May 14, 2025 · 1h 0m · big-technology
AI’s Drawbacks: Environmental Damage, Bad Benchmarks, Outsourcing Thinking
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of the Big Technology Podcast, host Alex Kantrowitz interviews 'The AI Con' authors Emily M. Bender and Alex Hanna to critically examine how artificial intelligence hype conceals severe environmental costs, labor exploitation, flawed evaluation benchmarks, and the degradation of human judgment.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 27.9% of the talking time here. How this is scored →
Alex as informed peer 5.5 Guest teaching 5.3 Guest disagreement 5.0 Alex pushing back 4.6
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 45:44 Sharply trading jabs over Bill Gates quote
Hanna emphatically interrupts and rejects the host's claim that quoting tech titans is fringe, insisting Gates represents the tech leadership driving the push.
Hardest push from Alex ▶ 45:37 Host rejects framing around extreme industry quotesKantrowitz explicitly refuses the guest's framing, arguing that citing Bill Gates's extreme statements distorts the nuanced consensus regarding workflow assistance.
Biggest teaching moment ▶ 49:23 Hanna details the cognitive burden of transcription auditingHanna draws on firsthand experience as an EMT and qualitative researcher to educate the host on why correcting machine errors is often slower and more error-prone than manual drafting.
Alex holds their own ▶ 13:00 Host brings data center and inference cost figuresKantrowitz demonstrates command of industry reporting by quoting Nvidia CEO Jensen Huang and executive discussions about the massive scaling requirements of inference compute.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| Examining AI's Environmental Costs and Lack of Transparency | 5 | 6 | 4 | 4 | Kantrowitz opens by asking the guests to quantify AI's energy and resource costs. When the guests cite data center water usage in Oregon, the host presses on whether those examples predate generative AI, prompting the guests to cite newer reporting on xAI's Memphis data center. | |
| Corporate Climate Backtracking and Inference Scaling | 6 | 5 | 4 | 4 | The host points out industry arguments around efficiency gains and inference scaling, citing Nvidia CEO Jensen Huang's estimate that inference will require dramatically more compute. The guests highlight corporate backtracking on climate commitments by Microsoft and Google. | |
| The Problem of Benchmark Gaming and Construct Validity | 6 | 6 | 4 | 3 | Bender and Hanna break down construct validity and critique benchmarks like Med-PaLM on medical exams. Kantrowitz engages constructively by bringing up the independent ARC-AGI benchmark and François Chollet's methodology. | |
| The Myth of General Intelligence and Unscoped Systems | 5 | 6 | 5 | 4 | Bender argues general intelligence is an unscoped engineering fallacy while Hanna draws historical analogies to chess as a narrow test. Kantrowitz pushes back with practical user experience, arguing models display cross-disciplinary utility. | |
| Deconstructing Synthetic Text and Information Retrieval | 6 | 5 | 6 | 6 | Bender dismisses LLMs as synthetic text extruders unsuitable for genuine information retrieval. Kantrowitz pushes back firmly using his firsthand experience in Paris using ChatGPT to synthesize friend recommendations and discover local cultural events. | |
| Evaluating Efficiency, Human Judgment, and Accountability | 7 | 6 | 7 | 7 | Bender critiques the Paris example by emphasizing missed human community connections and arguing bots lack judgment or accountability. Kantrowitz counters that newspapers also exercise biased editorial judgment and that LLM efficiency frees up real-world time. | |
| The Risks of Ambient AI Scribes in Healthcare | 6 | 7 | 6 | 5 | Kantrowitz suggests AI transcription scribes can alleviate severe administrative burdens for doctors. Bender counters that note synthesis is intrinsic to medical reflection and diagnostic care, also raising privacy and dialect bias issues. | |
| Technical Inaccuracies and Labor Casualization in Medical AI | 5 | 6 | 7 | 6 | Hanna discusses Whisper speech recognition errors, labor casualization, and quotes Bill Gates predicting the displacement of doctors. Kantrowitz pushes back, accusing Hanna of cherry-picking extreme quotes rather than addressing standard workflow use cases. | |
| The Realities of Verification and Administrative Burden | 6 | 7 | 6 | 6 | Kantrowitz maintains doctors could verify generated notes faster than writing them from scratch. Hanna counters with personal experience as an EMT and qualitative researcher, explaining that auditing flawed transcripts often takes more cognitive effort and time than drafting original notes. | |
| Bottom-Up Tool Adoption, Technical Debt, and Work Pressures | 5 | 5 | 5 | 4 | Kantrowitz raises bottom-up adoption where individual employees choose AI tools to ease their workload. Bender and Hanna argue this creates hidden technical debt in software and absurd synthetic email feedback loops. | |
| Critiquing AI Doomerism and Long-Termist Distractions | 6 | 5 | 6 | 6 | The host aligns with the guests on dismissing existential doomerism but pushes back on ignoring biosecurity threats and synthetic pathogen synthesis. Bender dismisses bio-risk scenarios as speculative science fiction that distracts from current material harms. | |
| Episode Conclusion and Parting Perspectives | 3 | 0 | 0 | 0 | Kantrowitz concludes the episode, thanking the authors, summarizing the spirit of balanced debate, and delivering housekeeping sign-offs. |
Statements from this episode (20)
Bender says large language models are parlor tricks exploiting human perception
“Right down at the bottom, you've got the fact that especially large language models are a technology that is, that's a parlor trick. It plays on our ability to make sense of language and makes it very easy to believe there's a thinking entity inside of there.”
Bender says AI search queries use 30 to 60 times more compute
“Each of those tokens has to be calculated individually, and so it's coming out one word at a time, and that is far more expensive. I think her number is somewhere between 30 and 60 times more expensive, just in terms of the compute, which then scales up for el…”
Hanna says xAI uses methane generators to power its Memphis supercomputer
“I would also say that speaking about existing effects, there's also a lot of reporting coming out of Memphis right now, especially around the methane generators that XAI has been using to power a particular supercomputer there called Colossus there, and specif…”
Hanna: Google data center consumed half the water in The Dalles, Oregon
“That is kind of pre AI in which we're focusing on the development of Google's hyperscaling And it wasn't until the Oregonians sued the city that we knew that half of the water consumption in the city was going to Google's data center.”
Bender says Microsoft abandoned net-zero carbon goals because of generative AI
“I would say that we've got Brad Smith at Microsoft giving up on the plans to become a net zero carbon since the beginning of Microsoft, and he said this ridiculous thing about we had a moonshot to get there, and turns out with generative AI, the moon is five t…”
Kantrowitz notes Nvidia CEO says AI inference needs 100x more compute
“And I mean, we have Jensen Wang, the CEO of Nvidia saying inference is going to take a hundred times more compute than traditional LLM inference.”
Bender says most AI benchmarks fail to measure actual capabilities
“Most of the benchmarks that are out there are not reasonable. They lack what's called construct validity, and construct validity is this two-part test of the thing that we are trying to measure is a real thing, and this measurement correlates with it interesti…”
Hanna argues USMLE scores do not prove Med-PaLM is ready for medicine
“When we get something where you have something like MedPalm II or MedPalm I and II being measured on the US medical licensing exam, that's not really a test that determines whether one is sufficient, you know, is prepared to be a medical practitioner. There's …”
Bender: Building general AI is not sound engineering practice
“So the ability to be general and here I'm thinking of the work of Dr. Tamik Gibru is not an engineering practice. That's an unscoped system. So what Dr. Gibru says is the first step in engineering is your specifications. What is it that you're building? If wha…”
Bender: ChatGPT only mimics language form and fails genuine information needs
“What ChatGPT can do is it can mimic human language use across many different domains. And so it can produce the form of a poem. It can produce the form of a travel itinerary. It can produce the form of a Wikipedia page on the history of some event. It is an ex…”
Bender: AI Bots Cannot Have Judgment or Accountability
“A bot is not the kind of thing that can have judgment, nor is it the kind of thing that can have accountability for exercising judgment.”
Hanna: People Lack Knowledge to Verify Chatbots Outside Their Specialties
“Once it gets into those areas in which verifiability just escapes me, which is most areas, because we're not professionals in most areas, and although a lot of us want to be jacks of all trades, jacks and jills of all trades, then we lose that ability, and we …”
Kantrowitz: AI's top use is companionship and therapy per recent study
“And we know now that AI is the number one use according to a recent study is companionship and therapy, and the therapy side really scares me, and I think the companionship isn't the best thing in the world either.”
Bender argues clinical note-taking is essential medical thinking, not clerical work
“Writing the clinical note is actually part of the process of care. It is the doctor reflecting on what came out of that conversation with the patient and thinking it through, writing it down, plans for next treatment. That is not something that I want doctors …”
Bender: Speech-to-text systems perform unequally across language varieties
“Thirdly, you've got the fact that automatic transcription systems, which are the first step in this, do not work equally well for different language varieties.”
Hanna: AI efficiency gains will increase doctor appointment pressure, not ease workloads
“And these efficiency gains from doctors isn't going to, like, Make their jobs necessarily easier. It's going to put more of a pressure on them. Not that you're in a position where you don't have to take medical notes. You're going to be running from position t…”
Hanna: Verifying AI transcription is not easier than transcribing directly
“Is verification an easier task than transcription? I guess that's my question. I would proffer no.”
Bender: AI coding assistants accumulate undocumented technical debt
“As coding assessments, assistants, so that's sort of a machine translation problem between natural language and some programming language and there I really worry about technical debt where you have, you know, output code that was not written by a person that'…”
Hanna: Employers cut jobs overestimating AI proficiency
“And I think we've already seen such a justification for this as being a place where employers have been Reducing positions by the scores, because there's a notion that these tools can do these jobs suitably and to a certain kind of degree of proficiency, which…”
Bender argues AI bioweapon risks are speculative fiction distracting from real harms
“Bad actors could use it to, you know, more quickly design, you know viruses and stuff like that. That's still speculative, right? So anytime we are taking the focus away, it's like, has that happened? Right? This is still people writing science fiction, fan fi…”