Discussion Topic |
|
This thread has been locked |
Ed Hartouni
Trad climber
Livermore, CA
|
|
Science 17 Dec 2017:
eaao1733
DOI: 10.1126/science.aao1733
http://science.sciencemag.org/content/early/2017/12/15/science.aao1733
Superhuman AI for heads-up no-limit poker: Libratus beats top professionals
Noam Brown, Tuomas Sandholm
Abstract
No-limit Texas hold’em is the most popular form of poker. Despite AI successes in perfect-information games, the private information and massive game tree have made no-limit poker difficult to tackle. We present Libratus, an AI that, in a 120,000-hand competition, defeated four top human specialist professionals in heads-up no-limit Texas hold’em, the leading benchmark and long-standing challenge problem in imperfect-information game solving. Our game-theoretic approach features application-independent techniques: an algorithm for computing a blueprint for the overall strategy, an algorithm that fleshes out the details of the strategy for subgames that are reached during play, and a self-improver algorithm that fixes potential weaknesses that opponents have identified in the blueprint strategy.
Conclusions
Libratus presents an approach that effectively addresses the challenge of game-theoretic reasoning under hidden information in a large state space. The techniques that we developed are largely domain independent and can thus be applied to other strategic imperfect-information interactions, including non-recreational applications. Owing to the ubiquity of hidden information in real-world strategic interactions, we believe the paradigm introduced in Libratus will be important for the future growth and widespread application of AI.
http://static.ijcai.org/proceedings-2017/0772.pdf
|
|
MH2
Boulder climber
Andy Cairns
|
|
I may have missed it, but I did not find a consideration of whether Libratus can improve its play by playing against itself. I suppose the strategy for improvement depends heavily on the strategies used by opponents, so for a focus on human poker players there may be no point to having the program play against itself.
Sounds like an interesting development. Sooner or later such programs seem likely to be in competition against each other, as perhaps in rapidly changing investment opportunities, as may be happening now with today's algorithms.
|
|
Ed Hartouni
Trad climber
Livermore, CA
|
|
The algorithm does not "learn" the way the algorithms for perfect-information games do. However, this algorithm does adapt to the competitors' strategy.
When looking at the rather staggering number of decision points, even with the benefit of "abstraction" one thinks about how humans do so well. Same with games like chess and Go. The evolution of these sorts of abilities seem to be in the parallel network architecture of the brain, and the idea that many different pathways compete in a "vote," that is, we don't normally "think" in an organized algorithmic process with a single conclusion, but that we actually work out many different outcomes from which the final "answer" is a weighted sum of those different outcomes.
Think about the ability of a crowd to correctly estimate the weight of a bull, where any individual's estimate might be less accurate. The advantage of "crowd sourcing" such analysis is that many more calculations are available upon which to base an answer.
My guess is that computer power is now approaching and surpassing those human capabilities by raw computation (also parallel, it should be noted). Though the amount of energy used by the computers is huge compared to brains.
This also informs on the difficulty in unravelling the evolutionary path to human brain capability, that a huge part of the "fitness landscape" is the energy required to provide those capabilities has to return survival advantage. The complicated brain has to provide a "return on investment" at a rate higher than it's energy requirement.
If we had to design computers to manage on much less energy (and we will eventually) we'd have very different architectures than we currently have today. It is an interesting side question to ask if the energy consumption of modern computing is returning a net positive. My guess is that it is not yet, and this rather warm winter is evidence for that.
Perhaps we can look forward to Libratus like AI to help figure out a way forward. Time for this computational power to start earning its keep.
|
|
MH2
Boulder climber
Andy Cairns
|
|
When looking at the rather staggering number of decision points, even with the benefit of "abstraction" one thinks about how humans do so well. Same with games like chess and Go. The evolution of these sorts of abilities seem to be in the parallel network architecture of the brain, and the idea that many different pathways compete in a "vote," that is, we don't normally "think" in an organized algorithmic process with a single conclusion, but that we actually work out many different outcomes from which the final "answer" is a weighted sum of those different outcomes.
A weighted sum is one way to describe what happens when a neuron does or does not generate an action potential at the axon hillock depending on the synaptic inputs to its dendrites. The strength of the synaptic inputs can change. An important question is: how should the synaptic strengths change in order to produce a "better" outcome? In a massively interconnected network this is no easy question to answer. Libratus may make use of some kind of answer.
There are interesting things to be learned from comparing brains and computers. There may be similarities in the goals to be sought, in some cases.
However, there may be big differences in the energy it takes to achieve a sought-for outcome. As long as humans are motivated to up the ante, the energy cost won't matter to the machines.
|
|
eeyonkee
Trad climber
Golden, CO
|
|
I finally got back to looking at Ed's last posts, which are great ones, may I say.
Let's start with this.
The evolution of these sorts of abilities seem to be in the parallel network architecture of the brain, and the idea that many different pathways compete in a "vote," that is, we don't normally "think" in an organized algorithmic process with a single conclusion, but that we actually work out many different outcomes from which the final "answer" is a weighted sum of those different outcomes. Everything seems to point to something along these lines based on my readings and understanding. Weighted sums get right to the architecture of neural networks and memory, which depend on relative signal strengths. There are, however, some kind of neural processes that are at a higher level than weighted sum neural network processing (I got most of this from Pinker -- How the Mind Works) -- that use the results of the neural network processes as their input.
There is just no question that thousands maybe hundreds of thousands of different, individual, brain-related processes are going on all at the same time as we experience the world. Weighted-sum-type neural network processing probably explains the bulk of our automatic behavior, but something like the interpreter involves some processes working at a higher level than this.
This also informs on the difficulty in unraveling the evolutionary path to human brain capability, that a huge part of the "fitness landscape" is the energy required to provide those capabilities has to return survival advantage. The complicated brain has to provide a "return on investment" at a rate higher than it's energy requirement. Exactly! There is no reason for intelligence to evolve over and above what is needed -- mainly because intelligence, like everything else is a trade-off. If you devote this much to intelligence you need to take away that much from something else when you have an energy budget. There is every reason to believe that we will be able to make machines more intelligent than us. Consciousness is not intelligence, of course. Consciousness uses intelligence.
I must admit that I don't quite understand this.
It is an interesting side question to ask if the energy consumption of modern computing is returning a net positive. My guess is that it is not yet, and this rather warm winter is evidence for that.
MH2 said
There are interesting things to be learned from comparing brains and computers. There may be similarities in the goals to be sought, in some cases. What I would surmise about this is that computers need to be given goals. Humans do not. The nature of how human goals have come about is an evolutionary story, I'm sure.
|
|
eeyonkee
Trad climber
Golden, CO
|
|
Since I got this thread back to the front page already, anyhow, I feel that I need to include a separate post admitting that I was wrong about looking at genes as winners or losers. Ed was right, it is better to look at genes as temporarily successful or not based on their relative numbers in the current genome. Also, the genome does play a role in regulating genes, just like the individual organism does, so characterizing the genome as happenstance was also an incorrect statement based even on my own understanding. Just had to get that off of my chest.
It is interesting from a different levels of organization standpoint to consider how genes give rise to organisms, genomes, epigenomes and ecologies which, themselves, reach out from the future and regulate the gene.
This is how Siddartha Mukhergee sees it.
|
|
Ed Hartouni
Trad climber
Livermore, CA
|
|
I must admit that I don't quite understand this.
It is an interesting side question to ask if the energy consumption of modern computing is returning a net positive. My guess is that it is not yet, and this rather warm winter is evidence for that.
how much computer power is required to run climate models world wide?
The German Climate Center has a 3.6 petaflop computer, the average power consumption for the top 500 computers in the world is something like 10 MFlops/Watt so to get to 3.6 e15 flops you have to multiple the 1e7 flops/watts by 3.6e8 watts, something like 360 MWatts. These are more efficient machines, presumably, but the most efficient are use two orders of magnitude less power, so maybe something like 10 MWatts of power....
if there are 10 such centers around the world we're using about 100 MWatts for all the computation.
this translates (if I got the calculation right) into about 1e5 Tonnes of CO2 emitted by the power generation required to do the calculations per year, or about 0.2 microCentigrade/yr surface temperature increase.
How much decrease do the calculations produce?
|
|
WBraun
climber
|
|
Some guy on this forum resurrected an old thread and posted in it "How did I ever miss this"
All while we walk/drive down the path and miss an infinite number of things around us at every second.
And simultaneously make arrogant statements there is nothing beyond what we can measure or see with our puny little selves.
The conscious level of awareness of the living entity is always very limited and given on a need to know basis according to time and circumstance.
The foolish arrogant gross materialists can never gain complete knowledge independently without the higher authorities that they so constantly, arrogantly and blindly dismiss.
Using computer analogies for the living entity is the height of scientific arrogance for the gross materialists has no clue to the living entity itself.
The foolish gross materialists are like people who study the machine (gross physical and subtle body) and all its parts down to the minutest details and never study the driver of the machine.
The machine (human body) is never the living entity itself ......
|
|
Ed Hartouni
Trad climber
Livermore, CA
|
|
want to improve forgetfulness?
A Brain Implant Improved Memory, Scientists Report
"Scientists have developed a brain implant that noticeably boosted memory in its first serious test run, perhaps offering a promising new strategy to treat dementia, traumatic brain injuries and other conditions that damage memory.
The device works like a pacemaker, sending electrical pulses to aid the brain when it is struggling to store new information, but remaining quiet when it senses that the brain is functioning well..."
Nature Communications volume 9, Article number: 365 (2018)
doi:10.1038/s41467-017-02753-0
https://www.nature.com/articles/s41467-017-02753-0
Closed-loop stimulation of temporal cortex rescues functional networks and improves memory
Youssef Ezzyat, Paul A. Wanda, Deborah F. Levy, Allison Kadel, Ada Aka, Isaac Pedisich, Michael R. Sperling, Ashwini D. Sharan, Bradley C. Lega, Alexis Burks, Robert E. Gross, Cory S. Inman, Barbara C. Jobst, Mark A. Gorenstein, Kathryn A. Davis, Gregory A. Worrell, Michal T. Kucewicz, Joel M. Stein, Richard Gorniak, Sandhitsu R. Das, Daniel S. Rizzuto & Michael J. Kahana
Abstract
Memory failures are frustrating and often the result of ineffective encoding. One approach to improving memory outcomes is through direct modulation of brain activity with electrical stimulation. Previous efforts, however, have reported inconsistent effects when using open-loop stimulation and often target the hippocampus and medial temporal lobes. Here we use a closed-loop system to monitor and decode neural activity from direct brain recordings in humans. We apply targeted stimulation to lateral temporal cortex and report that this stimulation rescues periods of poor memory encoding. This system also improves later recall, revealing that the lateral temporal cortex is a reliable target for memory enhancement. Taken together, our results suggest that such systems may provide a therapeutic approach for treating memory dysfunction.
|
|
Ward Trotter
Trad climber
|
|
Here is a much less intrusive way to improve brain function, especially if you spend much time in dimly lit rooms, and happen to be a Nile rat:
Spending too much time in dimly lit rooms and offices may actually change the brain's structure and hurt one's ability to remember and learn, indicates groundbreaking research by Michigan State University neuroscientists.
https://m.medicalxpress.com/news/2018-02-dim-dumber.html
|
|
eeyonkee
Trad climber
Golden, CO
|
|
I like this from Ward Trotter's link.
"In other words, dim lights are producing dimwits." Who knew?
|
|
Largo
Sport climber
The Big Wide Open Face
|
|
|
Topic Author's Reply - Feb 7, 2018 - 05:09pm PT
|
Gazzaniga goes on to explain that the unified story is put together in a very certain part of the left brain hemisphere dubbed the interpreter.
The evolutionary/scientific model of human consciousness easily incorporates these findings. I would think that both Largo's and Tom's worldviews cannot explain these findings so easily.
The amazing thing to me is that you believe "the interpreter" explains anything, and easily at that. Everything in consciousness is tied up to the brain, and the brain is tired up with everything else. But none of this "explains" why we are aware of anything, or why "the interpreter" is aware of what he/she is interpreting.
Again, when awareness is conflated with content, the whole works get muddled.
Conflation is always the turd in the punchbowl in these discussions. For example, Ed said:
but for those shared experiences we might be in substantial agreement.
sounds like the beginnings of objective experience.
Here, Ed has conflated an objective cognitive take (agreement) of experience, and has gone on to attribute that objectivity to subjective experience itself.
Strangely, there IS such a thing as objective experience, but that's another story.
|
|
eeyonkee
Trad climber
Golden, CO
|
|
Welcome back, Largo! Yeah, I think that the interpreter explains a lot. I think that the interpreter represents a higher level of processing in the brain (by higher, I mean that it uses the unconscious, “lower” processes as inputs), and that somehow this is what we experience as consciousness or mind. That’s my current assessment, although I am not married to it. Also, I would say that genes are the ultimate movers and creators of intelligence and mind. Without an understanding of genetics and evolution, you will be without moorings in trying to understand mind.
|
|
WBraun
climber
|
|
I would say that genes are the ultimate movers and creators of intelligence and mind.
Again the gross materialist using "Ulitmate" and always simultaneously denying any existence of Ultimate.
Shows you really don't even know what you doing, to begin with.
You have no clue about the actual foundation of life itself ......
|
|
MH2
Boulder climber
Andy Cairns
|
|
Charles Darwin was a wonderful observer and analyst.
My Dad was an embryologist. He learned techniques and got direction on which questions were of interest from Viktor Hamburger.
https://embryo.asu.edu/pages/viktor-hamburger-1900-2001
Rita Levi-Montalcini once came to visit us in Springville, NY.
|
|
WBraun
climber
|
|
Charles Darwin was a wonderful observer and analyst that made a fatal and epic mistake.
He could NOT see things as they are but only saw them as he thought they are .......
|
|
zBrown
Ice climber
|
|
Not to throw water on the parade, but suppose just for the sake of argument [doesn't everybody love a good argument as well as a parade] suppose we all reach an agreement as to what mind is.
What we gonna do when some fool asks "isn't mind evolving?" And our current level of understanding out of date?
|
|
WBraun
climber
|
|
Mind doesn't evolve, consciousness evolves or devolves.
Mind is only a servant of consciousness .......
|
|
zBrown
Ice climber
|
|
^
Doesn't that imply that what mind is then has been determined?
I guess we can all get some much deserved rest!
:)
BTW
is this legal syntax
~:)
|
|
|
SuperTopo on the Web
|
Let us know!