[R] Off-topic: ChatGPT Code Interpreter

Tue Jul 18 20:21:44 CEST 2023

Just to bring it back to R, I want to point out that what many R programmers do is not that different. If you develop some skills at analyzing some kinds of data and have a sort of toolchest based on past work, then a new project along similar lines may move very quickly.  After a while, you may put out a package that allows even novices to do a decent job as the code handles so many of the details or provides default options you can over-ride.

So if an AI has the ability to use the same tools you provided, and applies them properly, is there much difference?

I have seen EXPERTS do horribly when yanked just outside their field of expertise. They may use a tool outside the bounds it was designed for, as an example, or pick an unfamiliar model that is not applicable or optimal.

The problem with AI is compounded as some kinds of peer feedback may be missing while other things have not yet been programmed well that allow some selectivity and so on.

But if a simple regression often works well enough, why can't an AI use it too? 

I will say a lot of what people do in R is cleaning the data and getting it into the right form. Sometimes humans struggle to detect if say two names are the same and someone made a spelling mistake. And humans will still get things like that wrong unless they can go back and consult external resources to see if say the school has two teachers with similar names or whether all the students named should be combined under the same teacher for some purposes.

I will point out there is no good reason to think an Ai would necessarily use R or use base R rather than some of the functions in the tidyverse. I am studying Wolfram and an amazing amount of work can be done in one liners that rely on many thousands of built-in functionality. So why would you want an AI to write assembly code or do it in a language like C but rather pick a language well suited for whatever task.

-----Original Message-----
From: R-help <r-help-bounces using r-project.org> On Behalf Of Spencer Graves
Sent: Monday, July 17, 2023 3:13 PM
To: Bert Gunter <bgunter.4567 using gmail.com>; R-help <R-help using r-project.org>
Subject: Re: [R] Off-topic: ChatGPT Code Interpreter

	  I don't know about ChatGPT, but Daniel Kahneman won the 2002 Nobel 
Memorial Prize in Economics,[1] even though he's not an economist, for 
his leadership in creating a new subfield in the intersection of human 
psychology and economics now called "behavioral economics".[2] Then in 
2009 Kahneman and Gary Klein published an article on, "Conditions for 
intuitive expertise: a failure to disagree", which concluded that expert 
intuition is learned from frequent, rapid, high-quality feedback. 
People you do not learn from frequent, rapid, high-quality feedback can 
be beaten by simple heuristics developed by intelligent lay people.[3] 
That includes most professions, which Kahneman Sibony and Sunstein call 
"respect-experts".

	  Kahneman Sibony and Sunstein further report that with a little data, 
a regression model can outperform a simple heuristic, and with massive 
amounts of data, artificial intelligence can outperform regression 
models.[4]

	  An extreme but real example of current reality was describe in an 
article on "Asylum roulette":  With asylum judges in the same 
jurisdiction with cases assigned at random, one judge approved 5 percent 
of cases while another approved 88 percent.[5] However, virtually all 
"respect-experts" are influenced in their judgements by time of day and 
whether their favorite sports team won or lost the previous day.  That 
level of noise can be reduced dramatically by use of appropriate 
artificial intelligence.

	  Comments?
	  Spencer Graves

[1]

https://en.wikipedia.org/wiki/Daniel_Kahneman

[2]

https://en.wikipedia.org/wiki/Behavioral_economics

[3]

https://www.researchgate.net/publication/26798603_Conditions_for_Intuitive_Expertise_A_Failure_to_Disagree

[4]

Daniel Kahneman; Olivier Sibony; Cass Sunstein (2021). Noise: A Flaw in 
Human Judgment (Little, Brown and Company).

[5]

https://en.wikipedia.org/wiki/Refugee_roulette

On 7/17/23 1:46 PM, Bert Gunter wrote:
> This is an **off-topic** post about the subject line, that I thought
> might be of interest to the R Community. I hope this does not offend
> anyone.
> 
> The widely known ChatGPT software now offers what  is called a "Code
> Interpreter," that, among other things, purports to do "data
> analysis."  (Search for articles with details.) One quote, from the
> (online) NY Times, is:
> 
> "Arvind Narayanan, a professor of computer science at Princeton
> University, cautioned that people should not become overly reliant on
> code interpreter for data analysis as A.I. still produces inaccurate
> results and misinformation.
> 
> 'Appropriate data analysis requires just a lot of critical thinking
> about the data,” he said.' "
> 
> Amen. ... Maybe.
> 
> (As this is off-topic, if you wish to reply to me, probably better to
> do so privately).
> 
> Cheers to all,
> Bert
> 
> ______________________________________________
> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.

______________________________________________
R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.