It's your human hubris holding back AI acceptance
- Reference: 1675449006
- News link: https://www.theregister.co.uk/2023/02/03/human_hubris_may_prevent_people/
- Source link:
But with training, we can learn to overcome our biases and trust our automated advisors.
In [1]a preprint paper titled "Knowing About Knowing: An Illusion of Human Competence Can Hinder Appropriate Reliance on AI Systems," Gaole He, Lucie Kuiper, and Ujwal Gadiraju, from Delft University of Technology, examine whether the Dunning-Kruger effect hinders people from relying on recommendations from AI systems.
[2]
The Dunning-Kruger effect (DKE) dates back to [3]research from 1999 by psychologists David Dunning and Justin Kruger, "Unskilled and unaware of it: How difficulties in recognizing one's own incompetence lead to inflated self-assessments."
[4]
[5]
Dunning and Kruger posit that incompetent people lack the capacity to recognize their incompetence and thus tend to overestimate their abilities.
Assuming DKE exists – [6]something not everyone agrees on – the Delft researchers suggest this cognitive condition means AI guidance may be lost on us. That's not ideal since AI systems presently tend to be pitched as assistive systems that augment human decision-making rather than autonomous systems that operate without oversight. Robo help doesn't mean much if we don't accept it.
[7]
"This a particularly important metacognitive bias to understand in the context of human-AI decision making, since one can intuitively understand how inflated self-assessments and illusory superiority over an AI system can result in overly relying on oneself or exhibiting under-reliance on AI advice," state He, Kuiper, and Gadiraju in their paper, which has been conditionally accepted to [8]CHI 2023 . "This can cloud human behavior in their interaction with AI systems."
To test this, the researchers asked 249 people to answer a series of multiple choice questions to test their reasoning. The respondents were asked to answer questions first by themselves and then with the help of an AI assistant.
[9]The questions , available in the [10]research project GitHub repository , consisted of a series of questions like this:
Physician: In comparing our country with two other countries of roughly the same population size, I found that even though we face the same dietary, bacterial, and stress-related causes of ulcers as they do, prescriptions for ulcer medicines in all socioeconomic strata are much rarer here than in those two countries. It's clear that we suffer significantly fewer ulcers, per capita, than they do.
The study participants were then asked, Which one of the following, if true, most strengthens the physician's argument?
The two countries that were compared with the physician's country had approximately the same ulcer rates as each other.
The physician's country has a much better system for reporting the number of prescriptions of a given type that are obtained each year than is present in either of the other two countries.
A person in the physician's country who is suffering from ulcers is just as likely to obtain a prescription for the ailment as is a person suffering from ulcers in one of the other two countries.
Several other countries not covered in the physician's comparisons have more prescriptions for ulcer medication than does the physician's country.
After respondents answered, they were presented with the same questions as well as an AI system's recommended answer (D for the question above), and were given the opportunity to change their initial answer. This approach, the researchers say, has been [11]validated by past research [PDF].
Based on the answers they received, the three computer scientists conclude "that DKE can have a negative impact on user reliance on the AI system…"
[12]
But the good news, if that's the right term, is that DKE is not destiny. Our mistrust of AI can be trained away.
"To mitigate such cognitive bias, we introduced a tutorial intervention including performance feedback on tasks, alongside manually crafted explanations to contrast the correct answer with the users’ mistakes," the researchers explain. "Experimental results indicate that such an intervention is highly effective in calibrating self-assessment (significant improvement), and has some positive effect on mitigating under-reliance and promoting appropriate reliance (non-significant results)."
Yet if tutorials helped those exhibiting overconfidence (DKE), corrective re-education had the opposite effect on those who initially underestimated their capabilities: It made them either overconfident or possibly algorithm averse – [13]a known consequence [PDF] of seeing machines make mistakes.
[14]ChatGPT has mastered the confidence trick, and that's a terrible look for AI
[15]Machismo is ruining the tech industry for all of us. Equally
[16]How the Dunning-Kruger effect will stop techies buying houses
[17]Criminal mastermind signed name as 'Thief' on receipts after buying stuff with stolen card
In all, the researchers conclude that more work needs to be done to understand how human trust of AI systems can be shaped.
We'd do well to recall the words of HAL, from 2001: A Space Odyssey .
I know I've made some very poor decisions recently, but I can give you my complete assurance that my work will be back to normal. I've still got the greatest enthusiasm and confidence in the mission. And I want to help you.
®
Get our [18]Tech Resources
[1] https://arxiv.org/abs/2301.11333
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Y92SEYyvsElKGuObagedxQAAAFM&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubmed.ncbi.nlm.nih.gov/10626367/
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Y92SEYyvsElKGuObagedxQAAAFM&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Y92SEYyvsElKGuObagedxQAAAFM&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.mcgill.ca/oss/article/critical-thinking/dunning-kruger-effect-probably-not-real
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Y92SEYyvsElKGuObagedxQAAAFM&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://chi2023.acm.org/
[9] https://github.com/RichardHGL/CHI2023_DKE/blob/main/anonymous_data/selected_samples.csv
[10] https://github.com/RichardHGL/CHI2023_DKE
[11] https://faculty.wharton.upenn.edu/wp-content/uploads/2016/08/Dietvorst-Simmons-Massey-2018.pdf
[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Y92SEYyvsElKGuObagedxQAAAFM&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[13] https://repository.upenn.edu/cgi/viewcontent.cgi?article=1392&context=fnce_papers
[14] https://www.theregister.com/2022/12/12/chatgpt_has_mastered_the_confidence/
[15] https://www.theregister.com/2016/03/04/machismo_is_ruining_tech/
[16] https://www.theregister.com/2009/11/18/dunning_kruger/
[17] https://www.theregister.com/2019/08/16/card_thief_signed_name_as_thief_on_sales_receipts/
[18] https://whitepapers.theregister.com/
In all, the researchers conclude that more work needs to be done to understand how human trust of AI systems can be shaped.
Human interactions tend to involve consideration about various courses of action and the potential risks of each choice. We are pretty hardwired to do that; if an AI simply provides a recommended course of action that appears more risky than another "obvious" option then that's more likely to get picked.
It would have to be said that personally, I don't feel that an AI is actually intelligent. It parses data at an incredible speed and can produce results that you don't expect, but that's also true of a complicated excel spreadsheet. Fundamentally existing AI's are simply executing programmed commands within a framework without actually understanding what they are doing, and that can occasionally be dangerous.
If you want me to "trust" an AI then program it to provide the "thinking" behind the courses of action it's suggesting and discuss the tradeoffs in it's recommendations.
I don't feel that an AI is [currently] actually intelligent.
Agreed. The research must provide its definition of intelligence. Until that refutes the rest of your paragraph, we're agreed that it's an expert system at best, or maybe just Excel!
If you want me to "trust" an AI [is a physician] then program it to provide [1]certificates from the GMC (or similar).
Was it a blind study, or did the victims know they were talking to software? If it's testing trust in software, not the resultant response, that's not DKE. Maybe better to have been Turing compliant, eg. could be AI software, could be a qualified human physician. What biases you to change of stay, the alternative suggestion, or its source? If you don't trust your GP because they're a female, that's bigotry. (If you don't trust your "GP" because it's "fucking AI woot!" you're probably sane at least!)
A human with superior knowledge/experience to another can still be wrong. A persuasive but incorrect argument is still wrong. Being swayed by "an authority" isn't a given (ref. SW1A 0AA).
[1] https://www.gmc-uk.org/registration-and-licensing/managing-your-registration/certificates
Which is all fine and dandy...
...but the current crop of consumer-grade AI frequently provide incorrect answers and remain incapable of basic error checking, confidence rating, or source citation, so you'd be an absolute fool to rely on them.
If you're going to train a machine to mimic a human, the best possible outcome is a perfectly human machine. There are billions of perfectly human humans who will be happy to present thoughts and opinions of little to no worth in any given context. You may even think you've just spotted another one.
Er... Is it "C"?
I answered all Cs, as in a bunch of.
The "Dune" Butlerian Jihad cannot come too soon.
The Worm Turned?
A huge turning circle for a... woah! Now that is a rather large worm! (No Dougal It's not close at all! Run!!!)
I was a bit confused by the sample question presented. I wasn't sure if we were being asked to conclude that the lower level of ulcers (asserted in the final sentence) should be inferred from the prior text. If we were, then this is clearly not true and really none of the answer really work.
The prior text only talks about ulcer medication and not quantity of incidents of ulcer medications.
Since I don't believe the last sentence, questions about which "evidence" is more convincing seem pretty moot.
Outsourcing human experience.
If we outsource fundamental human experience to AI, the humans have no more relevance, and would be better replaced with AI drones entirely.
Human beings have a pesky tendency to become miserable unthinking suicidal creatures when when life overwhelmingly demonstrates that they are merely an accessory object to an inhuman overlord.
If everyone's miserable then then all lives in all births become an exercise in pointless suffering.
Should we encourage this socially-cloaked quest to make everyone lost in seeming pointlessness and miserable apathy unto self?
I say no.
Happy accidents, or recycled mistakes into innovation by our own ideas, is the source frown which the majority of human Joy comes from.
If things become too stabilized, life loses any savour as an adventure it once held.