University Study of AI Powerhouse ChatGPT Reveals Clear Leftist Bias (Read 1117 times)

aletheuo137

https://www.breitbart.com/tech/2023/08/17/university-study-of-ai-powerhouse-chatgpt-reveals-clear-leftist-bias/

A study of OpenAI’s ChatGPT, conducted by researchers at the University of East Anglia in the UK, shows that the market-leading AI chatbot has a clear bias towards leftist political parties.

Sent from my moto g power (2021) using Tapatalk

eyeeatingfish

Re: University Study of AI Powerhouse ChatGPT Reveals Clear Leftist Bias
« Reply #1 on: August 22, 2023, 10:59:47 PM »
https://www.breitbart.com/tech/2023/08/17/university-study-of-ai-powerhouse-chatgpt-reveals-clear-leftist-bias/

A study of OpenAI’s ChatGPT, conducted by researchers at the University of East Anglia in the UK, shows that the market-leading AI chatbot has a clear bias towards leftist political parties.

Sent from my moto g power (2021) using Tapatalk

I don't doubt it but I am curious how this came to be. Could be intentional (programming it to be bias), could be unintentional (feeding it disproportionate levels of one side), or a natural product of what is on the internet.

Flapp_Jackson

Re: University Study of AI Powerhouse ChatGPT Reveals Clear Leftist Bias
« Reply #2 on: August 22, 2023, 11:35:32 PM »
I don't doubt it but I am curious how this came to be. Could be intentional (programming it to be bias), could be unintentional (feeding it disproportionate levels of one side), or a natural product of what is on the internet.

Did you read the entire article?

Quote
While the research project did not set out to determine the reasons for
the political bias, the findings did point towards two potential sources.

The first was the training dataset – which may have biases within it, or
added to it by the human developers, which the developers’ ‘cleaning’
procedure had failed to remove. The second potential source was the
algorithm itself, which may be amplifying existing biases in the training
data.

When developing a software system, test data is normally generated to verify design requirements.  For example, if the program must differentiate between homonym  misspellings (their, there, here, hear, plus words with the same spelling but totally different meanings, like river bank versus financial bank versus aircraft bank, ...), then the context must be analyzed.  Then the results can be used to verify if all cases were handled properly.

I know I've created test data with my name, famous people, current events, pop culture references, bands, and so on.  It breaks up the monotony, serves the purpose for a test, and self-documents itself as test data -- immediately recognizable and bogus data.  But a live system should start with live data, not a collection of test cases.

Funny story.  One of the reasons my company chose me to move to Hawaii was a demo I helped give for some very high up DC dignitaries visiting Langley AFB.  The demo went very well, and I figured that was the last time i'd see that data since i was moving to HI.

After 3 years, when the contract I was under ended, I was flown to VA to renegotiate my continued work here.  While waiting for the operations manager to see me, I picked up a packet that was part of a current contract proposal.  On the first page was a screenshot of the subject system.  I grinned, because the picture was from my demo 3 years earlier.  It was easy to notice, since the air tracks on the display were named after me.   :rofl:

i guess it was easier to give demos without having to start from scratch.   :geekdanc:
The reasonable man adapts himself to the world;
the unreasonable one persists in trying to adapt the world to himself.
Therefore, all progress depends on the unreasonable man.
-- George Bernard Shaw