Global Solutions vs. Local Solutions for the AI Safety Problem

Round  1

Reviewer 1 Report

1. Please enounce the presuppositions / assumptions of the paper.

2. Please explicitly announce, in the text, the main thesis of the paper.

3. Please shortly define before/in Introduction: intelligence, human intelligence, artificial intelligence and general artificial intelligence and explain the differences between AI/AGI, human intelligence and AI, why/if AI is or not intelligence, in the same sense, as for the human beings.

4. Please clarify the following concepts: human level intelligence (lines 60-61); human analogue (line 66); benevolent AI (line 91); AI/AGI’s commercial purposes

5. Is the theoretical interest less valuable than a practical one? (line 99). Please explain.

6. AI can be used or will act autonomously? (lines 22 sqq). Please explain.

7. Sending message to future (line 512) – what for? A superior intelligence receiving a message from an inferior one will evaluate the request accordingly to another scale of understanding.  Please explain.

8. What is a seed AI (line 680)? Please explain.

9. It is not usual, in a scientific paper, to introduce a reference that is a manuscript, as the  reader cannot check that source. Please correct this anomaly.  (line 518; reference no 68)

10. Humans, so diverse, so different, so rich in various and contradictory characteristics will be incorporated into a much simpler AI/AGI? Please explain.

11. Lines 756-758  Seems needed to be corrected, by introducing “of the” before ALLFED

12.  References 41, 42, 48 should have some links  Please introduce links to all the online available references!

13. Please explain where is the Big Data theme in the AI/AGI in this paper.

Generally, assume the presuppositions of your paper explicitly, better define your concepts, improve the argumentation, add all the data missing in the references

Author Response

Reviewer 1 

Please enounce the presuppositions / assumptions of the paper.
Reply: We have done this.

Please explicitly announce, in the text, the main thesis of the paper.
Reply: We have done this.

Please shortly define before/in Introduction: intelligence, human intelligence, artificial intelligence and general artificial intelligence and explain the differences between AI/AGI, human intelligence and AI, why/if AI is or not intelligence, in the same sense, as for the human beings.
Reply: We have done this.

Please clarify the following concepts: human level intelligence (lines 60-61); human analogue(line 66); benevolent AI (line 91); AI/AGI’s commercial purposes - replaced with mind-models.
Reply: We have done this.

Is the theoretical interest less valuable than a practical one? (line 99). Please explain.
Reply: Removed and replaced with explanation about Luddism

6. AI can be used or will act autonomously? (lines 22 sqq). Please explain.
Reply: - We have now explained.

7. Sending message to future (line 512) – what for? A superior intelligence receiving a message from an inferior one will evaluate the request accordingly to another scale of understanding.  Please explain.
Reply:  We have removed this.

8. What is a seed AI (line 680)? Please explain.
Reply: We have removed this.

9. It is not usual, in a scientific paper, to introduce a reference that is a manuscript, as the reader cannot check that source. Please correct this anomaly.  (line 518; reference no 68)
Reply: We have removed this.

10. Humans, so diverse, so different, so rich in various and contradictory characteristics will be incorporated into a much simpler AI/AGI? Please explain. Reply: We haveexplained as participation in human-AI teams.

11. Lines 756-758  Seems needed to be corrected, by introducing “of the” before ALLFED
Reply: We have added this.

12.  References 41, 42, 48 should have some links  Please introduce links to all the online available references!
Reply: We added the links.

13. Please explain where is the Big Data theme in the AI/AGI in this paper.  
Reply: We added a clarification in the Introduction.

Generally, assume the presuppositions of your paper explicitly, better define your concepts, improve the argumentation, add all the data missing in the references
Thank you for the feedback – it has made this a stronger paper.

Reviewer 2 Report

The text is somewhat chaotic and messy, without a clear sound structure. I would suggest the authors to create chapters with subchapters and discuss the contents in the introduction. Moreover, the text is hard to classify as an academic paper. In some of the sub-chapters it resembles a policy paper with additional recommendations. Moreover, the text mainly surveys other academic works. Thus, a strong innovative component is lacking. Could the authors provide any original perspective on the issue or discuss some of the solutions in more detail, taking more critical perspective?

Author Response

Reviewer 2

The text is somewhat chaotic and messy, without a clear sound structure. I would suggest the authors to create chapters with subchapters and discuss the contents in the introduction. Moreover, the text is hard to classify as an academic paper. In some of the sub-chapters it resembles a policy paper with additional recommendations. Moreover, the text mainly surveys other academic works. Thus, a strong innovative component is lacking. Could the authors provide any original perspective on the issue or discuss some of the solutions in more detail, taking more critical perspective?

Reply: - We have added the perspective and discussion to the Introduction and into the new final Section. 

Reviewer 3 Report

The papers offers an intelligent discussion of an interesting and relevant subject.  I’m inclined to recommend acceptance on that basis alone but there is quite a lot that worries me …

Firstly, we have to question the actual ‘contribution’ of the paper.  The title suggests a ‘classification’, the abstract, a ‘discussion’.  I’m not sure the former can be justified: the top-level 1,2,3,4 doesn’t take much thinking about and the presentation in the paper (some tables, some bulleted list, some text) is a bit too haphazard.  The latter is better but the discussion needs a little more focus – or at least some stronger conclusions beyond the obvious, ‘tricky, this, isn’t it?’  What will the major drivers be that determine which of these outcomes we end up with?

This leads to the second major point: a number of assumptions on which the paper seems to be based.  Aside from a slight (but evident) Western or first world perspective, the paper seems to have some very fixed views on the definition of AI and these impact on its understanding of how it might ‘emerge’.  I’ve no wish to argue with these assumptions but they *are* assumptions.  Depending on the nature of the AI, it may or may not be possible to do some of the things discussed.  [Just to take a sci-fi analogy, the paper has a reasonably standard view of where AI might come from (e.g. www.amazon.com/WWW-Trilogy-Robert-J-Sawyer/dp/044101853X/) but there are others (e.g. www.amazon.com/Conscious-Vic-Grout/dp/1520590121/) ]

There’s also an implicit assumption that much of the wider world will go on unchanged.  For example, our current geo-political structures will essentially remain in place and that the future still adheres to a capitalist (or technocapitalist) model in which the greatest driver is profit.  Once again, this may well be but, considering the changes that technology itself is likely to bring about, more fundamental social shifts perhaps can’t be ruled out entirely?

Finally, a minor point: the overuse of the word ‘like’ bugs me.  It’s inappropriate in most cases.  Examples: Line 138 (‘such as’ is better) and Line 159 (‘as with’), etc.

OK, so I’m going to recommend it for publication anyway but, the more of these points that can be addressed in the final version, the better it will be.

Author Response

Reviewer 3

The papers offers an intelligent discussion of an interesting and relevant subject.  I’m inclined to recommend acceptance on that basis alone but there is quite a lot that worries me …

Firstly, we have to question the actual ‘contribution’ of the paper.  The title suggests a ‘classification’, the abstract, a ‘discussion’.  I’m not sure the former can be justified: the top-level 1,2,3,4 doesn’t take much thinking about and the presentation in the paper (some tables, some bulleted list, some text) is a bit too haphazard.  The latter is better but the discussion needs a little more focus – or at least some stronger conclusions beyond the obvious, ‘tricky, this, isn’t it?’  What will the major drivers be that determine which of these outcomes we end up with?

Reply: The focus of the paper was changed from mere clarification to finding the best relation between local and global solutions.

This leads to the second major point: a number of assumptions on which the paper seems to be based.  Aside from a slight (but evident) Western or first world perspective, the paper seems to have some very fixed views on the definition of AI and these impact on its understanding of how it might ‘emerge’.  I’ve no wish to argue with these assumptions but they *are* assumptions.  Depending on the nature of the AI, it may or may not be possible to do some of the things discussed.  [Just to take a sci-fi analogy, the paper has a reasonably standard view of where AI might come from (e.g. www.amazon.com/WWW-Trilogy-Robert-J-Sawyer/dp/044101853X/) but there are others (e.g. www.amazon.com/Conscious-Vic-Grout/dp/1520590121/) ].

Reply: These assumptions were added into Introduction. 

There’s also an implicit assumption that much of the wider world will go on unchanged.  For example, our current geo-political structures will essentially remain in place and that the future still adheres to a capitalist (or technocapitalist) model in which the greatest driver is profit.  Once again, this may well be but, considering the changes that technology itself is likely to bring about, more fundamental social shifts perhaps can’t be ruled out entirely?

Reply: The assumption is explicitly mentioned in the new section at the end.

Finally, a minor point: the overuse of the word ‘like’ bugs me.  It’s inappropriate in most cases.  Examples: Line 138 (‘such as’ is better) and Line 159 (‘as with’), etc.

Reply: Most “likes” are replaced with synonymous.

OK, so I’m going to recommend it for publication anyway but, the more of these points that can be addressed in the final version, the better it will be.

Round  2

Reviewer 2 Report

I think the article is sufficiently improved.