On Fri, Dec 8, 2017 at 7:30 PM, Brent Meeker <[email protected]> wrote: > > > On 12/8/2017 2:09 AM, Telmo Menezes wrote: >> >> On Thu, Dec 7, 2017 at 4:08 PM, Bruno Marchal <[email protected]> wrote: >>> >>> On 07 Dec 2017, at 10:01, Telmo Menezes wrote: >>> >>>> On Wed, Dec 6, 2017 at 11:50 PM, Brent Meeker <[email protected]> >>>> wrote: >>>>> >>>>> >>>>> >>>>> On 12/6/2017 1:46 AM, Bruno Marchal wrote: >>>>>> >>>>>> >>>>>> I suspect that this is perhaps why Brent want to refer to the >>>>>> environment >>>>>> for relating consciousness to the machine, and in Artificial >>>>>> Intelligence, >>>>>> some people defend the idea that (mundane) consciousness occur only >>>>>> when >>>>>> the >>>>>> environment contradicts a little bit the quasi automatic persistent >>>>>> inference we do all the time. >>>>> >>>>> >>>>> >>>>> That's Jeff Hawkins model of consciousness: one becomes conscious of >>>>> something when all lower, more specialized levels of the brain have >>>>> found >>>>> it >>>>> not to match their predictions. >>>> >>>> >>>> In that sort of model, how does matter "know" that it is being used to >>>> run a forecasting algorithm? Surely it doesn't right? >>> >>> >>> Imagine a society which builds some objects. When everything go well, the >>> boss can sleep in his office. But then there is some accident or >>> something >>> unusual. That is the time to wake up the boss. In this analogy, >>> consciousness is played by the (incorrigible) boss. >>> >>> >>> >>>> The only way this could work is if the forecasting algorithm and the >>>> cascading effects of failing predictions have the side effect of >>>> creating the "right" sort of interactions at a lower level that >>>> trigger consciousness. >>> >>> >>> After a moment of panic, the sub-entities dare to awake the ultimate >>> judge: >>> the one capable of "going out of the box" to take a (perhaps risky) >>> decision >>> in absence of complete information, and to take on its shoulder the >>> responsibility. >>> >>> >>> >>>> Then I want to know what these interactions >>>> are, and what if the "atom" of consciousness, what is the first >>>> principle. Without this, I would say that such hypothesis are not even >>>> wrong. >>> >>> >>> The sub-unities have specialized task, and does not need evolved >>> forecasting >>> ability. You can think them as ants, when they do their usual jobs >>> triggered >>> by the local pheromones left by their close neighbors. But if the nest is >>> attacked, or if some important food is missing, some species will needs >>> some >>> order of the queen (ike to fight or to move away. Some societies can >>> delegate most of the power to the sub-unities, but in complex unknown >>> situation, if they have to make important decision, they will need a >>> centralization of the power, which can act much more quickly to convince >>> the >>> whole society of some unusual option, like running away, closing the >>> doors, >>> fighting the enemy, etc. That will happen when *many* ants complain on >>> something. >>> >>> In this case, the role of consciousness is focusing the attention on what >>> is >>> important (with respect to survival), and to speed-up planning, decision, >>> etc. >>> >>> I am not sure this answer the question (we are in the "easy" part of the >>> problem here). >>> >>> But you will help me by telling me what is missing. I am not sure we need >>> to >>> dig on the difficult part of the consciousness problem here, which is >>> handed >>> at a different level, and concerned with the fact that the boss/queen is >>> confined in his office/chamber and can never be sure if the ants panic is >>> genuine, or an illusion, and still decide ... >> >> Yes, I agree with this model and what you say. I am just criticizing >> the "trick" of confusing the several meanings of consciousness. >> I would say that here we are in the realm of intelligence / learning. >> This is about attention, and how attention is directed. Several AI >> models already work like this. When an artificial neural network fails >> a prediction, this triggers a cascade of changes. It wakes up the >> boss, as you say. >> >> In short, I feel that some scientists tend to propose an answer to the >> easy problem and that try to smuggle it as a solution for the hard >> problem, by relying on the overloading of terms. > > > Progress is made by solving the problems you can.
Sure. And it is also true that progress is not made by pretending to have solved problems that were not solved. > But as you know I think > "the hard problem" will go away when the "easy problem" is solved. When we > can produce AI's that are creative, humorous, compassionate, imaginative, > etc and adjust those attributes and understand how they are > implemented...the "hard problem" will be seen as the wrong question. > > Instead AI engineers will ask, "Well, how much consciousness do you want? > We recommend more subconscious competence for that task," I guess you also know my position about this. You are taking an a priori position of what consciousness is, and how it fits reality and just running with it. It is irrelevant here if the human-level AI already exists or not. Knowing how to make something happen is not the same thing of understanding how it works. This is still true with engineering. Take the humble bike: it is a very old (and elegant) technology. We certainly know how to build them. However, the precise dynamics that make them work are still an open research problem. An article was published on the in Science in 2011! http://science.sciencemag.org/content/332/6027/339 With AI the situation is even trickier. I'm sure you follow modern AI research, and realize that modern AI is "grown", more than it is designed. Heuristic search, evolutionary approaches, statistical learning, these are all self-adaption mechanisms that look more like an artificial biology than XIX-century style engineering. As with the bikes, the reason why the artificial neural networks *that we already have* are so effective is also an open research problem. Humans have been planting seeds, watering and fertilizing them to grow plants. They did this for a very long time, while being completely ignorant of genetics, cell biology, photosynthesis and so forth. > Already there seems to be a consensus that a philosophical zombie is > impossible. That entails that any AI with human level (or greater) > intelligence must be conscious. That is also my bet, but it is no more than a bet at this point. Consensus have been spectacularly wrong many times. None of this replaces testable theories. I don't see a lot of them. > The AI engineers will develop different > realizations of intelligent machines and invent terms for the different ways > in which they are conscious. Then "consciousness" will be seen as a vague > generalization covering many somewhat different processes. Probably they will make bets, yes. Very tricky ethical problems will arise. The bet that philosophical zombies are impossible does not replace a scientific theory, but it is enough to have ethical consequences. I think there we can agree. Telmo. > Brent > "One cannot guess the real difficulties of a problem before having solved > it." > --- Carl Ludwig Siegel > > > -- > You received this message because you are subscribed to the Google Groups > "Everything List" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at https://groups.google.com/group/everything-list. > For more options, visit https://groups.google.com/d/optout. -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at https://groups.google.com/group/everything-list. For more options, visit https://groups.google.com/d/optout.
