In this final guest episode of the year, we explore AI as a kind of Trojan horse: a technology that promises one thing while carrying hidden costs inside it. Those costs show up in data centers, energy and water systems, local economies, and the communities asked to host the infrastructure that makes AI possible.

We’re joined by Jon Ippolito and Joline Blais from the University of Maine for a conversation that starts with AI’s environmental footprint and expands into questions of extraction, power, education, and ethics.

In this episode, we discuss:

• Why AI can function as a Trojan horse for data extraction and profit
• What data centers actually do, and why they matter
• The environmental costs hidden inside “innovation” narratives
• The difference between individual AI use and industrial-scale impact
• Why most data center activity isn’t actually AI
• How communities are pitched data centers—and what’s often left out
• The role of gender in ethical decision-making in tech
• What AI is forcing educators to rethink about learning and work
• Why asking “Who benefits?” still cuts through the hype
• And how dissonance can be a form of clarity

Resources mentioned:

• IMPACT Risk framework: https://ai-impact-risk.com
• What Uses More:
  https://what-uses-more.com

Guests:

• Jon Ippolito – artist, writer, and curator who teaches New Media and Digital Curation at the University of Maine.
• Joline Blais – researches regenerative design, teaches digital storytelling and permaculture, and advises the Terrell House Permaculture Center at the University of Maine.

Leave us a comment or a suggestion!

Support the show

Contact us: https://www.womentalkinboutai.com/

Why can AI crush law exams and chess grandmasters, yet still struggle with word games? In this episode, Kimberly and Jessica use Moravec's Paradox to unpack why machines and humans are "smart" in such different ways—and what that means for how we use AI at work and in daily life.

They start with a practical fact-check on agentic AI: what actually happens to your data when you let tools like ChatGPT or Gemini access your email, calendar, or billing systems, and which privacy toggles are worth changing. From there, they dive into why AI fails at the New York Times' Connections game, how sci-fi anticipated current concerns about AI psychology decades ago, and what brain-computer interfaces like Neuralink tell us about embodiment and intelligence.

Along the way: sycophantic bias, personality tests for language models, why edtech needs more friction, and a lighter "pit and peach" segment with unexpected life hacks.

Resources by Topic

Privacy & Security (ChatGPT)

• OpenAI Memory & Controls (Official Guide): https://openai.com/index/memory-and-new-controls-for-chatgpt/
• OpenAI Data Controls & Privacy FAQ: https://help.openai.com/en/articles/7730893-data-controls-faq
• OpenAI Blog: Using ChatGPT with Agents: https://platform.openai.com/docs/guides/agents
  
  

Moravec's Paradox & Cognitive Science

• Moravec's Paradox (Wikipedia): https://en.wikipedia.org/wiki/Moravec's_paradox
• "The Moravec Paradox" - Research Paper: https://en.wikipedia.org/wiki/Hans_Moravec

Sycophancy & LLM Behavior

• "Sycophancy in Large Language Models: Causes and Mitigations" (arxiv): https://arxiv.org/abs/2411.15287
  
  
• "Personality Testing of Large Language Models: Limited Temporal Stability, but Highlighted Prosociality": https://royalsocietypublishing.org/doi/10.1098/rsos.240180
  
  

Brain-Computer Interfaces & Embodied AI

• Neuralink: "A Year of Telepathy" Update: https://neuralink.com/updates/a-year-of-telepathy/

Leave us a comment or a suggestion!

Support the show

Contact Jessica or Kimberly on LinkedIn:

• Jessica's LinkedIn
• Kimberly's LinkedIn

What happens when you automate away a six-hour task? You don't get more free time ... you just do more work.

In this impromptu conversation, Kimberly and Jessica break down what agentic AI actually does, why the "time savings" narrative misses the point entirely, and how to figure out which workflows are worth automating.

WHAT WE COVER:

• What agentic AI actually is (and how it's different from ChatGPT)
• Jessica's real invoice automation workflow: how she turned 6 hours of manual work into an AI agent task
• The framework for identifying automatable workflows (repetitive, skill-free, multi-step tasks)
• Why this beats creative AI work: no judgment calls, just execution
• The Blackboard experiment: what happens when an agent does something you didn't ask it to do
• Security & trust: passwords, login credentials, and where your data actually goes
• Enterprise-level agent solutions (and why they're not quite ready yet)
• The uncomfortable truth: freed-up time doesn't mean fewer hours—it means more output
• How detailed instruction manuals prepared Jessica for prompt engineering
• The human bottleneck: why your whole organization has to move at the same speed
• Why marketing and research are next on the chopping block

TOOLS MENTIONED:

• ChatGPT Pro with Agents — https://openai.com/chatgpt/
• Perplexity Comet (agentic browser) — https://www.perplexity.ai/comet
• Zoho Billing — https://www.zoho.com/billing/
• Constant Contact — https://www.constantcontact.com
• Zapier — https://zapier.com
• Elicit (systematic reviews & literature analysis) — https://elicit.com
• Corpus of Contemporary American English — https://www.english-corpora.org/coca/
• Descript — https://www.descript.com
• Canva — https://www.canva.com
• Riverside.fm — https://riverside.fm

TIMESTAMPS:

• 0:00 — Opening & guest cancellation
• 1:18 — Podcast website & jingle development (and why music taste is complicated)
• 6:34 — What is agentic AI? Jessica's invoice automation example
• 10:33 — Why this use case actually works
• 14:15 — The Blackboard incident (when the agent went off-script)
• 16:21 — Security concerns: passwords, login credentials, and trust
• 18:35 — Why speed doesn't matter (as long as it's faster than human bottleneck)
• 19:27 — Enterprise solutions on the horizon
• 20:57 — United Airlines cease-and-desist letters for replica training sites
• 22:27 — Why Kimberly can't use agents in her CCRC work
• 25:21 — How to identify your automatable workflows (the practical framework)
• 27:57 — Research automation with Elicit & corpus linguistics
• 30:45 — The core insight: AI shifts time, it doesn't save it
• 34:10 — Organizational bottlenecks & human capacity limits
• 35:08 — Pit & Peach (staying in your own canoe)

Leave us a comment or a suggestion!

Support the show

Contact Jessica or Kimberly on LinkedIn:

• Jessica's LinkedIn
• Kimberly's LinkedIn