5 ms·

> Tay is an artificial intelligent chat bot developed by Microsoft's Technology and Research and Bing teams to experiment with and conduct research on conversat

> Tay is an artificial intelligent chat bot developed by Microsoft's Technology and Research and Bing teams to experiment with and conduct research on conversational understanding. The next A.I. winter will be a very cold one. For those wondering what I mean exactly: we're seeing the term A.I. being used in marketing, in the papers, in the news. Yes, we are making great strides in weak A.I. but strong A.I.? The kind we read about in stories? The kind of A.I. the public thinks of when we say A.I.? Asimovian Robotics A.I.? Smoke and mirrors[1]. People develop new techniques and algorithms which are moderately self-learning in a focused way. The general public presumes this to be the basis of a general intelligence which can evolve (magically) to be like another form of life. Soon, everyone jumps on the A.I. bandwagon. The future must just be around the corner! Then the uncomfortable details emerge; strong A.I. is not a matter of faster processors, more memory, or even more advanced/well-designed programming models. Rather, it is that there is still some fundamental aspect of real, human-like, or even animal-like intelligence that, to this day, eludes our understanding of intelligence. A.I. winters have occurred many times before in many countries. The United States in the early 80s, for example, was pulling hair over the cybernization of the Soviet Economy. The US highest levels of government gloomily predicted that massive mainframes, given enough information and processing power, would become self-learning and turn the communist laggard economy into a powerhouse.[2] I think maybe one day A.I. could happen. I think one day I will be proved wrong. Regardless of how A.I. comes about, it will not be due to the label of "A.I." slapped on any kind of product that remotely resembles intelligence. [3] [1]https://en.wikipedia.org/wiki/AI_winter https://en.wikipedia.org/wiki/AI_winter [2]http://nautil.us/issue/23/dominoes/how-the-computer-got-its-revenge-on-the-soviet-union http://nautil.us/issue/23/dominoes/how-the-computer-got-its-... [3] Between the winters, people call their stuff A.I. for the sexiness factor. When called out on the implications of the term, those same people retreat to the textbook definition. "It's A.I!.... well, technically it's weak A.I..."

Delmania 11y ago
> And then the uncomfortable details emerge: that it's not a matter of faster processors, more memory, or even more advanced programming models: that there is still some fundamental aspect of real, human-like, or even animal-like intelligence that eludes us. I believe this was answered in GEB. Hofstadter mentions how humans (and some animals) can step out of a system and analyze it objectively. Meaning, we can take the rules of a system, analyze, and determine that we will never be able to generate the desired results. We can objectively lot at things (even ourselves) and reason about them. A computer, on the other hand, even the most advanced AI, is still just blindly executing the commands given to it.
- schoen 11y ago
  > I believe this was answered in GEB. In my reading of GEB, Hofstadter was criticizing other authors who disagreed with his understanding of the Church-Turing thesis (that human intelligence is, or at least is not more powerful than, some kind of rule-following system). Hofstadter thinks that there is no inherent contradiction or essential difference in kind between the human who "can step out of a system" and the computer "just blindly executing the commands given to it". (But Hofstadter didn't explain at a technical level how to make a computer that's as intelligent as a human being.)
  - sgt101 11y ago
    I think that's the exact opposite of my reading, I think he flagged that Godel had exactly pointed to why humans and CT machines were different.
- sklivvz1971 11y ago
  According to GEB, humans have an internal model of their own mind, ergo a conscience. This is what allows them to step out of the system - they can step out of the system within the model.
  - swombat 11y ago
    I'd say the human model of one's own mind is extremely faulty. Bugs in said self-understanding frequently result in conflicts (as we blame everything but our own actions), wars, etc. Anyone who does any coaching work is intensely aware of these limits (in everyone, btw, the coach included!)...
    - fao_ 11y ago
      Perhaps it could instead be rephrased to 'the ability to build an internal model of their own, and other people's mind'?
    - sklivvz1971 11y ago
      Yep that goes without saying. The model of something will always be a limited, faulty version for the original.
putaside 11y ago
I think this is an unfair assessment. The industry talks of AI, not AGI. It's not Microsoft's fault that the public thinks of Terminator robots when hearing "AI", nor that science-fiction stories have been written a 100 years ago. The general public does not set budgets for AI research. AI research benefits from faster processors, more memory and more advanced algorithms. Only philosophers (the domain of AGI is philosophy, not so much engineering or maths) are uncomfortable with those details. Please refer to AGI, if you are talking of the hypothetical strong AI. The current textbook definition is alright. (And yeah, the hype around AI is palpable and annoying. That's why many researchers called their work "cognitive science", "machine learning", "optimization", "logic" or "applied maths" and avoided "AI". Because else, they'd have to argue semantics, or defend why they haven't build an artificial God yet...)
- romaniv 11y ago
  >It's not Microsoft's fault that the public thinks of Terminator robots when hearing "AI" Large corporation definitely could do better when it comes to explaining the limits of their AI/ML technologies and putting those technologies in perspective (especially historic perspective). Scaling down on hyperbole, buzzwords and personification would help as well.
- TheLogothete 11y ago
  Why is a glorified calculator considered "AI"?
  - davmre 11y ago
    Why shouldn't it be? You don't even need the glorification. A normal calculator is performing what most people would consider to be a difficult cognitive task, and doing so at a superhuman level. AI is not some magical ineffable thing that will someday appear out of nowhere. It's the name we give to the gradual progression of our efforts to build machines to perform cognitive tasks. Someone from the pre-computer age would have had no problem recognizing even the ENIAC as intelligent, in a limited sense; it could solve problems that were previously too hard even for the smartest human mathematicians. As someone who actually does AI research on a daily basis, I have no problem with granting intelligence to even very basic chatbots. That goes hand in hand with recognizing that there are many kinds and levels of intelligence, and plenty of opportunity to build smarter and more flexible systems.
mikeash 11y ago
Seems to me that there's a big difference now, in that even limited as it is, current AI advances are finding many important practical uses. For example, automakers are shipping all sorts of fancy driver assistance systems, and a lot of that is driven by fancy image recognition and deep learning systems. Looking beyond, there's the constantly-discussed self-driving car. It's not here yet, and it's going to take a ton of work, but at this point it looks like there's nothing fundamental standing in the way. It's no longer a question of whether it will happen, but whether it's happening in five, ten, or fifteen years. Outside of transportation, businesses are applying AI techniques to analytics to squeeze ever more money out of their client base. There's the infamous story about how Target knew a teenage girl was pregnant before her father did. AT&T just sent me a letter saying they're giving me an extra 5GB/month bonus data as long as I keep my plan, which I assume happened because their data indicated I was a risk for downgrading or switching carriers. Improving these systems is worth a ton of money, and as far as I understand it fits really well with where current research advancements are being made. As long as there are clear, immediate, profitable uses for AI advancements, there shouldn't be any worry of another winter. I guess the question is whether current advancements will peter out eventually, or whether we've reached a point where more research will always produce more immediate dividends. It sure feels like the latter to me, but I could easily be wrong.
- fao_ 11y ago
  > There's the infamous story about how Target knew a teenage girl was pregnant before her father did. > AT&T just sent me a letter saying they're giving me an extra 5GB/month bonus data as long as I keep my plan, which I assume happened because their data indicated I was a risk for downgrading or switching carriers. This isn't Artificial Intelligence though. These seem like they fall under 'Reactive systems' or 'Data Analytics'.
  - mikeash 11y ago
    Throughout the history of AI, anything that's developed to the point where it actually works ceases to be "AI." In any case, it doesn't matter what you call it. Current research is producing results that can improve these systems, which means that research is worth a lot of money.

Delmania 11y ago

> And then the uncomfortable details emerge: that it's not a matter of faster processors, more memory, or even more advanced programming models: that there is still some fundamental aspect of real, human-like, or even animal-like intelligence that eludes us. I believe this was answered in GEB. Hofstadter mentions how humans (and some animals) can step out of a system and analyze it objectively. Meaning, we can take the rules of a system, analyze, and determine that we will never be able to generate the desired results. We can objectively lot at things (even ourselves) and reason about them. A computer, on the other hand, even the most advanced AI, is still just blindly executing the commands given to it.

schoen 11y ago
> I believe this was answered in GEB. In my reading of GEB, Hofstadter was criticizing other authors who disagreed with his understanding of the Church-Turing thesis (that human intelligence is, or at least is not more powerful than, some kind of rule-following system). Hofstadter thinks that there is no inherent contradiction or essential difference in kind between the human who "can step out of a system" and the computer "just blindly executing the commands given to it". (But Hofstadter didn't explain at a technical level how to make a computer that's as intelligent as a human being.)
- sgt101 11y ago
  I think that's the exact opposite of my reading, I think he flagged that Godel had exactly pointed to why humans and CT machines were different.
sklivvz1971 11y ago
According to GEB, humans have an internal model of their own mind, ergo a conscience. This is what allows them to step out of the system - they can step out of the system within the model.
- swombat 11y ago
  I'd say the human model of one's own mind is extremely faulty. Bugs in said self-understanding frequently result in conflicts (as we blame everything but our own actions), wars, etc. Anyone who does any coaching work is intensely aware of these limits (in everyone, btw, the coach included!)...
  - fao_ 11y ago
    Perhaps it could instead be rephrased to 'the ability to build an internal model of their own, and other people's mind'?
  - sklivvz1971 11y ago
    Yep that goes without saying. The model of something will always be a limited, faulty version for the original.

putaside 11y ago

I think this is an unfair assessment. The industry talks of AI, not AGI. It's not Microsoft's fault that the public thinks of Terminator robots when hearing "AI", nor that science-fiction stories have been written a 100 years ago. The general public does not set budgets for AI research. AI research benefits from faster processors, more memory and more advanced algorithms. Only philosophers (the domain of AGI is philosophy, not so much engineering or maths) are uncomfortable with those details. Please refer to AGI, if you are talking of the hypothetical strong AI. The current textbook definition is alright. (And yeah, the hype around AI is palpable and annoying. That's why many researchers called their work "cognitive science", "machine learning", "optimization", "logic" or "applied maths" and avoided "AI". Because else, they'd have to argue semantics, or defend why they haven't build an artificial God yet...)

romaniv 11y ago
>It's not Microsoft's fault that the public thinks of Terminator robots when hearing "AI" Large corporation definitely could do better when it comes to explaining the limits of their AI/ML technologies and putting those technologies in perspective (especially historic perspective). Scaling down on hyperbole, buzzwords and personification would help as well.
TheLogothete 11y ago
Why is a glorified calculator considered "AI"?
- davmre 11y ago
  Why shouldn't it be? You don't even need the glorification. A normal calculator is performing what most people would consider to be a difficult cognitive task, and doing so at a superhuman level. AI is not some magical ineffable thing that will someday appear out of nowhere. It's the name we give to the gradual progression of our efforts to build machines to perform cognitive tasks. Someone from the pre-computer age would have had no problem recognizing even the ENIAC as intelligent, in a limited sense; it could solve problems that were previously too hard even for the smartest human mathematicians. As someone who actually does AI research on a daily basis, I have no problem with granting intelligence to even very basic chatbots. That goes hand in hand with recognizing that there are many kinds and levels of intelligence, and plenty of opportunity to build smarter and more flexible systems.

mikeash 11y ago

Seems to me that there's a big difference now, in that even limited as it is, current AI advances are finding many important practical uses. For example, automakers are shipping all sorts of fancy driver assistance systems, and a lot of that is driven by fancy image recognition and deep learning systems. Looking beyond, there's the constantly-discussed self-driving car. It's not here yet, and it's going to take a ton of work, but at this point it looks like there's nothing fundamental standing in the way. It's no longer a question of whether it will happen, but whether it's happening in five, ten, or fifteen years. Outside of transportation, businesses are applying AI techniques to analytics to squeeze ever more money out of their client base. There's the infamous story about how Target knew a teenage girl was pregnant before her father did. AT&T just sent me a letter saying they're giving me an extra 5GB/month bonus data as long as I keep my plan, which I assume happened because their data indicated I was a risk for downgrading or switching carriers. Improving these systems is worth a ton of money, and as far as I understand it fits really well with where current research advancements are being made. As long as there are clear, immediate, profitable uses for AI advancements, there shouldn't be any worry of another winter. I guess the question is whether current advancements will peter out eventually, or whether we've reached a point where more research will always produce more immediate dividends. It sure feels like the latter to me, but I could easily be wrong.

fao_ 11y ago
> There's the infamous story about how Target knew a teenage girl was pregnant before her father did. > AT&T just sent me a letter saying they're giving me an extra 5GB/month bonus data as long as I keep my plan, which I assume happened because their data indicated I was a risk for downgrading or switching carriers. This isn't Artificial Intelligence though. These seem like they fall under 'Reactive systems' or 'Data Analytics'.
- mikeash 11y ago
  Throughout the history of AI, anything that's developed to the point where it actually works ceases to be "AI." In any case, it doesn't matter what you call it. Current research is producing results that can improve these systems, which means that research is worth a lot of money.