[Lex Computer & Tech Group/LCTG] AI bias

[Robert Primak]

 About certain replies getting tagged as Spam by email servers:
I get this too. Not really consistent in Yahoo Web Mail. Occasionally one or more Lexington Technology mail list replies will fall into Spam. Happens with other mail list services too. Does not track with how frequent the automatic sending of emails to my account from the mail list is. 
I just have to be careful when checking my Spam, so as not to miss things like this which get routed there by mistake. I blame the AI algorithms for all of this. 
AI isn't as "intelligent" as I might hope!
If anyone subscribes to the paid version of AskWoody Newsletter, today's Brian Livingston Silicon item (Post Number 3 in the Newsletter) touches upon the problems with AI, focusing on how advances in chip design are contributing to increasing development of AI. 
Hopefully, AI gets better with time. And regarding Spam Filters, maybe it learns that we want to read ALL the emails from our mail lists, not just a few per day. 
-- Bob Primak 
    On Monday, April 11, 2022, 09:13:20 AM EDT, Jerry Harris <jerryharri at gmail.com> wrote:  
 
 Bob, I'm curious about something. Some emails coming through the LCTG group get tagged as Spam by Gmail, and others don't. Your emails, for example, get tagged. But I've whitelisted the alias, so Gmail will still display your emails with an annoying "Is this spam?" message. 
Are my emails getting through to you? Eg, did you get this email I sent on Sat? 

Regards, Jerry
On Sun, Apr 10, 2022 at 7:01 PM Robert Primak <bobprimak at yahoo.com> wrote:

 Jerry -- 
>From your mouth (or keyboard) to God's ears! Absolutely this could be done. But do we have a prayer of a chance in real life? Maybe, and there is always hope.
-- Bob Primak 

    On Sunday, April 10, 2022, 12:53:44 PM EDT, Jerome Slate <slatemd at comcast.net> wrote:  
 
  There is the remote possibility that the same AI values about crime might be used for benevolent purposes. It could be used to signal areas to develop afternoon programs for at risk children, places to open job training schools and subsidized centers for parent training and child care. Unlikely, but possible. 
 Jerry Slate
 
 On 4/9/2022 2:29 PM, Jerry Harris wrote:
  
 
 > wouldn't surprise me if zip codes correlate with a crime. (Probably not with the amount stolen per capita!) It wouldn't surprise me at all if criminal records of relatives correlate more strongly with crime 
  The bias comes into play by using this data to give people from certain zip codes or certain families longer sentences. Or to assign more police officers to certain regions, where in turn they make more arrests for minor crimes they observe (eg, broken tail light, etc), and thus feedback misleading data into the models.  
  Cathy imagines these algorithms to be used to make different decisions:  
Imagine if you used recidivist models to provide the at-risk inmates with counseling and job training while in prison. Or if police doubled down on foot patrols in high crime zip codes -- working to build relationships with the community instead of arresting people for minor offenses. 
 
You might notice there's a human element to these solutions. Because really that's the key. Algorithms can inform and illuminate and supplement our decisions and policies. But to get not-evil results, humans and data really have to work together.
 
"Big Data processes codify the past," O'Neil writes. "They do not invent the future. Doing that requires moral imagination, and that's something only humans can provide."  
 
  https://money.cnn.com/2016/09/06/technology/weapons-of-math-destruction/index.html
 
  Jerry   
  On Sat, Apr 9, 2022 at 12:47 PM Jon Dreyer <jon at jondreyer.org> wrote:
  
  
Not every instance of apparent bias represents bias. John's example seems like an open-and-shut case of bias, but Jerry's may not be. Since zip codes correlate with poverty and poverty correlates with crime, it wouldn't surprise me if zip codes correlate with crime. (Probably not with amount stolen per capita!) It wouldn't surprise me at all if criminal records of relatives correlate more strongly with crime.
 
Whether it's ethical to use that data for a given purpose is a separate question.
 
-- 
 Jon "Bias Sometime" Dreyer
 Math tutor/Computer Science tutor
 Jon Dreyer Music
 On 4/9/22 11:22, Jerry Harris wrote:
  
 John,  You might find the book by Cathy O'Neil, "Weapons of Math Destruction," interesting since it contains more examples in the same vein as the article. The author attended Lexington High School and gave a talk about her book at Follen Church several years ago. Stories such as an algorithm that uses data like zip code and criminal records of relatives to calculate a prisoner's likelihood to commit another crime if released on parole, etc.      https://www.amazon.com/Weapons-Math-Destruction-Increases-Inequality/dp/0553418815 
  Jerry  
  On Sat, Apr 9, 2022 at 10:47 AM john rudy <jjrudy1 at comcast.net> wrote:
  
   
I am taking an AI course and a major issue is the bias build into some AI systems.  If, for example, you pick new hires from those who were successful in the past you’ll mostly hire white males because those were who you mostly had in the past.  Here is a fascinating article from the course
 
https://www.wired.com/story/excerpt-from-automating-inequality/
   
  
  ===============================================
 ::The Lexington Computer and Technology Group Mailing List::
 Reply goes to sender only; Reply All to send to list.
 Send to the list: LCTG at lists.toku.us      Message archives: http://lists.toku.us/private.cgi/lctg-toku.us
 To subscribe: email lctg-subscribe at toku.us  To unsubscribe: email lctg-unsubscribe at toku.us
 Future and Past meeting information: http://LCTG.toku.us
 List information: http://lists.toku.us/listinfo.cgi/lctg-toku.us
 This message was sent to jerryharri at gmail.com.
 Set your list options: http://lists.toku.us/options.cgi/lctg-toku.us/jerryharri@gmail.com
  
  ===============================================
::The Lexington Computer and Technology Group Mailing List::
Reply goes to sender only; Reply All to send to list.
Send to the list: LCTG at lists.toku.us      Message archives: http://lists.toku.us/private.cgi/lctg-toku.us
To subscribe: email lctg-subscribe at toku.us  To unsubscribe: email lctg-unsubscribe at toku.us
Future and Past meeting information: http://LCTG.toku.us
List information: http://lists.toku.us/listinfo.cgi/lctg-toku.us
This message was sent to slatemd at comcast.net.
Set your list options: http://lists.toku.us/options.cgi/lctg-toku.us/slatemd@comcast.net 
 ===============================================
::The Lexington Computer and Technology Group Mailing List::
Reply goes to sender only; Reply All to send to list.
Send to the list: LCTG at lists.toku.us      Message archives: http://lists.toku.us/private.cgi/lctg-toku.us
To subscribe: email lctg-subscribe at toku.us  To unsubscribe: email lctg-unsubscribe at toku.us
Future and Past meeting information: http://LCTG.toku.us
List information: http://lists.toku.us/listinfo.cgi/lctg-toku.us
This message was sent to bobprimak at yahoo.com.
Set your list options: http://lists.toku.us/options.cgi/lctg-toku.us/bobprimak@yahoo.com  ===============================================
::The Lexington Computer and Technology Group Mailing List::
Reply goes to sender only; Reply All to send to list.
Send to the list: LCTG at lists.toku.us      Message archives: http://lists.toku.us/private.cgi/lctg-toku.us
To subscribe: email lctg-subscribe at toku.us  To unsubscribe: email lctg-unsubscribe at toku.us
Future and Past meeting information: http://LCTG.toku.us
List information: http://lists.toku.us/listinfo.cgi/lctg-toku.us
This message was sent to jerryharri at gmail.com.
Set your list options: http://lists.toku.us/options.cgi/lctg-toku.us/jerryharri@gmail.com
  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.toku.us/pipermail/lctg-toku.us/attachments/20220411/7ca046d1/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 56927 bytes
Desc: not available
URL: <http://lists.toku.us/pipermail/lctg-toku.us/attachments/20220411/7ca046d1/attachment.png>