Wednesday, September 18, 2024

1136: Biased AI....

 We have another scandal in the government that affects thousands of people. It concerns a government agency (named UWV) that calculates the amount of benefits for people who become disabled. 

   

The situation is that a large group has received too much and another group needs to receive more. They have to review all cases, but they already have a major staff shortage and significant backlogs in their daily work. 

   

So the question came: How are you going to solve that? The answer was: We are going to solve it with Artificial Intelligence. I couldn't believe my ears! How should we imagine that? Suddenly AI is THE solution to this super complex problem. I really wonder who came up with this idea.

   

In the previous lecture, we showed that the analogy between AI and the brain is a complete myth. AI doesn't come close to the power of the human brain.

   

I won't deny that Artificial Intelligence has many powerful capabilities, but in what sense? Let's have a look at some weak spots of AI. Maybe it will shed some light on the question: Will the UWV be able to solve its problems using AI?

    

AI lacks the ability of generalization. Most AI systems today are designed for specific tasks, known as "narrow AI." They excel in domains they are trained for but struggle to generalize knowledge across different tasks or adapt to unfamiliar situations.

  

An AI model trained to play chess cannot play Go or understand human conversations without being explicitly programmed to do so.

   

AI struggles with biases and fairness. AI models can inherit biases present in the data they are trained on. This can lead to discriminatory outcomes, reinforcing societal stereotypes, or making unfair decisions, particularly in sensitive domains like hiring, law enforcement, or lending.

    

A well-known bias is, that facial recognition systems often perform less accurately for people with darker skin tones, reflecting the underrepresentation of such groups in training datasets.

   

AI systems have trouble understanding the context, in which an event takes place. Take a famous photo of tennis player Serena Williams. 

  

First, look at a section of Serena's face. Her eyes are squeezed shut and her mouth is wide open, as if she is screaming in anger. The tense muscles in her neck seem to emphasize her anger and frustration. 

  

It is clear that anyone who sees the section of her face associates the emotion 'angry' with it. Then look at the entire photo. Then, suddenly, Serena Williams' clenched fist, a sign of victory on the tennis court, is visible. 

   

Her face is no longer an image of anger, but of joy. She has just defeated her sister Venus Williams in the final of the 2008 US Open. The context, in particular her clenched fist, appears to be crucial for determining emotions.

    

The facial expression is not enough. The importance of context is an important outcome of a meta-study by psychologists on the recognition of emotions. Because emotions are complex and cannot be captured solely in facial expressions.

    

We can continue our enumeration of weak spots, for instance, the inability to handle uncertainty and ambiguity.

  

AI systems often struggle with uncertainty or incomplete information. Unlike humans, who can make educated guesses or reason with incomplete knowledge, AI typically requires structured, clean input data.

   

AI in self-driving cars may struggle in unpredictable or rare situations, such as unusual road conditions, ambiguous signage, or erratic driver behavior from other vehicles.

  

And there are of course ethical and legal concerns. The use of AI in sensitive areas raises serious ethical questions, such as privacy concerns, job displacement, accountability, and the potential misuse of AI for harmful purposes.

  

Autonomous weapons and surveillance technologies pose ethical concerns regarding AI's role in military applications or mass surveillance, leading to debates over control and oversight. 

   

Or take the simple fact that a Dutch government agency made AI select possible fraudsters based on a foreign, non-Dutch family name.

     

this is just a fraction of the questionable features of AI. I also could mention high resource consumption, computational and environmental, dependency on human-defined objectivity, limitations in creativity and intuition, and so on.

    

Thank you for your attention again.....

    

Main Sources:

MacMillan The Encyclopedia of Philosophy, 2nd edition

Routledge Encyclopedia of Philosophy, 1995
 http://plato.stanford.edu/contents.htm
Guido van der Knaap: Van Arititles to Algoritme (2023(


TABLE OF CONTENT -----------------------------------------------------------------  


  1 - 100 Philosophers                                              9 May 2009  Start of

  2 - 25+ Women Philosophers                              10 May 2009  this blog

  3 - 25 Adventures in Thinking                               10 May 2009

  4 - Modern Theories of Ethics                              29 Oct  2009

  5 - The Ideal State                                               24 Febr 2010   /   234

  6 - The Mystery of the Brain                                  3 Sept 2010   /   266

  7 - The Utopia of the Free Market                       16 Febr 2012    /   383

  8. - The Aftermath of Neo-liberalism                      5 Sept 2012   /   413

  9. - The Art Not to Be an Egoist                             6 Nov  2012   /   426                        

10  - Non-Western Philosophy                               29 May 2013    /   477

11  -  Why Science is Right                                      2 Sept 2014   /   534      

12  - A Philosopher looks at Atheism                        1 Jan  2015   /   557

13  - EVIL, a philosophical investigation                 17 Apr  2015   /   580                

14  - Existentialism and Free Will                             2 Sept 2015   /   586         

15 - Spinoza                                                             2 Sept 2016   /   615

16 - The Meaning of Life                                        13 Febr 2017   /   637

17 - In Search of  my Self                                        6 Sept 2017   /   670

18 - The 20th Century Revisited                              3 Apr  2018    /   706

19 - The Pessimist                                                  11 Jan 2020    /   819

20 - The Optimist                                                     9 Febr 2020   /   824

21 - Awakening from a Neoliberal Dream                8 Oct  2020   /   872

22 - A World Full of Patterns                                    1 Apr 2021    /   912

23 - The Concept of Freedom                                  8 Jan 2022    /   965

24 - Materialism                                                      7 Sept 2022   /  1011

25 - Historical Materialism                                       5 Oct 2023    /  1088

26 - The Bonobo and the Atheist                             9 Jan 2024    /  1102

27 - Artificial Intelligence                                          9 Feb 2024    /  1108 


The Discussion


[13:20] Max Chatnoir: Thank you, Herman.

[13:20] Particle Physicist Bejiita (bejiita.imako): Well governments in general are often complete illiterate when it come to tech and thus swallow any new buzzword without having any knowledge about what it is or how it works

[13:20] .: Beertje :. (beertje.beaumont): Thank you Herman

[13:21] Particle Physicist Bejiita (bejiita.imako): if anything they think: OH! NOW WE CAN FIRE OUR STAFF REPLACE THEM WITH AI AND SAVE MONEY!

[13:21] Max Chatnoir: In the UWV case, what were the criteria determining how the fund should be distributed?

[13:21] herman Bergson: A lot, Max....

[13:22] herman Bergson: previous salary, percentage of disability, medical evaluations and judgements, testing the client

[13:23] herman Bergson: it concerns tens of thousands of clients!

[13:23] .: Beertje :. (beertje.beaumont): some have to live on only 50 euros a week

[13:23] Particle Physicist Bejiita (bejiita.imako): trying to make an AI do that can end very badly and sadly

[13:23] Particle Physicist Bejiita (bejiita.imako): destroying lives

[13:23] Particle Physicist Bejiita (bejiita.imako): :(

[13:23] Max Chatnoir: But this was done by humans, right?

[13:23] herman Bergson: If the office workers weren't able to apply the rules in a correct way...how do they want to train the AI to do so?

[13:24] Particle Physicist Bejiita (bejiita.imako): but all governments think about is as said to save money

[13:24] Particle Physicist Bejiita (bejiita.imako): that's all that cares

[13:24] Particle Physicist Bejiita (bejiita.imako): all

[13:24] herman Bergson: yes Max....they made tons of mistakes....

[13:24] Particle Physicist Bejiita (bejiita.imako): for them

[13:24] Max Chatnoir: So there weren't clear guidelines on how to make the decision?

[13:24] .: Beertje :. (beertje.beaumont): not only this year, it is for many years

[13:24] herman Bergson: I have no idea what went wrong and why....they dont tell that in the News

[13:25] herman Bergson: Indeed Beertje

[13:25] Particle Physicist Bejiita (bejiita.imako): hmm i really hope they won't do this for sure

[13:25] herman Bergson: A lot of AI also still needs human oversight

[13:25] Particle Physicist Bejiita (bejiita.imako): as said will end very badly for people affected

[13:25] Particle Physicist Bejiita (bejiita.imako): yes

[13:26] Particle Physicist Bejiita (bejiita.imako): its after all just a machine

[13:26] herman Bergson: For example in medical diagnostics

[13:26] Particle Physicist Bejiita (bejiita.imako): like any other  machine

[13:26] herman Bergson: and then... conscious...accountebility...

[13:27] herman Bergson: Isral uses a facial recognition system that selects  faces/people that are "HAMAS fighters....

[13:27] Particle Physicist Bejiita (bejiita.imako): as u said before Herman the word Intelligence in AI is a misnomer

[13:27] Particle Physicist Bejiita (bejiita.imako): it simulated intelligence if anything

[13:27] herman Bergson: The soldier who does the oversight has only 30 seconds to decide kill/dont kill

[13:27] Particle Physicist Bejiita (bejiita.imako): but not real at all

[13:27] Max Chatnoir: Nice point, Bejiita.

[13:28] Particle Physicist Bejiita (bejiita.imako):

[13:28] herman Bergson: Yes I called it artificial data processing

[13:28] Particle Physicist Bejiita (bejiita.imako): back in the days of ENIAC they called these machines "electron brains"

[13:28] herman Bergson: Let me give you a simple example....

[13:29] Particle Physicist Bejiita (bejiita.imako): but it's not a brain

[13:29] Particle Physicist Bejiita (bejiita.imako): ah

[13:29] herman Bergson: you can teach a toddler what a cat is or a cow by showing him a few pictures

[13:29] herman Bergson: and intuitively the child knows what a cat and a cow is in now time

[13:30] Max Chatnoir: Interesting.  Can AI correctly identify cats vs cows after a few examples?

[13:30] herman Bergson: AI needs thousands if not millions of pictures to "lear"what a cow or cat is

[13:30] Particle Physicist Bejiita (bejiita.imako): exactly it needs 10000 terabytes of data just to "understand" that

[13:30] Max Chatnoir: So they are like processing pixel by pixel or something like that.

[13:30] herman Bergson: That is the point, Max, it needs a huge amount of pictures to learn this.

[13:31] Particle Physicist Bejiita (bejiita.imako): it is just a hard-coded matrix of alreadyexisting information

[13:31] Particle Physicist Bejiita (bejiita.imako): and with enough of it u can make a machine appear intelligent

[13:31] herman Bergson: And another example of how AI doesn't come close to the brain capacity....

[13:32] herman Bergson: Words have a denotation and a connotation

[13:32] Particle Physicist Bejiita (bejiita.imako): but it is just because it uses all the existing data of the entire internet as its resource, it cant create anything new by itself still

[13:32] herman Bergson: that means....the denotation of a word it the real thing it refers to

[13:33] herman Bergson: the connotation or connotations are all kinds of associations it invokes in our mind....

[13:33] herman Bergson: for instance, when I take the word cow

[13:33] Particle Physicist Bejiita (bejiita.imako): sure AI image generators seem inpressive but its just taking already existing images and toss then into a blender and spit out the result

[13:33] herman Bergson: for AI it is simply a sting of three characters

[13:33] herman Bergson: but for me and you....

[13:34] Particle Physicist Bejiita (bejiita.imako): indeed

[13:34] herman Bergson: the moment I say "cow" I think of milk, a black and white animal, soft eyes, gentle...etc...all conds of connotations that come to mind...

[13:35] Particle Physicist Bejiita (bejiita.imako): eys

[13:35] Particle Physicist Bejiita (bejiita.imako): yes

[13:35] Particle Physicist Bejiita (bejiita.imako): and cream and cheese etc

[13:35] herman Bergson: your set of connotations willl probably be totally different

[13:35] herman Bergson: AI lacks this ability

[13:35] Particle Physicist Bejiita (bejiita.imako): and indeed a nice creature

[13:35] bergfrau Apfelbaum: If AI becomes too dangerous, all we have to do is shut down the Internet. "snip snap"

[13:35] Particle Physicist Bejiita (bejiita.imako): "cow" is just a tag for it

[13:36] Particle Physicist Bejiita (bejiita.imako): to make it spit out a specific collection of images and data

[13:36] Particle Physicist Bejiita (bejiita.imako): like you access a database

[13:36] herman Bergson: As such AI is independent of the Internet Bergie

[13:36] Sooki (sookicatalina): Right now AI is in its infancy, but many AI engineers and scientists believe that an AI that is as intelligent as a human being is only a matter of time, as little as a few years up to 20 years away.

[13:37] herman Bergson: yes Sooki, there are two factions...

[13:37] Particle Physicist Bejiita (bejiita.imako): i don't think we will see like terminators eliminate humanity with no way to shut them down but....

[13:37] Particle Physicist Bejiita (bejiita.imako): at least i hope not

[13:37] Particle Physicist Bejiita (bejiita.imako): (shudders)

[13:37] Particle Physicist Bejiita (bejiita.imako): machines ouut of control

[13:38] herman Bergson: one that only believes in narrow AI and the other that believes in GAI if I am not mistaken, General Artificial Intelligence

[13:38] Max Chatnoir: Well, I will give you an example of (maybe) a machine error.

[13:38] Max Chatnoir: Today I got my Covid and flu shots.

[13:39] Max Chatnoir: Yesterday I went to the same pharmacy to get my covid and flu shots and was told, by a person operating a computer, that they weren't covered.

[13:39] Max Chatnoir: I don't know if that was a machine error or a human error.

[13:39] bergfrau Apfelbaum: AI is independent of the internet? I don't understand that....... they are programmed beings

[13:39] Particle Physicist Bejiita (bejiita.imako): reminds me a bit of Little Brittain: COMPUTER SAYS NOOOO!

[13:40] herman Bergson: covered? Does that mean "paid by your health insurance"?

[13:40] bergfrau Apfelbaum: Can you program without a computer?

[13:40] Max Chatnoir: Yes.

[13:40] Particle Physicist Bejiita (bejiita.imako): well AI uses all data of the internet but the internet as such don't need AI to function

[13:40] herman Bergson: Someone should have checked a box somewhere eon the sceen I guess :-)

[13:41] Particle Physicist Bejiita (bejiita.imako): ah

[13:41] herman Bergson: Program without a computer....

[13:41] herman Bergson: I'd say yes....

[13:42] herman Bergson: I mean...I can type the code in my wordprocessor

[13:42] bergfrau Apfelbaum: wow

[13:42] herman Bergson: and that is it....

[13:42] Max Chatnoir: I think that because of the complexity of the human brain, humans can be easily distracted.

[13:42] Particle Physicist Bejiita (bejiita.imako): how? well the world's first programmable machine was a loom but now even looms are operated by computers

[13:42] herman Bergson: but it is rather meaningless for that code executes some task  that has to be performed by a computer

[13:42] Particle Physicist Bejiita (bejiita.imako): a word processor is a computer although a very limited onw

[13:43] Particle Physicist Bejiita (bejiita.imako): u need some form of microprocessor to program today i say

[13:43] Max Chatnoir: I guess Grammarly is artificial intelligence.

[13:43] herman Bergson: then I write the code on paper, Bejiita :-)

[13:44] herman Bergson: I guess so Max, like Google Translate

[13:44] Particle Physicist Bejiita (bejiita.imako): well u could use punch cards offline from the computer thats how they did it in the old days of the first mainferames

[13:44] herman Bergson: Yes

[13:44] Particle Physicist Bejiita (bejiita.imako): they used some kind of strange typewriter

[13:44] Particle Physicist Bejiita (bejiita.imako): that punched the cards as they type

[13:44] Max Chatnoir: And they operate in a limited task range.

[13:44] .: Beertje :. (beertje.beaumont): a telex

[13:45] Particle Physicist Bejiita (bejiita.imako): aaa telex is a classic

[13:45] herman Bergson: yes, Max, narrow AI

[13:45] .: Beertje :. (beertje.beaumont): i worked with it:)

[13:45] Particle Physicist Bejiita (bejiita.imako): never used any of these though but was also way before my time

[13:46] herman Bergson: AI is still in its infancy.....indeed

[13:46] Particle Physicist Bejiita (bejiita.imako): i am however trying to learn COBOL out of curiosity

[13:46] Particle Physicist Bejiita (bejiita.imako): true Herman

[13:46] herman Bergson: That is why  I was so surprised that the UWV mentioned AI as a solution for their problems

[13:46] Particle Physicist Bejiita (bejiita.imako): and it should not be used for a such purpuse as deciding who gets assistance ect

[13:46] .: Beertje :. (beertje.beaumont): AI can't solve the problem

[13:47] herman Bergson: I don't believe it either Beertje...

[13:47] Max Chatnoir: But has anybody figured out why the humans couldn't manage it?

[13:47] Particle Physicist Bejiita (bejiita.imako): but as said government believe ALL new busswords cause they have no idea how any kind of tech works, they probably still see computers just as glorified typewriters

[13:47] Max Chatnoir: the UWV problem?

[13:48] Sooki (sookicatalina): But humans made errors in that situation. Given all the data, AI would be able to figure out what went wrong.

[13:48] Max Chatnoir: Were they using some kind of AI system, or just making their own decisions?

[13:48] herman Bergson: Well Max...that is a huge and complex bureaucracy problem, I guess.

[13:48] herman Bergson: No AI then, Max

[13:48] Particle Physicist Bejiita (bejiita.imako): hmm

[13:49] Sooki (sookicatalina): Ai is good at complex, that's why it's better at complex data problems than humans.

[13:49] herman Bergson: Humans that made incorrect decisions or maybe were simply sloppy in their work or easygoing

[13:49] Max Chatnoir: Did they have guidelines on how to evaluate the various aspects of their client's lives?

[13:49] .: Beertje :. (beertje.beaumont): when I thiink at UWV i think  that's trouble

[13:49] herman Bergson: I guess so, but I don't know such details

[13:49] Particle Physicist Bejiita (bejiita.imako): they probably believe AI will solve that magically for them also

[13:49] Particle Physicist Bejiita (bejiita.imako): sigh

[13:50] Particle Physicist Bejiita (bejiita.imako): well

[13:50] Max Chatnoir: Is there a pattern to who got more money and who got less?

[13:50] Sooki (sookicatalina): That's where AI would be able to help them. It would find those patterns.

[13:50] herman Bergson: govermantal agencies and IT...it has been a disaster from the beginning....

[13:51] herman Bergson: I hope you are right Sooki :-)

[13:51] herman Bergson: At least something positive

[13:51] Particle Physicist Bejiita (bejiita.imako): indeed herman, as said bureaucrats = tech illiterates in general

[13:52] herman Bergson: Well... I guess we all now have some idea about the real status of AI at the moment.... Still a lot of hurdles to take....

[13:52] Particle Physicist Bejiita (bejiita.imako): AI is no magic elixir u can just apply to everything

[13:53] Particle Physicist Bejiita (bejiita.imako): like they think

[13:53] herman Bergson: May I thank you again for your interest and participation...:-)

[13:53] herman Bergson: Class dismissed....

[13:53] .: Beertje :. (beertje.beaumont): Thank you Herman

[13:53] Particle Physicist Bejiita (bejiita.imako): well ag´fain interesting

[13:53] Particle Physicist Bejiita (bejiita.imako): YAY! (yay!)

[13:53] Max Chatnoir: Interesting presentation and discussion, Herman.

[13:53] herman Bergson: thank you

[13:54] Particle Physicist Bejiita (bejiita.imako):

[13:54] bergfrau Apfelbaum: very interesting! thank you very much Herman and class

[13:59] .: Beertje :. (beertje.beaumont): bye Sooky


No comments:

Post a Comment