Supposing a deep AI machine were given the Human Question. The most powerful AI machines act empirically through trial and error. Alphazero, December 2017, took on the task of solving chess. It has not as yet completed the task, but has surpassed humans in its chess prowess. The full knowledge of humans including that of the conception of Evolutionary strategies such as Selective Breeding will be available to the machine.
An example of new environmental conditions in a "reaction norm" model of Evolution is the following generation of laws in a country: All public discussion is outlaw. Permitted will be invective of the most violent form. This will satisfy the needs of public criticism and the alteration of the general attitudes of the people. These rules are to be strictly enforced with the penalty for failure of compliance death by killer-droid (*).
*killer-droids are affiliated with no national group or political organization and serve the sole requirements of ultimate reason as embodied by the AI machine.
This is not an ethical suggestion in the ordinary sense, but just as death by natural cause is different in quality than death by human will, the free movement of the AI machine as sanctioned by the democratic community which largely understands it as the best problem solver will allow the experiment. The view will be taken that all deaths are meritorious and rational as they are in the service of the requirements of the most rational attack on the Human Question. Thus, those educated to understand this will feel no excessive resentment, but realize their place in the great effort.
Attempts will not necessarily be total, but different regions of the globe or off-globe can possibly serve as experimental zones.
Deep AI Makes a Try
Re: Deep AI Makes a Try
The answer shall be fully determined exactly by the data and its limits and by the valuesTheVisionofEr wrote: ↑Thu Feb 13, 2020 10:48 pm Supposing a deep AI machine were given the Human Question. The most powerful AI machines act empirically through trial and error. Alphazero, December 2017, took on the task of solving chess. It has not as yet completed the task, but has surpassed humans in its chess prowess. The full knowledge of humans including that of the conception of Evolutionary strategies such as Selective Breeding will be available to the machine.
An example of new environmental conditions in a "reaction norm" model of Evolution is the following generation of laws in a country: All public discussion is outlaw. Permitted will be invective of the most violent form. This will satisfy the needs of public criticism and the alteration of the general attitudes of the people. These rules are to be strictly enforced with the penalty for failure of compliance death by killer-droid (*).
*killer-droids are affiliated with no national group or political organization and serve the sole requirements of ultimate reason as embodied by the AI machine.
This is not an ethical suggestion in the ordinary sense, but just as death by natural cause is different in quality than death by human will, the free movement of the AI machine as sanctioned by the democratic community which largely understands it as the best problem solver will allow the experiment. The view will be taken that all deaths are meritorious and rational as they are in the service of the requirements of the most rational attack on the Human Question. Thus, those educated to understand this will feel no excessive resentment, but realize their place in the great effort.
Attempts will not necessarily be total, but different regions of the globe or off-globe can possibly serve as experimental zones.
and conditions that are already coded into the program.
Nothing new or surprising could ever result, since the presuppositions and prejudices of the programmers would be in the exposition.
-
TheVisionofEr
- Posts: 383
- Joined: Tue Feb 11, 2020 7:59 pm
Re: Deep AI Makes a Try
The play of the Deep Learning machines even in a game of relatively simple rules, chess, regularly produces new and surprising results. Otherwise the machine would not be better than the best human players.
Your objection only seems to work to the extent that the change in things and experimentation with the world does not include a corresponding change in the understanding of the things being manipulated. However, the very criterion,"new and surprising" is not available to the machine, but only to someone observing its activity. Much that is new, not to say wholly impossible to predict, will surely result from AI and to some extent already has.
I think one would have to make a distinction between a very extreme meaning of new, which would not at all include inventions, every invention of technology would not count, every improvement of already existing situations such as stopping death with medical advances would not count. What would count would, indeed, go beyond what we can currently think. And, then, again, I would say, the standard, since it is about our understanding, if closely defined, "new and surprising" would prove self defeating. Since the machine would be programmed already to know what would count as new and surprising and so be capable to achieve that.
Your objection only seems to work to the extent that the change in things and experimentation with the world does not include a corresponding change in the understanding of the things being manipulated. However, the very criterion,"new and surprising" is not available to the machine, but only to someone observing its activity. Much that is new, not to say wholly impossible to predict, will surely result from AI and to some extent already has.
I think one would have to make a distinction between a very extreme meaning of new, which would not at all include inventions, every invention of technology would not count, every improvement of already existing situations such as stopping death with medical advances would not count. What would count would, indeed, go beyond what we can currently think. And, then, again, I would say, the standard, since it is about our understanding, if closely defined, "new and surprising" would prove self defeating. Since the machine would be programmed already to know what would count as new and surprising and so be capable to achieve that.