentally realized simple versions of a physical object that can make decisions using the TOW method in related work.
"The TOW is suited for physical implementations," Kim said. "In fact, we have already implemented the TOW in quantum dots, single photons, and atomic switches."
By showing that decision-making is not limited to biological systems, the new method has potential applications in artificial intelligence.
"The proposed method will introduce a new physics-based analog computing paradigm, which will include such things as 'intelligent nanodevices' and 'intelligent information networks' based on self-detection and self-judgment," Kim said. "One example is a device that can make a directional change so as to maximize its light-absorption." This ability is similar to how a young sunflower turns in the direction of the sun.
Another possibility that the researchers recently explored is an analogue computer that harnesses natural fluctuations in order to maximize the total rewards "without paying the conventionally required computational cost."
Explore further: Quantum dots make efficient decisions
More information:
Song-Ju Kim, et al. "Efficient decision-making by volume-conserving physical object." New Journal of Physics. DOI: 10.1088/1367-2630/17/8/083023
docile
ppnlppnl
LariAnn
surprisingly
ACoffeeDrinker
Jaeherys
I think this is a hint that everything is "alive" to some degree; I had more or less been leaning towards the ability to metabolize as a mechanism to maintain some level of stability denoted as alive but that itself has not questioned the underlying question of is there such a thing as a live or dead?
I really think we have totally misunderstood nature in a fundamental way; we are part of the whole, not above it.
Jaeherys
Imagine light impacting some molecules with the appropriate atomic switches to determine whether it changes its buoyancy or not based on conformational changes. The one thing I've learned from molecular biology, all decision making networks can and will be used in one way or another within the cell. Nature does not limit itself in this sense, i.e. processes are not hidden from one another.
SciTechdude
lrshultis
"Decide" presupposes the concept of "consciousness" in any living thing that decides to take an action.
If they want to redefine "decide" to include non conscious physical activity then create a new concept that does so and not pervert perfectly valid concept like "decide" which presupposes the concept "consciousness".
docile
ppnlppnl
Good bye stupid stupid world.
sascoflame
meerling
I'm sorry, but I don't really see anything profound or new about using a physical object for this function in any way.
If there's something I'm missing, I'd appreciate it if someone would point out just what it is.
Guy_Underbridge
Eikka
I can't see the point either. We could just as well move a tiny stone left and right depending on which machine wins, and the end result of that random walk will place the pebble nearer to the machine that wins more.
There's absolutely no need for any "volume conserving" mechanisms. It's just a tally of the number of wins subtracted. If you kick a stone, the stone will "count" how many kicks it has recieved. What decision making is there to be had?
Eikka
"Volume conserving" is just a roundabout way of saying that the bar is rigid. When you move one end, the other end moves along.
What they're essentially saying is that an abacus-like device can be used to count, which is completely obvious and that's not the main point of the article. They're really discussing a different way (tug-of-war dynamics) of using an abacus to solve a statistical problem, and the iron bar is completely irrelevant to the question.
The difference in this case is that instead of just rewarding the winning machine as in a random walk situation, you punish the machine for losing as well, and that makes for an efficient algoritm.
Eikka
Basically, their proposed method can be applied to any sufficiently volume-conserving object - such as a person.
Following the suggested operations, in the simple case of two slot machines A and B, whenever machine A wins the person is moved to machine A unless they are already there, and whenever machine A loses, the person is moved to machine B unless already there. The same is done with machine B in reverse.
That way the person keeps playing the winning machine until he loses, then switches machines until he gets another winning streak, and this is proves to be an efficient means to keep the player at the winning machine.
Earthman
qitana
Eikka
Indeed. The article is grossly misrepresenting their research.
OdinsAcolyte
enjoyed the article
cgsperling
PhotonX
.
This looks about as brilliant as the rain-predicting rock.
.
.
Whydening Gyre
"Random-ness" works in rotation. But - it does it in 2 directions...:-) Back to forward (or top to bottom, if you prefer), (it's)right to left... the 3rd direction (axis, if you will) is just a combination of the first 2. Take a little time to absorb this... you'll get it...
Ludwig99b0