9 ms·

Ask HN: Are you concerned about AI?

13 points by nvahalik 9y ago

Elon Musk says that if you aren't concerned, you should be. If you are concerned, why? If not, then why not?

paulcole 9y ago
Not at all. I'll be long dead before anything happens.
nigrioid 9y ago
No - we have only a very basic understanding of how the brain works, have no idea what consciousness even is, and know little about human intelligence. The only thing I can think of is there's a sinister agenda pushing these constant fear-mongering stories about how we're all going to die from AI. Show me a neural net that even comes close to doing anything resembling original, organized thought and I might take these warnings more seriously.
- ljk 9y ago
  stupid question, but the AI/neural net these days require training right? look at what happened to microsoft's AI bot that turned rogue after just a few days and had to be turned off
  - paulcole 9y ago
    https://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist https://www.theverge.com/2016/3/24/11297050/tay-microsoft-ch... It didn't "go rogue" it did what it was supposed to. People were just being shits to it.
    - ljk 9y ago
      exactly, so "real" AI can be trained by the same people too and have the same outcome?
      - paulcole 9y ago
        somehow i think that "real" AI won't be trained by the minds of 4chan
- jostmey 9y ago
  We still don't know everything about the biology of bird feathers but we can make flying machines. True AI is not going to happen tomorrow, but I expect it to come within my lifetime. It's easy to get hung up on the complexity of biological systems like the brain, but sometimes all that complexity is not necessary to the function. Complexity is the result of an evolutionary process, and may not be a necessary component of intelligence
  - studentrob 9y ago
    Evolution has taken us through an uncountable number of neurons and lifespans. In my opinion, even if we had an AI that matched performance of the human brain, it'd still be missing the historical data on earth (and the universe) we used to arrive at this point, the level of parallelization that exists across the earth across all life forms, and the capacity to input as much data as all life on earth can. Data and input capacity are as important as the model in machine learning. I think we sometimes forget how interdependent we are. We learn a lot through sharing our experiences, and when there are more of us, we make more breakthroughs as a species. And, maybe the answer is 42.
  - oldandtired 9y ago
    I started to reply in detail to your comment, but I have decided to stop because there is so much wrong with it. So, I'll leave it with the following, Biological systems are: highly complex self-replicating self-refueling, low-power, low-speed, highly manoeuvrable Our attempts at replication of the abilities lead to brute force high powered, simple non self-replicating, non self-refueling low manoeuvrability needing high speed to achieve same goal And we know very little about biological systems and most of what we know is probably wrong as well.
nvahalik 9y ago
I think thus far the only thing I'm really concerned about is our trust/dependence on it. Articles like this one (https://www.autoblog.com/2017/08/04/self-driving-car-sign-hack-stickers/ https://www.autoblog.com/2017/08/04/self-driving-car-sign-ha...) show just how fragile some of this ML stuff is. Of course, I'm also assuming that ML = AI... that is a correct assumption, right?
- larkeith 9y ago
  IIRC AI will probably emerge from and incorporate ML, but it's unlikely that AI will be entirely ML-based.
Animats 9y ago
We're already at "machines should think, people should work" for low-level jobs, like fast food and order fulfillment. Robots are still not good at, or cost-effective at, unstructured manipulation, but computers are great at organizing work, communicating, selling, and accounting. What management used to do.
- oldandtired 9y ago
  Computers are not great at organising anything, cannot communicate, cannot sell and cannot do accounting. It requires a person to "program" these facilities. Without the human generated programs, no computer is capable of doing anything. They are no different to a hammer, a crane or crowbar. They are simply tools in the hands of a human. They are completely useless if not picked up and used appropriately. What is unfortunate is that there are many people who act like the computer and get blocked by GIGO. They are unable to work around this.
barbarian 9y ago
"Super Intelligence" made a nice point that AI will likely be very stupid until it's suddenly very, very smart. I thought the AI worries I read about in the media were a bit hyperbolic till I read this book - it makes some good and sensible arguments as to how a human dangerous AI might come about. It doesn't claim general AI is necessarily very likely, but more that, in the event it does come about, it will be very sudden, very swift - it won't be the gradual curve of innovation and improvement we've been used to elsewhere - and so our time to react to it's birth and implications will be very short.
pryelluw 9y ago
Yes, because even basic AI can be hacked at the hardware level.
rodgerd 9y ago
Actual AI? No. But if I were super-rich I would be, because I'd be concerned an actual AI would look at the problems that the world has and conclude that resource hoarding at the top of the wealth pyramid would be the first problem to solve. The things being sold as AI? Absolutely. Weapons systems, policing systems, and the like, being run by poorly-understood algorithms infected with the unconsidered biases and assumptions of their creators seem like a great way to make things worse, not better.
darylteo 9y ago
My worry with any new technological leap is that we simply forget how to do things that we needed to do prior. When it comes to the AI we have now: * when self driving cars are mainstream, will we forget how to drive? * when ai can answer all questions we need and want will we forget how to think and solve problems? * when robots can print anything and everything, will we forget how to manufacture things with our own hands? What happens to agriculture and food production as well? We don't even need to get to the point where we build weaponised AIs that kill us SkyNet/Horizon style; We might just forget how to do stuff. My theory for the near future is that society will need to form guilds of arts/sports/craftsmanship to preserve the skills that brought us here.
- jacknews 9y ago
  Of course we won't stop thinking, creating, conversing, and there will always be hobbyist and academic interest in even obsolete skills. But we have already forgotten how to make, how to farm, etc. I don't think any single person can make even something as simple as a pencil, from scratch, let alone something like a cell-phone, advanced drugs etc. Even bringing together a collection of all the necessary experts, the task would likely be almost impossible without computers, and the rest of modern infrastructure. If humanity were to be somehow detached from even today's infrastructure, we'd be in trouble.
- oldandtired 9y ago
  This has been a problem for millennia, not just a problem for today. There are so many examples in the last 150 years of knowledge lost because of a change in how we do things. AI's won't be the trouble here. It will be people who no longer care about skills that were once taught. As far as driving is concerned, there are many today who do not or cannot drive due to where they live. Self-driving vehicles will just mean a different form of transport is now available for use. AI's will never be able to answer all questions we need or want. Less so than today. Even with the current computer systems we have, there are so many questions that cannot be answered, we don't know the answers and in many cases we don't even know the questions. It takes a great deal of skill to manufacture anything and though we have machines that can do the duplication, there are many instances where humans have to check that the machines (the robotic systems) have done the correct job. It takes a human judgement to see if everything in acceptable limits. No machine system, as yet, has been given free reign to do an entire process from start to finish without human intervention, as there is far too much that can go wrong. We have such a lack of appreciation of our past that we happily go into the future based on the current or next fad that comes along. As a programmer getting up to the four decades, I see so many things coming along that are touted as the Next Big Thing that are duplicates of previously rejected ideas that failed the test from years and decades ago. We don't learn from the past so we make the same mistakes. We already form guilds and have done so for millennia, not so that we protect the information we have accumulated but to censor and hide that information from those who are not guild members. Being open and sharing our knowledge will do more to preserve it than forming guilds. Actually documenting what we do is extremely important but we tend to get throw out old manuals and documents and source and drawings/blueprints, etc. History is just repeating itself. As humans, we do not value what we can learn form days pasts and often we consider ourself to be more technologically, socially and morally more advanced than those who have lived before us.
williamle8300 9y ago
Not afraid of an hostile take over by AI. Everything is data all the way down so I'm afraid of those that wield that data.
bikamonki 9y ago
I am concerned. People think AI = Skynet but in reality, the next couple of decades, AI = millions of humans out of jobs. I am concerned the change will be too fast for us to adapt. Not only do we need to invent a lot of new jobs, we may need to re-learn basic paradigms like job = goal of life = success. I suppose we can imagine millions of humans that are freed from work to devote their time to hobbies, art, sports, entertainment, etc. What if that does not happen? We plug them to media and keep them calm? Bored jobless youngsters are dangerous, no?
- gremlinsinc 9y ago
  The more I watch this unfold the more I feel we need a new type of party...something more technocratic (uses ai/tech to streamline processes/end bureacracies and make things run more efficiently) and socialize some things..like medicine/basic income/education -- but STILL keep production in the hands of businesses. One alternative to single-payer could be hospitals = insurer. Hospitals would have a max they could charge which would be a % of income for individuals above poverty line and would be required to treat poverty level families...maybe divy them out by hospital as best as they can to spread the costs. hospitals would benefit because they could set the price for themselves....since they only have x amount of recurring income and they know roughly x amount of patients come through door monthly for x, y, z ailments...etc.. They could form coalitions with other hospitals to negotiate drug rates that are more fair...etc... All while ending insurance industry, ending insurance CEO pay, sales agents, customer support jobs--all wasted money in healthcare... I think a MRR system makes more sense to their bottom line. -- traveling? Your host hospital/insurer would pay the out of area hospital.. they'd all set comparative prices that benefit everyone...because they all have the extra costs of making sure the system holds up...
- blacksmythe 9y ago
  Even cardiologist's jobs are not safe: https://news.ycombinator.com/item?id=15007208 https://news.ycombinator.com/item?id=15007208
larkeith 9y ago
True AI, not in the slightest. However, the current implementations of machine learning I believe have the potential for disastrous results, especially as ML is being used for ever more critical roles; We're due for autonomous weapon platforms this year, and self-driving vehicles are soon to follow, while there is limited research on the security of the algorithms, especially in the case of induced edge cases.
oldandtired 9y ago
No. As someone who has watch AI development since the late 70's as both a student, undergraduate and then real world programmer, every hyped-up break-through in AI has led to minor improvements but nothing that takes the programs beyond relative simplicity. Even the systems that beat grandmasters in there respective games are having to operate at gigahertz speeds, focussed on one simple task to beat the human who has to operate in a highly sensory loaded environment. The human has many things to focus on, is running at relatively low speeds (nerve impulses are glacial slow compared to the impulse sequences in silicon hardware) and is still only just being beaten. We have little or no real understanding of how the brain works, what sentience, intelligence and consciousness or free will is. Until we have this kind of understanding of ourselves, this concept that certain people are worried about (the singularity) will not happen. None of our computer systems (or all of them combined) are of the complexity of a single human baby brain. In addition, since I also believe that there is a Creator God who built and designed the universe and that new intelligences are solely in His domain, I have no concern over any such singularity ever being created by man.
- fellellor 9y ago
  What we do have to worry about is the power of hype and marketing and the stupidity of the management class falling prey to that. That stupidity and short sightedness may cause more loss of livelihood than the technology itself.
  - studentrob 9y ago
    Yes. It seems almost inevitable that companies will overhype their versions of the tech in order to get more investment. The more success the tech has, the more expectations the public will have. Musk's rhetoric has come from some relatively minor advancement, in my opinion, compared to what is coming, given that GPU/ASIC hardware is still improving. I think education can help, and maybe we should expect some dips. Perhaps in the next dip, research into the tech will not fall as flat as it did in the 1980s during the last AI winter. Who knows what lies ahead.
thesmallestcat 9y ago
I worry about "good enough" AI winning in the jobs race against humans mostly. You can see it now, like the other day where some non-extremist content was being purged from YouTube. If a human were doing that job it wouldn't have happened, or at least somebody would be accountable instead of "lol a bug." I can see a future where all kinds of interfaces are aggravating and obtuse, like when you have to speak carefully at a phone prompt for the speech recognizer to understand you.
studentrob 9y ago
In the sense that it will kill us? No. In the sense that it may cause a lot of people to retrain? Maybe. Regarding the first version, Musk says this in order to acquire more AI talent, or get it cheaper. Jürgen Schmidhuber does the same thing. What they have in common is they both run organizations that they founded in order to build "AGI". It is odd that they're warning against AGI and building it at the same time. My theory is, if they can convince people that they're doing something "causeworthy" or world-changing, then they can attract more AI talent. In Musk's case, OpenAI also serves as a pipeline, or at least networking location, for getting more in-house talent at Tesla. I think it hurts the field of AI that they say things like this because it gets people's expectations up. However, it also looks like it helps them recruit and network in an area of work that is very, very competitive among companies. The guy who went from Waymo to Uber's self driving car program, for example, had received a $100 million bonus from Google, and Google said this was not out of the ordinary for engineers at that level. Regarding #2, I'm a bit more optimistic than the average SV person. I don't think anyone can tell the future. We just need to focus on making education more accessible. There are no two ways about it, in my view.
Vanit 9y ago
I'm not concerned about AI, but I am concerned about AI being improperly used. In Australia there was a recent controversy with the govt using big data to improperly correlate tax returns with information submitted to centrelink (social security), to determine if people had received welfare when they were ineligible. It turned out that correlating the data was prone to error due to limitations in both datasets, and it was worsened by the high error tolerance, such that they automatically issued debt notices to all false positives and cut their payments with no human oversight. This was made worse by there being no way to appeal the debt to a human. I think you're only going to see this kind of blindfaith incompetence more.
KnightLore 9y ago
I'm not concerned about some kind of consciousness, but before machines were unable to recognize something else than a very specific type of data. Now machines can recognize faces, objects, even emotions. That attached to a program and the existence of "big data" can be even more dangerous, because human is dangerous.
macavity23 9y ago
I actually am, but not about AI becoming smarter than us, or the 'singularity', for good reasons others have already stated. I'm concerned what happens when human-sized combat robots become capable of defeating a trained and motivated human opponent. Such a thing is certainly a way off, but requires no fundamental advances in our understanding (very much unlike skynet/singularity-style general AI) so I see it as a certainty sooner or later, and the way everything is going probably sooner. Throughout human history, the inherent power the rich have over the poor has had an equally-inherent counterpower: there are many more poor people, and if the rich make things too miserable for them, if they all rise up together, they will win. To my knowledge this is a universal in human history (please prove me wrong!), and in general tyrants and elites everywhere have to devote SOME of their resources at least to maintaining the 'general welfare', which is really a nice way of saying 'giving the poor enough to stop them rising up and taking our stuff'. If robots become capable of beating serious human opponents one-on-one, then I think this age-old balance changes, because once you can build one such robot, you can build a hundred thousand, and what are the poor people going to do about that?

13 points by nvahalik 9y ago

Elon Musk says that if you aren't concerned, you should be. If you are concerned, why? If not, then why not?

paulcole 9y ago

Not at all. I'll be long dead before anything happens.

nigrioid 9y ago

No - we have only a very basic understanding of how the brain works, have no idea what consciousness even is, and know little about human intelligence. The only thing I can think of is there's a sinister agenda pushing these constant fear-mongering stories about how we're all going to die from AI. Show me a neural net that even comes close to doing anything resembling original, organized thought and I might take these warnings more seriously.

ljk 9y ago
stupid question, but the AI/neural net these days require training right? look at what happened to microsoft's AI bot that turned rogue after just a few days and had to be turned off
- paulcole 9y ago
  https://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist https://www.theverge.com/2016/3/24/11297050/tay-microsoft-ch... It didn't "go rogue" it did what it was supposed to. People were just being shits to it.
  - ljk 9y ago
    exactly, so "real" AI can be trained by the same people too and have the same outcome?
    - paulcole 9y ago
      somehow i think that "real" AI won't be trained by the minds of 4chan
jostmey 9y ago
We still don't know everything about the biology of bird feathers but we can make flying machines. True AI is not going to happen tomorrow, but I expect it to come within my lifetime. It's easy to get hung up on the complexity of biological systems like the brain, but sometimes all that complexity is not necessary to the function. Complexity is the result of an evolutionary process, and may not be a necessary component of intelligence
- studentrob 9y ago
  Evolution has taken us through an uncountable number of neurons and lifespans. In my opinion, even if we had an AI that matched performance of the human brain, it'd still be missing the historical data on earth (and the universe) we used to arrive at this point, the level of parallelization that exists across the earth across all life forms, and the capacity to input as much data as all life on earth can. Data and input capacity are as important as the model in machine learning. I think we sometimes forget how interdependent we are. We learn a lot through sharing our experiences, and when there are more of us, we make more breakthroughs as a species. And, maybe the answer is 42.
- oldandtired 9y ago
  I started to reply in detail to your comment, but I have decided to stop because there is so much wrong with it. So, I'll leave it with the following, Biological systems are: highly complex self-replicating self-refueling, low-power, low-speed, highly manoeuvrable Our attempts at replication of the abilities lead to brute force high powered, simple non self-replicating, non self-refueling low manoeuvrability needing high speed to achieve same goal And we know very little about biological systems and most of what we know is probably wrong as well.

nvahalik 9y ago

I think thus far the only thing I'm really concerned about is our trust/dependence on it. Articles like this one (https://www.autoblog.com/2017/08/04/self-driving-car-sign-hack-stickers/ https://www.autoblog.com/2017/08/04/self-driving-car-sign-ha...) show just how fragile some of this ML stuff is. Of course, I'm also assuming that ML = AI... that is a correct assumption, right?

larkeith 9y ago
IIRC AI will probably emerge from and incorporate ML, but it's unlikely that AI will be entirely ML-based.

Animats 9y ago

We're already at "machines should think, people should work" for low-level jobs, like fast food and order fulfillment. Robots are still not good at, or cost-effective at, unstructured manipulation, but computers are great at organizing work, communicating, selling, and accounting. What management used to do.

oldandtired 9y ago
Computers are not great at organising anything, cannot communicate, cannot sell and cannot do accounting. It requires a person to "program" these facilities. Without the human generated programs, no computer is capable of doing anything. They are no different to a hammer, a crane or crowbar. They are simply tools in the hands of a human. They are completely useless if not picked up and used appropriately. What is unfortunate is that there are many people who act like the computer and get blocked by GIGO. They are unable to work around this.

barbarian 9y ago

"Super Intelligence" made a nice point that AI will likely be very stupid until it's suddenly very, very smart. I thought the AI worries I read about in the media were a bit hyperbolic till I read this book - it makes some good and sensible arguments as to how a human dangerous AI might come about. It doesn't claim general AI is necessarily very likely, but more that, in the event it does come about, it will be very sudden, very swift - it won't be the gradual curve of innovation and improvement we've been used to elsewhere - and so our time to react to it's birth and implications will be very short.

pryelluw 9y ago

Yes, because even basic AI can be hacked at the hardware level.

rodgerd 9y ago

Actual AI? No. But if I were super-rich I would be, because I'd be concerned an actual AI would look at the problems that the world has and conclude that resource hoarding at the top of the wealth pyramid would be the first problem to solve. The things being sold as AI? Absolutely. Weapons systems, policing systems, and the like, being run by poorly-understood algorithms infected with the unconsidered biases and assumptions of their creators seem like a great way to make things worse, not better.

darylteo 9y ago

My worry with any new technological leap is that we simply forget how to do things that we needed to do prior. When it comes to the AI we have now: * when self driving cars are mainstream, will we forget how to drive? * when ai can answer all questions we need and want will we forget how to think and solve problems? * when robots can print anything and everything, will we forget how to manufacture things with our own hands? What happens to agriculture and food production as well? We don't even need to get to the point where we build weaponised AIs that kill us SkyNet/Horizon style; We might just forget how to do stuff. My theory for the near future is that society will need to form guilds of arts/sports/craftsmanship to preserve the skills that brought us here.

jacknews 9y ago
Of course we won't stop thinking, creating, conversing, and there will always be hobbyist and academic interest in even obsolete skills. But we have already forgotten how to make, how to farm, etc. I don't think any single person can make even something as simple as a pencil, from scratch, let alone something like a cell-phone, advanced drugs etc. Even bringing together a collection of all the necessary experts, the task would likely be almost impossible without computers, and the rest of modern infrastructure. If humanity were to be somehow detached from even today's infrastructure, we'd be in trouble.
oldandtired 9y ago
This has been a problem for millennia, not just a problem for today. There are so many examples in the last 150 years of knowledge lost because of a change in how we do things. AI's won't be the trouble here. It will be people who no longer care about skills that were once taught. As far as driving is concerned, there are many today who do not or cannot drive due to where they live. Self-driving vehicles will just mean a different form of transport is now available for use. AI's will never be able to answer all questions we need or want. Less so than today. Even with the current computer systems we have, there are so many questions that cannot be answered, we don't know the answers and in many cases we don't even know the questions. It takes a great deal of skill to manufacture anything and though we have machines that can do the duplication, there are many instances where humans have to check that the machines (the robotic systems) have done the correct job. It takes a human judgement to see if everything in acceptable limits. No machine system, as yet, has been given free reign to do an entire process from start to finish without human intervention, as there is far too much that can go wrong. We have such a lack of appreciation of our past that we happily go into the future based on the current or next fad that comes along. As a programmer getting up to the four decades, I see so many things coming along that are touted as the Next Big Thing that are duplicates of previously rejected ideas that failed the test from years and decades ago. We don't learn from the past so we make the same mistakes. We already form guilds and have done so for millennia, not so that we protect the information we have accumulated but to censor and hide that information from those who are not guild members. Being open and sharing our knowledge will do more to preserve it than forming guilds. Actually documenting what we do is extremely important but we tend to get throw out old manuals and documents and source and drawings/blueprints, etc. History is just repeating itself. As humans, we do not value what we can learn form days pasts and often we consider ourself to be more technologically, socially and morally more advanced than those who have lived before us.

williamle8300 9y ago

Not afraid of an hostile take over by AI. Everything is data all the way down so I'm afraid of those that wield that data.

bikamonki 9y ago

I am concerned. People think AI = Skynet but in reality, the next couple of decades, AI = millions of humans out of jobs. I am concerned the change will be too fast for us to adapt. Not only do we need to invent a lot of new jobs, we may need to re-learn basic paradigms like job = goal of life = success. I suppose we can imagine millions of humans that are freed from work to devote their time to hobbies, art, sports, entertainment, etc. What if that does not happen? We plug them to media and keep them calm? Bored jobless youngsters are dangerous, no?

gremlinsinc 9y ago
The more I watch this unfold the more I feel we need a new type of party...something more technocratic (uses ai/tech to streamline processes/end bureacracies and make things run more efficiently) and socialize some things..like medicine/basic income/education -- but STILL keep production in the hands of businesses. One alternative to single-payer could be hospitals = insurer. Hospitals would have a max they could charge which would be a % of income for individuals above poverty line and would be required to treat poverty level families...maybe divy them out by hospital as best as they can to spread the costs. hospitals would benefit because they could set the price for themselves....since they only have x amount of recurring income and they know roughly x amount of patients come through door monthly for x, y, z ailments...etc.. They could form coalitions with other hospitals to negotiate drug rates that are more fair...etc... All while ending insurance industry, ending insurance CEO pay, sales agents, customer support jobs--all wasted money in healthcare... I think a MRR system makes more sense to their bottom line. -- traveling? Your host hospital/insurer would pay the out of area hospital.. they'd all set comparative prices that benefit everyone...because they all have the extra costs of making sure the system holds up...
blacksmythe 9y ago
Even cardiologist's jobs are not safe: https://news.ycombinator.com/item?id=15007208 https://news.ycombinator.com/item?id=15007208

larkeith 9y ago

True AI, not in the slightest. However, the current implementations of machine learning I believe have the potential for disastrous results, especially as ML is being used for ever more critical roles; We're due for autonomous weapon platforms this year, and self-driving vehicles are soon to follow, while there is limited research on the security of the algorithms, especially in the case of induced edge cases.

oldandtired 9y ago

No. As someone who has watch AI development since the late 70's as both a student, undergraduate and then real world programmer, every hyped-up break-through in AI has led to minor improvements but nothing that takes the programs beyond relative simplicity. Even the systems that beat grandmasters in there respective games are having to operate at gigahertz speeds, focussed on one simple task to beat the human who has to operate in a highly sensory loaded environment. The human has many things to focus on, is running at relatively low speeds (nerve impulses are glacial slow compared to the impulse sequences in silicon hardware) and is still only just being beaten. We have little or no real understanding of how the brain works, what sentience, intelligence and consciousness or free will is. Until we have this kind of understanding of ourselves, this concept that certain people are worried about (the singularity) will not happen. None of our computer systems (or all of them combined) are of the complexity of a single human baby brain. In addition, since I also believe that there is a Creator God who built and designed the universe and that new intelligences are solely in His domain, I have no concern over any such singularity ever being created by man.

fellellor 9y ago
What we do have to worry about is the power of hype and marketing and the stupidity of the management class falling prey to that. That stupidity and short sightedness may cause more loss of livelihood than the technology itself.
- studentrob 9y ago
  Yes. It seems almost inevitable that companies will overhype their versions of the tech in order to get more investment. The more success the tech has, the more expectations the public will have. Musk's rhetoric has come from some relatively minor advancement, in my opinion, compared to what is coming, given that GPU/ASIC hardware is still improving. I think education can help, and maybe we should expect some dips. Perhaps in the next dip, research into the tech will not fall as flat as it did in the 1980s during the last AI winter. Who knows what lies ahead.

thesmallestcat 9y ago

I worry about "good enough" AI winning in the jobs race against humans mostly. You can see it now, like the other day where some non-extremist content was being purged from YouTube. If a human were doing that job it wouldn't have happened, or at least somebody would be accountable instead of "lol a bug." I can see a future where all kinds of interfaces are aggravating and obtuse, like when you have to speak carefully at a phone prompt for the speech recognizer to understand you.

studentrob 9y ago

In the sense that it will kill us? No. In the sense that it may cause a lot of people to retrain? Maybe. Regarding the first version, Musk says this in order to acquire more AI talent, or get it cheaper. Jürgen Schmidhuber does the same thing. What they have in common is they both run organizations that they founded in order to build "AGI". It is odd that they're warning against AGI and building it at the same time. My theory is, if they can convince people that they're doing something "causeworthy" or world-changing, then they can attract more AI talent. In Musk's case, OpenAI also serves as a pipeline, or at least networking location, for getting more in-house talent at Tesla. I think it hurts the field of AI that they say things like this because it gets people's expectations up. However, it also looks like it helps them recruit and network in an area of work that is very, very competitive among companies. The guy who went from Waymo to Uber's self driving car program, for example, had received a $100 million bonus from Google, and Google said this was not out of the ordinary for engineers at that level. Regarding #2, I'm a bit more optimistic than the average SV person. I don't think anyone can tell the future. We just need to focus on making education more accessible. There are no two ways about it, in my view.

Vanit 9y ago

I'm not concerned about AI, but I am concerned about AI being improperly used. In Australia there was a recent controversy with the govt using big data to improperly correlate tax returns with information submitted to centrelink (social security), to determine if people had received welfare when they were ineligible. It turned out that correlating the data was prone to error due to limitations in both datasets, and it was worsened by the high error tolerance, such that they automatically issued debt notices to all false positives and cut their payments with no human oversight. This was made worse by there being no way to appeal the debt to a human. I think you're only going to see this kind of blindfaith incompetence more.

KnightLore 9y ago

I'm not concerned about some kind of consciousness, but before machines were unable to recognize something else than a very specific type of data. Now machines can recognize faces, objects, even emotions. That attached to a program and the existence of "big data" can be even more dangerous, because human is dangerous.

macavity23 9y ago

I actually am, but not about AI becoming smarter than us, or the 'singularity', for good reasons others have already stated. I'm concerned what happens when human-sized combat robots become capable of defeating a trained and motivated human opponent. Such a thing is certainly a way off, but requires no fundamental advances in our understanding (very much unlike skynet/singularity-style general AI) so I see it as a certainty sooner or later, and the way everything is going probably sooner. Throughout human history, the inherent power the rich have over the poor has had an equally-inherent counterpower: there are many more poor people, and if the rich make things too miserable for them, if they all rise up together, they will win. To my knowledge this is a universal in human history (please prove me wrong!), and in general tyrants and elites everywhere have to devote SOME of their resources at least to maintaining the 'general welfare', which is really a nice way of saying 'giving the poor enough to stop them rising up and taking our stuff'. If robots become capable of beating serious human opponents one-on-one, then I think this age-old balance changes, because once you can build one such robot, you can build a hundred thousand, and what are the poor people going to do about that?