This depends on whether the problem is the basic complexity of modeling a neural network or learning how to do it. If the former, then we may be looking at a long time. But if it’s the latter, then we really just need more attempts, successful or not, to learn from and a framework which allows a leap in understanding could arrive.
But if it’s the latter, then we really just need more attempts,
I don’t know that repeatedly doing the wrong thing will help inform us how to do the right thing. This seems counterfactual to me. Certainly it informs us what the wrong thing is, but… without additional effort to more finely emulate the real-time biochemical actions of neurons, it seems that emulating what we already know won’t lead us to deeper insights as to what we don’t. The question becomes: how do we discern that missing information?
Emulations are certainly a vital part of that process, however: without them we cannot properly guage how close we are to ‘knowing enough for government work’.
Everything that fails does for a reason and in a way. In engineering, mere bugs aside, everything fails at the frontier of our knowledge and our failures carry information about the shape of that frontier back to us. We learn what problems need to be overcome and can, with many failures, generalize what the overall frontier is like, connect its problems and create concepts which solve many at once.
Everything that fails does for a reason and in a way.
Oh, absolutely. But if they keep failing for the same reason and in the same way, re-running the simulations doesn’t get you any unique or novel information. It only reinforces what you already know.
I acknowledged this as I said, “Emulations are certainly a vital part of that process, however: without them we cannot properly guage how close we are to ‘knowing enough for government work’.”
I think the problem here is that you think that each instance of a simulation is actually an “attempt”. A simulation is a model of some behavior; unlike climbing Everest (which I did in 2003), taming Pegasus (in −642) or repelling the Golden Horde (1257 − 1324, when I was called away on urgent business in Stockholm), each run of a model is a trial, not an attempt. Each iteration of the model is an attempt, as is each new model.
We need more attempts. We learn something different from each one.
I think the problem here is that you think that each instance of a simulation is actually an “attempt”.
No, the problem here is more that I don’t believe that it is any longer feasible to run a simulation and attempt to extract new information without direct observation of the simulated subject-matter.
We need more attempts. We learn something different from each one.
Yes, absolutely. But I don’t believe we can do anything other than repeat the past by building models based on modeled output without direct observation at this time.
So why not just say “to clarify, I believe that we do not have enough knowledge of C. elegans’ neuroanatomy to build new models at this time. We need to devote more work to studying that before we can build newer models”? That’s a perfectly valid objection, but it contradicts your original post, which states that C. elegans is well understood neurologically.
If you believe that we cannot build effective models “without [additional] direct observation”, then you have done two things: you’ve objected to the consensus that C. elegans is well understood and provided a criterion (and effective upload model of its neuroanatomy) for judging how well we understand.
That’s a perfectly valid objection, but it contradicts your original post, which states that C. elegans is well understood neurologically.
My original post stated, “without additional effort to more finely emulate the real-time biochemical actions of neurons, it seems that emulating what we already know won’t lead us to deeper insights as to what we don’t.”
Your assertion (in-line quoted, this comment) is false. I said what I meant the first time ’round: we don’t know enough about how neurons work yet and without that understanding any models we build now won’t yield us any new insights into how they do.
This, furthermore, has nothing to do with C. elegans in specific.
you’ve objected to the consensus that C. elegans is well understood and provided a criterion (and effective upload model of its neuroanatomy) for judging how well we understand.
Since the goal of these models is to emulate the behavior of C. elegans, and the models do not yet do this, it is clear that one of two things is true: either we do not understand C. elegans or we do not understand neurobiology sufficiently to achieve this goal.
I have made my assertion as to which this is, I have done so quite explicitly, and I have been consistent and clear in this from my first post in this thread.
I am not jkaufman. So I don’t know that I follow what you’re trying to say here. This means that either you or I are confused. In either case, no successful communication is currently occurring.
Could you please clarify what it is you’re trying to say?
Nothing to clarify, actually. I apologize; I’ve been busy and the header switch occasioned by using the context link threw me. It changes the title to “XXXX comments on YYYY”. Not being someone who comments consistently, this tends to make me mistake who originally posted because it plants an association between the person I’m replying to and the title of the post.
Maybe. I read a massive quantity of material daily, on the order of 80-90,000 words some weeks. This is combined with comment across a variety of forums and fields. I rely heavily on cues from websites to keep straight who I’m talking to and that I’m even on the right submission forms when I say something.
This depends on whether the problem is the basic complexity of modeling a neural network or learning how to do it. If the former, then we may be looking at a long time. But if it’s the latter, then we really just need more attempts, successful or not, to learn from and a framework which allows a leap in understanding could arrive.
I don’t know that repeatedly doing the wrong thing will help inform us how to do the right thing. This seems counterfactual to me. Certainly it informs us what the wrong thing is, but… without additional effort to more finely emulate the real-time biochemical actions of neurons, it seems that emulating what we already know won’t lead us to deeper insights as to what we don’t. The question becomes: how do we discern that missing information?
Emulations are certainly a vital part of that process, however: without them we cannot properly guage how close we are to ‘knowing enough for government work’.
Everything that fails does for a reason and in a way. In engineering, mere bugs aside, everything fails at the frontier of our knowledge and our failures carry information about the shape of that frontier back to us. We learn what problems need to be overcome and can, with many failures, generalize what the overall frontier is like, connect its problems and create concepts which solve many at once.
Oh, absolutely. But if they keep failing for the same reason and in the same way, re-running the simulations doesn’t get you any unique or novel information. It only reinforces what you already know.
I acknowledged this as I said, “Emulations are certainly a vital part of that process, however: without them we cannot properly guage how close we are to ‘knowing enough for government work’.”
I think the problem here is that you think that each instance of a simulation is actually an “attempt”. A simulation is a model of some behavior; unlike climbing Everest (which I did in 2003), taming Pegasus (in −642) or repelling the Golden Horde (1257 − 1324, when I was called away on urgent business in Stockholm), each run of a model is a trial, not an attempt. Each iteration of the model is an attempt, as is each new model.
We need more attempts. We learn something different from each one.
No, the problem here is more that I don’t believe that it is any longer feasible to run a simulation and attempt to extract new information without direct observation of the simulated subject-matter.
Yes, absolutely. But I don’t believe we can do anything other than repeat the past by building models based on modeled output without direct observation at this time.
So why not just say “to clarify, I believe that we do not have enough knowledge of C. elegans’ neuroanatomy to build new models at this time. We need to devote more work to studying that before we can build newer models”? That’s a perfectly valid objection, but it contradicts your original post, which states that C. elegans is well understood neurologically.
If you believe that we cannot build effective models “without [additional] direct observation”, then you have done two things: you’ve objected to the consensus that C. elegans is well understood and provided a criterion (and effective upload model of its neuroanatomy) for judging how well we understand.
My original post stated, “without additional effort to more finely emulate the real-time biochemical actions of neurons, it seems that emulating what we already know won’t lead us to deeper insights as to what we don’t.”
Your assertion (in-line quoted, this comment) is false. I said what I meant the first time ’round: we don’t know enough about how neurons work yet and without that understanding any models we build now won’t yield us any new insights into how they do.
This, furthermore, has nothing to do with C. elegans in specific.
Since the goal of these models is to emulate the behavior of C. elegans, and the models do not yet do this, it is clear that one of two things is true: either we do not understand C. elegans or we do not understand neurobiology sufficiently to achieve this goal.
I have made my assertion as to which this is, I have done so quite explicitly, and I have been consistent and clear in this from my first post in this thread.
So where’s the confusion?
“The first time around” for the OPer is the OP, from which it is absent and in which you identify the problem as incomplete attempts.
I am not jkaufman. So I don’t know that I follow what you’re trying to say here. This means that either you or I are confused. In either case, no successful communication is currently occurring.
Could you please clarify what it is you’re trying to say?
Nothing to clarify, actually. I apologize; I’ve been busy and the header switch occasioned by using the context link threw me. It changes the title to “XXXX comments on YYYY”. Not being someone who comments consistently, this tends to make me mistake who originally posted because it plants an association between the person I’m replying to and the title of the post.
Ahh. Much is explained. :)
Well, hopefully this incident will serve to reinforce this particular tidbit and prevent you from having a repeat occurrance.
Maybe. I read a massive quantity of material daily, on the order of 80-90,000 words some weeks. This is combined with comment across a variety of forums and fields. I rely heavily on cues from websites to keep straight who I’m talking to and that I’m even on the right submission forms when I say something.