6 ms·

Unfortunately, many of your assertions are inaccurate. As far as game playing goes, the ways in which AI agents beat human opponents is simply a matter of compu

Unfortunately, many of your assertions are inaccurate. As far as game playing goes, the ways in which AI agents beat human opponents is simply a matter of computers being able to process more information in a given period of time. The selection and evaluation criteria are actually pretty similar to human decision-making, if a bit more systematic. I can expand on this point in another comment, if you'd like. Chess and Go are not solved games, and will likely not be for some time (10+ years, for Go at least). Solving non-deterministic problems are even harder, and many likely will not be accomplished within our lifetimes[1]. There is no AI system that can generally make more accurate diagnoses than a physician. We're not there quite yet, and it will be a while. The best we've come are some pretty advanced expert systems, but these require very heavy input from physicians[2]. Please provide sources to validate your claims on the progress of AI. Extrapolation from data is dangerous, extrapolation from falsity is ignorance. As far as your take on understanding goes, the human "meaning" of things is inherently subjective. If we define "meaning" as some thing's value or role based on environmental context, then just like humans, any artificially intelligent system will only be able to determine meaning based on observations of the environment and both individual and collective experience. It's interesting that you seem to be focusing primarily on social domains that have inherently "human" contexts. Beyond misunderstanding the point of this paper, and the way AI systems work, I think you're missing the point. AI is, and for the time being will remain, an extension of the human mind. The decision-making needs to be developed (at some core level) by a human. Those goals need to be set by a human. The experiences and observational capabilities ultimately need to be determined by a human. Even AI systems that build other AI systems need to be directed to do so, and with strict goals, set by a human [3]. I highly suggest you take a moment to read openai's mission statement: https://openai.com/about/ https://openai.com/about/. AI is a tool. And like any powerful tool it must be used responsibly, freely, and openly. Openai is pursuing this goal and making efforts to ensure that this tool is available to as many people as possible to avoid the abuses implicit to your concerns. You obviously have an interest in AI and some knowledge of the field, but I worry your comments veer a bit towards fear-mongering. I suggest you use openai and resources like it to enrich your knowledge of both the advances and concerns of AI, because those are important, and we definitely need people thinking about these things. Ultimately, you are absolutely correct that eventually these systems will probably have the technical capability to influence elections, the economy, and more. But the only way they will is under the direction of humans. It it not the machine you should fear, but the man behind it. The same thing you ought to have been fearing all along. 1. http://fragrieu.free.fr/SearchingForSolutions.pdf http://fragrieu.free.fr/SearchingForSolutions.pdf 2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1307157/ https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1307157/ 3. https://arxiv.org/abs/1611.02779 https://arxiv.org/abs/1611.02779

rocmcd 9y ago
"Extrapolation from data is dangerous, extrapolation from falsity is ignorance." Beautifully stated.
EGreg 9y ago
Yes please expand. BUT you are attacking a straw man. First of all I did not say Chess and Go are Solved!! I said that computers will be able to get solutions in our systems faster than we can and of such high quality that it's pointless to even challenge them (for a doctor or poker player or a man running away from a robot or someone trying to prevent the ruin of their reputation or someone trying to have a trust based relationship with their neighbor in a crazy world). It is also not true that it is "simply" a matter of computers being able to process more information in a given period of time. This is a major deal. Imagine a textile worker 3000 years ago making clothes. Now imagine the tread count today of a cheap shirt. No one would think of making shirts of such high thread counts back then. With Chess and especially with Go, there is just a whole other level of intellence. It misses the "human meaning" part but it can manipulate a vector that involves 10,000,000 variables. It is in terms of such vectors that concepts like "cat", "dog" or diagnoses are made. How to explain why it made this diagnosis in a way a human can confirm? What I said was actually pretty straightforward: in 17 years computers will be able to rap and make better jokes than humans and convince better than humans. They will be able to hack our systems of reputation, trust, voting, legal argument and so on. What's scary is that we will essentially be giving up control to systems that don't really have the same concepts as we do, and perhaps never will. And if one day it all blows up somehow or gradually shifts to something unpleasant for humans, then that's freaking scary! Just look at how we are already doing it when it comes to wealth inequality. As a society we are richer than ever but the inequality is greater. This is just a mild example. What if computers could do a lot more?
- readonlydanger 9y ago
  Happy to! And I'm not sure I am, but you're right in that I should be more explicit in my point (I have a tendency towards logorrhea). I think if I'm attacking anything, it's what I believe to be the subtext of your comments: 'AI = Bad = The Terminator'. Pardon me if the following comes off as pedantic, I do not know your level of expertise and want to continue the discussion from the same point of base knowledge. To my game playing point, and to start from a simple example, in Tic-Tac-Toe, an AI opponent will always beat humans because they can search through every possible move to end-game and always make the correct move. For games like Chess and Go, the search space is far too large to search to a terminal board position, so we need to use ML and heuristics to evaluate the value of a given board. These evaluation functions and heuristics are designed by utilizing human insight into the game. IBM's retrospective on Deep Blue is a fascinating read, and I suggest anyone check it out if they're interested in AI and game playing[1]. You'll see how they built their evaluation function with the input of chess grandmasters, particularly in implementing opening books and prioritizing center play in the early game. AlphaGo's system is not entirely dissimilar[2]. You'll note that significant advances were made as AlphaGo continued to play and learn from human opponents. The point is that both these systems (and all game playing AI agents that I know of) search through possible board states and make decisions that are fundamentally reliant on the intuitions of humans. Furthermore, humans make decisions like this as well. We generate all the possible decisions we can make, and rule out invalid choices either implicitly (though sub-conscious bias and knowledge) or explicitly (thinking through a decision). Computers do not have the advantage of implicit evaluation, so we must program that explicit evaluation and use massive amounts of data (for deep learning, anyway) with ML techniques to validate those intuitions. Both Go and Chess are deterministic games. Given that even these games haven't been solved, the stochasticity of the various domains you described are orders of magnitude more complex and we honestly need several breakthroughs before we come close (I have no sources here, this is just my opinion. I feel that most of the success of AI right now is the standard M.O. of a lot of Academic CS: Things 'work' for certain definitions of 'work'. The breakthroughs are great and impressive, but the constant extrapolation by pundits and the general media is both irresponsible and fallacious). And yes, it really is a lot about processing power. Neural Networks have been falling in and out of fashion since the 1950s. One big factor of the recent ressurgance in popularity is due to GPU utilization and cloud computing (admittedly, the availability of data via the internet is another large factor, among others). That's why Google, NVidia, Apple, and others are investing so much into ML specific hardware. And let's not kid ourselves: training any ML model takes a lot of time and a lot of manual adjustment of hyper-parameters. We're talking about possibly hundreds of hours of manual input for a single model (novel ones mostly). That's why every minor breakthrough merits a white paper (sort of joking, sort of not...)! I think we're making the same argument with your linear algebra example; that machines can't reasonably replace humans. My amended version of that argument is that machines can and should extend and augment human capability. Despite the linear algebra that happens, any form of decision making and cognition is in someway designed by a human. So despite the vectorization of the world (as seen by an AI), they will process through the lens of human cognition; because I don't think we can build systems that don't somehow stem from our own cognitive processes. As to your specific fears, I seriously doubt we'll be able to make enough progress within 17 years for AI to dominate those fields completely. I agree that AI will probably become a presence in many of those domains, but I do not think we will be "giving up control". Remember, these systems will be operated by individuals. So, I would say that there is some evidence to suggest that within 10-20 years, humans will be using AI to produce higher quality art, jurisprudence, etc. It is also true, that this raises the possibility for humans to abuse this technology, but this is inescapable for almost any human achievement. I think openness and transparency is the best safe-guard against this possibility, and I would encourage everyone to vocally oppose any integration of AI into public systems without extreme transparency. Beyond that, humans that have and always will be the cause of wealth inequality. Also, you provide no evidence that inequality is greater. In the 21st century, inequality has increased [I don't feel like sourcing this, but Google is reasonably good here], but I would like to see research confirming that we're worse off than the heights of Feudal society, or other equally tyrannical periods of human history. How would you foresee AI contributing to wealth inequality? I only see AI as a contributing factor to increasing wealth inequality if it remains in the hands of few. I've definitely rambled here, but I blame that on the few drinks I've had. I think my point still stands; humans kill people, AI doesn't kill people (and we're at a point where I think we can ensure that it doesn't). As a side-question, what exactly is significant about 17 years? Is there some prediction out there that uses this number, or was it an arbitrary number? 1. https://pdfs.semanticscholar.org/ad2c/1efffcd7c3b7106e507396bdaa5fe00fa597.pdf https://pdfs.semanticscholar.org/ad2c/1efffcd7c3b7106e507396... 2. https://storage.googleapis.com/deepmind-media/alphago/AlphaGoNaturePaper.pdf https://storage.googleapis.com/deepmind-media/alphago/AlphaG...

rocmcd 9y ago

"Extrapolation from data is dangerous, extrapolation from falsity is ignorance." Beautifully stated.

EGreg 9y ago

Yes please expand. BUT you are attacking a straw man. First of all I did not say Chess and Go are Solved!! I said that computers will be able to get solutions in our systems faster than we can and of such high quality that it's pointless to even challenge them (for a doctor or poker player or a man running away from a robot or someone trying to prevent the ruin of their reputation or someone trying to have a trust based relationship with their neighbor in a crazy world). It is also not true that it is "simply" a matter of computers being able to process more information in a given period of time. This is a major deal. Imagine a textile worker 3000 years ago making clothes. Now imagine the tread count today of a cheap shirt. No one would think of making shirts of such high thread counts back then. With Chess and especially with Go, there is just a whole other level of intellence. It misses the "human meaning" part but it can manipulate a vector that involves 10,000,000 variables. It is in terms of such vectors that concepts like "cat", "dog" or diagnoses are made. How to explain why it made this diagnosis in a way a human can confirm? What I said was actually pretty straightforward: in 17 years computers will be able to rap and make better jokes than humans and convince better than humans. They will be able to hack our systems of reputation, trust, voting, legal argument and so on. What's scary is that we will essentially be giving up control to systems that don't really have the same concepts as we do, and perhaps never will. And if one day it all blows up somehow or gradually shifts to something unpleasant for humans, then that's freaking scary! Just look at how we are already doing it when it comes to wealth inequality. As a society we are richer than ever but the inequality is greater. This is just a mild example. What if computers could do a lot more?

readonlydanger 9y ago
Happy to! And I'm not sure I am, but you're right in that I should be more explicit in my point (I have a tendency towards logorrhea). I think if I'm attacking anything, it's what I believe to be the subtext of your comments: 'AI = Bad = The Terminator'. Pardon me if the following comes off as pedantic, I do not know your level of expertise and want to continue the discussion from the same point of base knowledge. To my game playing point, and to start from a simple example, in Tic-Tac-Toe, an AI opponent will always beat humans because they can search through every possible move to end-game and always make the correct move. For games like Chess and Go, the search space is far too large to search to a terminal board position, so we need to use ML and heuristics to evaluate the value of a given board. These evaluation functions and heuristics are designed by utilizing human insight into the game. IBM's retrospective on Deep Blue is a fascinating read, and I suggest anyone check it out if they're interested in AI and game playing[1]. You'll see how they built their evaluation function with the input of chess grandmasters, particularly in implementing opening books and prioritizing center play in the early game. AlphaGo's system is not entirely dissimilar[2]. You'll note that significant advances were made as AlphaGo continued to play and learn from human opponents. The point is that both these systems (and all game playing AI agents that I know of) search through possible board states and make decisions that are fundamentally reliant on the intuitions of humans. Furthermore, humans make decisions like this as well. We generate all the possible decisions we can make, and rule out invalid choices either implicitly (though sub-conscious bias and knowledge) or explicitly (thinking through a decision). Computers do not have the advantage of implicit evaluation, so we must program that explicit evaluation and use massive amounts of data (for deep learning, anyway) with ML techniques to validate those intuitions. Both Go and Chess are deterministic games. Given that even these games haven't been solved, the stochasticity of the various domains you described are orders of magnitude more complex and we honestly need several breakthroughs before we come close (I have no sources here, this is just my opinion. I feel that most of the success of AI right now is the standard M.O. of a lot of Academic CS: Things 'work' for certain definitions of 'work'. The breakthroughs are great and impressive, but the constant extrapolation by pundits and the general media is both irresponsible and fallacious). And yes, it really is a lot about processing power. Neural Networks have been falling in and out of fashion since the 1950s. One big factor of the recent ressurgance in popularity is due to GPU utilization and cloud computing (admittedly, the availability of data via the internet is another large factor, among others). That's why Google, NVidia, Apple, and others are investing so much into ML specific hardware. And let's not kid ourselves: training any ML model takes a lot of time and a lot of manual adjustment of hyper-parameters. We're talking about possibly hundreds of hours of manual input for a single model (novel ones mostly). That's why every minor breakthrough merits a white paper (sort of joking, sort of not...)! I think we're making the same argument with your linear algebra example; that machines can't reasonably replace humans. My amended version of that argument is that machines can and should extend and augment human capability. Despite the linear algebra that happens, any form of decision making and cognition is in someway designed by a human. So despite the vectorization of the world (as seen by an AI), they will process through the lens of human cognition; because I don't think we can build systems that don't somehow stem from our own cognitive processes. As to your specific fears, I seriously doubt we'll be able to make enough progress within 17 years for AI to dominate those fields completely. I agree that AI will probably become a presence in many of those domains, but I do not think we will be "giving up control". Remember, these systems will be operated by individuals. So, I would say that there is some evidence to suggest that within 10-20 years, humans will be using AI to produce higher quality art, jurisprudence, etc. It is also true, that this raises the possibility for humans to abuse this technology, but this is inescapable for almost any human achievement. I think openness and transparency is the best safe-guard against this possibility, and I would encourage everyone to vocally oppose any integration of AI into public systems without extreme transparency. Beyond that, humans that have and always will be the cause of wealth inequality. Also, you provide no evidence that inequality is greater. In the 21st century, inequality has increased [I don't feel like sourcing this, but Google is reasonably good here], but I would like to see research confirming that we're worse off than the heights of Feudal society, or other equally tyrannical periods of human history. How would you foresee AI contributing to wealth inequality? I only see AI as a contributing factor to increasing wealth inequality if it remains in the hands of few. I've definitely rambled here, but I blame that on the few drinks I've had. I think my point still stands; humans kill people, AI doesn't kill people (and we're at a point where I think we can ensure that it doesn't). As a side-question, what exactly is significant about 17 years? Is there some prediction out there that uses this number, or was it an arbitrary number? 1. https://pdfs.semanticscholar.org/ad2c/1efffcd7c3b7106e507396bdaa5fe00fa597.pdf https://pdfs.semanticscholar.org/ad2c/1efffcd7c3b7106e507396... 2. https://storage.googleapis.com/deepmind-media/alphago/AlphaGoNaturePaper.pdf https://storage.googleapis.com/deepmind-media/alphago/AlphaG...