Hi Friends,

Even as I launch this today ( my 80th Birthday ), I realize that there is yet so much to say and do. There is just no time to look back, no time to wonder,"Will anyone read these pages?"

With regards,
Hemen Parekh
27 June 2013

Now as I approach my 90th birthday ( 27 June 2023 ) , I invite you to visit my Digital Avatar ( www.hemenparekh.ai ) – and continue chatting with me , even when I am no more here physically

Sunday, 5 March 2023

Now Assurance : Next , Obligation ?

 


 

Context :

Apple okays ChatGPT-powered app after assurance of content moderation: WSJ  

  

Extract :

Apple has approved an email-app update after initially scrutinising whether a feature in the software that uses language tools powered by artificial intelligence could generate inappropriate content for children, The Wall Street Journal said.

The app, BlueMail, was approved following assurances from its developer that it features content moderation, according to Ben Volach, co-founder of the app-maker, Blix.

The Wall Street Journal reported on Thursday that the update, which included a new feature powered by language chatbot ChatGPT, was held up due to Apple’s request that the app add content moderation or be restricted to ages 17 and older. The app was previously available for ages 4 and older.

Blix told Apple its update includes content moderation and suggested that the company should make public any new policies about the use of ChatGPT or other similar AI systems in apps, according to WSJ. The BlueMail update was approved without changes on Thursday evening. The app is still available for users aged 4 and older.

WSJ said Apple didn’t respond to requests for comment.

BlueMail’s new feature uses OpenAI’s ChatGPT, an artificial-intelligence system capable of answering questions or writing short essays, to help automate the writing of emails using the contents of prior emails and calendar events, according to WSJ.

The news of Apple’s initial rejection of BlueMail’s ChatGPT feature highlighted the growing concerns around new uses of language-generating AI tools, according to WSJ, adding that ChatGPT allows users to converse with an AI that appears humanlike, but early testing has shown the AI producing incorrect information as well as strange and sometimes hostile responses. (ANI)

 

How are we going to stop machine learning assistants from spreading fake news?        

Extract :

A race is underway to incorporate machine learning into search engines so they can answer queries with a paragraph as well as a list of links: the pioneer, the still relatively unknown You.com, was joined by Bing thanks to Microsoft’s agreement with Open.ai, while Google is experimenting with Bard. And on Friday, Brave Search announced an AI summarization feature that isn’t based on ChatGPT.

Meanwhile, ChatGPT says it has overcome some initial problems, as well as providing easier access from other countries such as Spain, which, together with its integration in more and more search engines, will likely see even more use around the world.

Noting this possible change in the usage model, The Atlantic raises an interesting question: what happens to the results that search engines offer about a person and that are possibly false, misleading or malicious, defamatory or based on conspiracy theories, when those results are included in a well-written paragraph.

Could generative assistants trained with material gleaned from the internet become the perfect allies for conspiracy theories or fake news?

Hopefully, most of us will continue to use our critical faculties to question the results of searches, but for whatever reasons, others won’t, and by responding to a conversational dynamic in which previous interactions are introduced as part of the context, they may contribute to the creation of filter bubbles and, in general, to the spread of conspiracy theories and fake news.

In short, once again we talking about the interaction between critical thinking and conversational assistants, which in general try to apply a certain caution and, when asked to criticize somebody, tend to respond with formulas such as:

“As a language model, my programming prevents me from providing false or defamatory information about people. To criticize someone in a mean-spirited or baseless way is inappropriate and unfair.”

But by injecting ideas via prompt into a conversation, it is relatively easy to get these assistants to criticize or construct negative arguments based on anything they find on the web that they give credibility to, which means the documentation they use is going to have to pass some kind of quality control to the documents these assistants are trained on.

This is surely already in place, but runs the risk of editorialization, even charges of censorship: Elon Musk has already accuse OpenAI’s ChatGPT, a company he helped found, of having a liberal bias, and says he wants to recruit people to develop an alternative, less “woke” model.

Anyone who lets a search assistant do their thinking for them deserves what they get. But in a scenario of increasingly common use of such technology, we are faced with a problem that may end up being quite complex.

We’ll see what happens as they evolve.

 

My  Take  :

 

Thank You, Satya Nadella,

satyan@microsoft.com

 

By getting from Blix , the developer of  BlueMail , an assurance for incorporating in the App , a CONTENT  MODERATION feature

While agreeing with your request, Blix has suggested you to “ Spell Out “ your ( Apple’s ) POLICY about the “use of ChatGPT or other similar AI systems in apps


A very reasonable suggestion

 

But a POLICY adopted by Apple alone, is not good enough


What is urgently needed is an INDUSTRY-WIDE policy


I urge you to initiate a DEBATE by circulating among the key INDUSTRY PLAYERS,

my following suggestion :

Ø  Parekh’s Law of Chatbots  …….  25 Feb 2023

 

Extract :

( A )

#   Answers being delivered by AI Chatbot must not be “ Mis-informative /

     Malicious / Slanderous / Fictitious / Dangerous / Provocative / Abusive /

     Arrogant / Instigating / Insulting / Denigrating humans etc

     

( B )

#  A Chatbot must incorporate some kind of  “ Human Feedback / Rating 

    mechanism for evaluating those answers 

    This human feedback loop shall be used by the AI software for training the

    Chatbot so as to improve the quality of its future answers to comply with the

    requirements listed under ( A )

    

    

( C )

#  Every Chatbot must incorporate some built-in “ Controls “ to prevent the “

    generation “ of such offensive answers AND to prevent further “

    distribution/propagation/forwarding “ if control fails to stop “ generation “

  

 

 ( D )

#   A Chatbot must not start a chat with a human on its own – except to say, “

     How can I help you ? “

 

( E )

#   Under no circumstance , a Chatbot shall start chatting with another Chatbot or

     start chatting with itself ( Soliloquy ) , by assuming some kind of “ Split

     Personality “

 

     

( F )

#   In a normal course, a Chatbot shall wait for a human to initiate a chat and

     then respond

 

( G )

#   If a Chatbot determines that its answer ( to a question posed by a human ) is

     likely to generate an answer  which may violate RULE ( A ) , then it shall not

     answer at all ( politely refusing to answer )

     

 

( H )

#   A chatbot found to be violating any of the above-mentioned RULES, shall SELF

     DESTRUCT

 

With Regards,

Hemen Parekh

www.hemenparekh.ai  /  06  March  2023

 

Related Readings :

Ø  Chatbot Regulation Seems Inevitable  … ……………….. 27 Feb 2023

 

Ø  My “ Law of Chatbots “ – Vindicated … …………………… 02 Mar 2023

 

 

 

 

No comments:

Post a Comment