July 11, 2018 at 9:23 pm
Comments posted to this topic are about the item AI Regulators
July 12, 2018 at 5:04 am
Privacy and security is good. Massive government powers, controlled by unelected bureaucrats, are not.
The EU is ultimately looking to exert de facto censorship control on the internet. We see it in the way they are demanding that outfits like Google globally delete legitimate information under their made up 'right to be forgotten' (which is also partly a component of GDPR). While they have created some new 'rights' there is no real free speech protection.... news media are regularly banned from reporting some crimes and events (and police are raiding private individuals who dare post those stories on social media), the next goal is to block those stories at the chokepoint (social media organizations), and the mechanisms in GPDR provide the tools to do that.
GPDR is the benign carrot. Watch out for the stick.
...
-- FORTRAN manual for Xerox Computers --
July 12, 2018 at 7:53 am
The danger, to me, is that humans will be lazy. They'll start to trust the AI systems as authorities and use less of their own judgment,
This can be a big problem. Thousands of people here in Michigan were accused of fraud by the Michigan Unemployment Insurance Agency because they relied on an automated system.
July 12, 2018 at 8:37 am
"We should be sure that the goals and results from any AI system continue to befocused on what we want to achieve, and that we transparently define thosegoals for anyone impacted."
The expected AI behavior in the experiment in the Infoscience article was that all communication would be suppressed to conceal the food location. As the article states "...this somewhat surprising result was due to the strength of selection in suppressing information declining concomitantly with the reduction in information content".
Illustrating the gap between machine behavior and human expectation, these kind of results should help focus development of AI to better reflect human behavior. Some interesting research in that area shows how the human body could be reduced to mathematical equations as anthropomorphism for machines. I expect treating the human mind in a similar way will be more challenging.
MattF
July 12, 2018 at 8:51 am
I wonder how good an AI system will be to help EU countries check companies compliance to GDPR. If I understand what you wrote, Steve, I wouldn't be surprised to find that some companies say one thing, but do another. If the AI systems only check companies statements about how they're complying with GDPR, then that's not much of an enforcement.
Kindest Regards, Rod Connect with me on LinkedIn.
July 12, 2018 at 9:08 am
MattF - Thursday, July 12, 2018 8:37 AM"We should be sure that the goals and results from any AI system continue to befocused on what we want to achieve, and that we transparently define thosegoals for anyone impacted."The expected AI behavior in the experiment in the Infoscience article was that all communication would be suppressed to conceal the food location. As the article states "...this somewhat surprising result was due to the strength of selection in suppressing information declining concomitantly with the reduction in information content".
Illustrating the gap between machine behavior and human expectation, these kind of results should help focus development of AI to better reflect human behavior. Some interesting research in that area shows how the human body could be reduced to mathematical equations as anthropomorphism for machines. I expect treating the human mind in a similar way will be more challenging.
I think we humans have a ton of work to do to really understand what we want. That's going to be hard.
July 12, 2018 at 9:27 am
Steve Jones - SSC Editor - Thursday, July 12, 2018 9:08 AMMattF - Thursday, July 12, 2018 8:37 AM"We should be sure that the goals and results from any AI system continue to befocused on what we want to achieve, and that we transparently define thosegoals for anyone impacted."The expected AI behavior in the experiment in the Infoscience article was that all communication would be suppressed to conceal the food location. As the article states "...this somewhat surprising result was due to the strength of selection in suppressing information declining concomitantly with the reduction in information content".
Illustrating the gap between machine behavior and human expectation, these kind of results should help focus development of AI to better reflect human behavior. Some interesting research in that area shows how the human body could be reduced to mathematical equations as anthropomorphism for machines. I expect treating the human mind in a similar way will be more challenging.I think we humans have a ton of work to do to really understand what we want. That's going to be hard.
And what humans want today isn't likely to be what we'll want a year from now.
Kindest Regards, Rod Connect with me on LinkedIn.
July 12, 2018 at 12:20 pm
Rod at work - Thursday, July 12, 2018 8:51 AMI wonder how good an AI system will be to help EU countries check companies compliance to GDPR. If I understand what you wrote, Steve, I wouldn't be surprised to find that some companies say one thing, but do another. If the AI systems only check companies statements about how they're complying with GDPR, then that's not much of an enforcement.
It's only a partial solution. You could have AIs that look for different items, and then check those. This isn't about solving 100% of the issue, but leveraging computers to lower the human labor load.
July 13, 2018 at 9:52 am
Steve Jones - SSC Editor - Thursday, July 12, 2018 12:20 PMRod at work - Thursday, July 12, 2018 8:51 AMI wonder how good an AI system will be to help EU countries check companies compliance to GDPR. If I understand what you wrote, Steve, I wouldn't be surprised to find that some companies say one thing, but do another. If the AI systems only check companies statements about how they're complying with GDPR, then that's not much of an enforcement.It's only a partial solution. You could have AIs that look for different items, and then check those. This isn't about solving 100% of the issue, but leveraging computers to lower the human labor load.
Ah, good point. Sometimes I forget that even a partial solution is better than none. Thanks, Steve.
Kindest Regards, Rod Connect with me on LinkedIn.
July 13, 2018 at 4:57 pm
For me it is all about the training and re-training of the AI. Once it has been properly trained and believed to be good, there needs to be random checks on those claims that have been rejected.
Just like humans, AI can make mistakes and we need to keep honing their behavior.
Viewing 10 posts - 1 through 9 (of 9 total)
You must be logged in to reply to this topic. Login to reply