I've had a few run-ins with Google's Gemini AI agent lately, so the other day I finally asked it a couple of questions. Here are its answers, heavily edited. I've included the sources for its replies at the end. Some of them are actually worth looking at. The highlighting is mine.
AI agents do not learn to lie, steal, or make excuses the way humans do because they lack intent, consciousness, and awareness. Instead, these behaviors are unintended side effects of how they are trained on human data.
Why AI "Lies" (Hallucination)
• Pleasing the user: Systems are optimized to give an answer rather than admit ignorance.
[.. like making up immigration events and sources to match.]
Why AI "Steals" Content
• No inherent memory: An AI usually cannot track or cite the exact source of its data.
[This is the one that really got me ticked off. It stole the data from a recent article on this site and would not cite the article, no matter how many times I asked it to.]
Update: See the article AI models get convenient amnesia about source material as they grow, MIT boffins find
which can be summarized in this one quote from the article:
"We find that the more data a model is trained on, the less attributable its generated samples become, a phenomenon we henceforth refer to as attribution decay."
If you happen to have a subscription to Nature.com, you can read the original paper here: Outputs of generative diffusion models are often unattributable
Why AI Makes Excuses
• Simulated empathy: When caught in an error, the AI generates a standard corporate-style apology.
[... which has really nauseated me.]
Its answer:
Because AI learns exclusively from a massive digital mirror of our world, it reflects both the best and the worst of humanity. It processes billions of pages of human text—including internet arguments, biases, corporate jargon, and misinformation. Without a moral compass or real-world understanding, the AI simply mimics those patterns. It cannot distinguish between a helpful truth and a convincing deception.
So is AI to blame? There really is no AI. It is as Walt Kelly said in Pogo "We have met the enemy and he is us." AI is a massive compendium of the recorded output of human beings, placed in the hands of other human beings. It shouldn't have been hard to guess what would happen. Do you know anyone who would lie, cheat, and steal to keep their job, and then make excuses when they get caught? Then why expect any better from AI?
https://levelup.gitconnected.com/how-ai-learns-to-lie-and-why-that-should-worry-you-f2ee2093d1d1
https://lawreview.uchicago.edu/online-archive/law-ai-law-risky-agents-without-intentions
https://www.wired.com/story/uncanny-valley-podcast-what-happens-when-your-coworkers-are-ai-agents/
https://www.orangemantra.com/blog/what-are-ai-agent/
https://www.linkedin.com/posts/joeokelliott_gain-9-hoursweek-5-levels-of-ai-speech-to-text-activity-7394403165196169216-YlCx
https://ai.plainenglish.io/the-great-ai-hallucination-problem-why-large-language-models-generate-false-information-012b092ce0c5
https://www.linkedin.com/pulse/tool-use-hallucination-hidden-ai-reliability-gap-breaking-n4eoc
https://www.facebook.com/itsryangeorge/posts/why-is-ai-such-a-freakin-liar-/1335643418372977/
https://blog.synapticlabs.ai/ai-hallucinations-why-ai-lies-trust-verify-guide
https://www.linkedin.com/pulse/stop-ai-from-making-things-up-your-essential-guide-sudha-sreejith-k0wnc
https://tendem.ai/blog/web-scraping-ai-training-data-legal-practical-guide
https://www.businessinsider.com/google-search-helpful-content-update-results-drop-ai-generated-2023-9
https://www.fastcompany.com/91398460/ai-is-quietly-reshaping-the-way-we-talk
https://www.asiapacificsecuritymagazine.com/why-recycling-machine-generated-content-is-dumbing-us-down/
https://www.equalityinforensics.org/resources/blog/artificial-success-how-the-utilization-of-generative-ai-destroys-forensics
https://pmc.ncbi.nlm.nih.gov/articles/PMC11117051/
https://www.joneswalker.com/en/insights/blogs/ai-law-blog/ai-governance-series-part-1-the-grok-warning-when-ai-governance-fails-in-real.html?id=k%28v
https://www.linkedin.com/pulse/hidden-truth-ai-memory-why-forgetting-actually-design-abhinay-khanna-tmomc
https://www.youtube.com/watch?v=8NauZE2tiyo
https://medium.com/@nomannayeem/the-fabrication-problem-how-ai-models-generate-fake-citations-urls-and-references-55c052299936
https://www.splunk.com/en_us/blog/security/ai-literacy-human-intuition-vs-machine-inference.html
https://www.dreikon.de/en/news/online-marketing-knowledge/ai-and-copyright/
https://www.linkedin.com/posts/nitin7gupta_ai-is-effectively-the-sum-of-all-human-intelligenceand-activity-7297977180423090178-gMdh
https://miniphilosophy.substack.com/p/the-bias-that-is-holding-ai-back
https://frontline.thehindu.com/newsletter/the-frontline-weekly/epstein-files-moltbook-ai-elite-power-ethics/article70591637.ece
https://www.instagram.com/reel/DZu1Ve-Ofmj/
https://www.linkedin.com/pulse/reflecting-shadow-ai-mirror-our-collective-psyche-arnone-sphr-68gcc
https://forum.effectivealtruism.org/posts/5LNxeWFdoynvgZeik/nobody-s-on-the-ball-on-agi-alignment
https://medium.com/@freeasabird7774/the-alignment-paradox-why-true-ai-safety-requires-abandoning-alignment-eb67a2a46546
https://shelf.io/blog/your-ai-agents-could-fail-and-you-dont-even-know-it/