Writing with a personal experience is actually a good way to grow. People normally follow individuals not outlets, so it makes the personal relationship stronger.
Same! There is a 0% chance I would have been able to write all the stuff I've wanted to hear if I didn't own the publication. No one with an editor would have been OK with this... for better or worse!
I am not sure what you mean by mathematics is a “ground truth”. My first reaction is to say, no. Turing (and others) realized the insufficiency of mathematics, as described in Erik J. Larson’s book, The Myth of Artificial Intelligence.
What I mean is that mathematics is the best way we have of modeling reality with high fidelity. As to whether the universe literally *is* mathematical, I am undecided, but I suspect that it is not, and so in this sense I agree with your point. As I wrote:
"And that’s unsatisfying in a sense, because ultimately mathematics can merely be used to represent these inscrutable emergent orders with greater and greater fidelity (also known as modeling reality)."
Interesting post! I found almost all of it reasonable but I'm skeptical about this part:
"You’ll talk to an orchestrating agent, and other agents will be deputized on the fly to do research, write code, etc. At first it will be just a few agents, but there is no reason in principle that this parallelization cannot continue indefinitely. It will be possible to ask a question, and, in essence, create a company of digital minds exclusively for the purpose of solving your specific task. The company will exist as long as it needs to: a day, a week, indefinitely."
Most computing systems today spend a lot of time idling, waiting for something to happen in the external world. My iPhone sits in my pocket waiting for me to pull it out and use it. Web servers sit around waiting for inbound requests. Compute is plentiful enough that in most cases it makes sense to have a lot of spare capacity, which allows these systems to respond quickly even in periods of heavy demand.
I expect that the same true will be true for AI: that AI systems will get fast and powerful enough that they can perform almost any task within a matter of seconds or at most minutes. You'll say "plan me an itinerary for my trip to San Francisco next month" and in 3 or 10 or 30 seconds it will come back with a possible itinerary—or maybe a menu of 3 or 4 possible ones that the user gets to choose from. Under the hood of course your AI travel assistant might consist of dozens of specialized AI modules designed to perform various aspects of the travel-planning process. But from our perspective, they won't be an organization of autonomous digital agents, it'll just be an app or a chatbot or whatever that has near-instant answers to most of the questions we ask it.
Maybe this is mostly a matter of semantics, but I think it might be misleading to think of it as a "company of digital minds." I think it'll be more like existing software (say the Google search engine, the Uber app, etc.) than a lot of people think. Just as with current software, the bottleneck will be interaction with the physical world—often getting feedback from the user—and so there will be no particular reason to take human beings out of the loop for most tasks.
I'm still thinking through my own views on what this is likely to look like so I might be thinking about this wrong. But that's what make sense to me right now.
I do think that's true for a lot of tasks that seem simple to us but are in fact complex (booking a flight, planning a trip, etc.). It may be multiple agents under the hood, but it will just look like a chatbot or Siri or similar looks today.
I think this will get more complicated, though. For instance, if I say "write me an academic-level report analyzing the Small Business Investment Corporations created by Congress in 1958, looking specifically at what those corporations invested in and how they performed relative to standard market benchmarks," well to actually do that at a very high quality level is going to involve more than Googling, a la perplexity.
It may involve writing some code to scrape some data, or to do some statistics. It might involve contacting people who hold access to certain data. It would involve reading a lot of different sources.
As the eventual author of this piece, or someone using this research as an input into something else I'm doing, I may well need to know the specific actions taken by the various agents, so I think there will need to be some UI to at least give me the option of auditing arbitrary actions taken by the agent(s).
But again, this is pie in the sky for now--let's get them booking Ubers first!
Writing with a personal experience is actually a good way to grow. People normally follow individuals not outlets, so it makes the personal relationship stronger.
I for one try to include bursts of it when I can.
Same! There is a 0% chance I would have been able to write all the stuff I've wanted to hear if I didn't own the publication. No one with an editor would have been OK with this... for better or worse!
I am not sure what you mean by mathematics is a “ground truth”. My first reaction is to say, no. Turing (and others) realized the insufficiency of mathematics, as described in Erik J. Larson’s book, The Myth of Artificial Intelligence.
What I mean is that mathematics is the best way we have of modeling reality with high fidelity. As to whether the universe literally *is* mathematical, I am undecided, but I suspect that it is not, and so in this sense I agree with your point. As I wrote:
"And that’s unsatisfying in a sense, because ultimately mathematics can merely be used to represent these inscrutable emergent orders with greater and greater fidelity (also known as modeling reality)."
Interesting post! I found almost all of it reasonable but I'm skeptical about this part:
"You’ll talk to an orchestrating agent, and other agents will be deputized on the fly to do research, write code, etc. At first it will be just a few agents, but there is no reason in principle that this parallelization cannot continue indefinitely. It will be possible to ask a question, and, in essence, create a company of digital minds exclusively for the purpose of solving your specific task. The company will exist as long as it needs to: a day, a week, indefinitely."
Most computing systems today spend a lot of time idling, waiting for something to happen in the external world. My iPhone sits in my pocket waiting for me to pull it out and use it. Web servers sit around waiting for inbound requests. Compute is plentiful enough that in most cases it makes sense to have a lot of spare capacity, which allows these systems to respond quickly even in periods of heavy demand.
I expect that the same true will be true for AI: that AI systems will get fast and powerful enough that they can perform almost any task within a matter of seconds or at most minutes. You'll say "plan me an itinerary for my trip to San Francisco next month" and in 3 or 10 or 30 seconds it will come back with a possible itinerary—or maybe a menu of 3 or 4 possible ones that the user gets to choose from. Under the hood of course your AI travel assistant might consist of dozens of specialized AI modules designed to perform various aspects of the travel-planning process. But from our perspective, they won't be an organization of autonomous digital agents, it'll just be an app or a chatbot or whatever that has near-instant answers to most of the questions we ask it.
Maybe this is mostly a matter of semantics, but I think it might be misleading to think of it as a "company of digital minds." I think it'll be more like existing software (say the Google search engine, the Uber app, etc.) than a lot of people think. Just as with current software, the bottleneck will be interaction with the physical world—often getting feedback from the user—and so there will be no particular reason to take human beings out of the loop for most tasks.
I'm still thinking through my own views on what this is likely to look like so I might be thinking about this wrong. But that's what make sense to me right now.
I do think that's true for a lot of tasks that seem simple to us but are in fact complex (booking a flight, planning a trip, etc.). It may be multiple agents under the hood, but it will just look like a chatbot or Siri or similar looks today.
I think this will get more complicated, though. For instance, if I say "write me an academic-level report analyzing the Small Business Investment Corporations created by Congress in 1958, looking specifically at what those corporations invested in and how they performed relative to standard market benchmarks," well to actually do that at a very high quality level is going to involve more than Googling, a la perplexity.
It may involve writing some code to scrape some data, or to do some statistics. It might involve contacting people who hold access to certain data. It would involve reading a lot of different sources.
As the eventual author of this piece, or someone using this research as an input into something else I'm doing, I may well need to know the specific actions taken by the various agents, so I think there will need to be some UI to at least give me the option of auditing arbitrary actions taken by the agent(s).
But again, this is pie in the sky for now--let's get them booking Ubers first!