The tech-industry bestowed name, “Artificial Intelligence (AI)”, is a head-fake; there is no intelligence, just algorithms. Sales are based more on fear of missing out than efficiently-usable machines. The authors of “The AI Con: How to Fight Big Tech’s Hype and Create the Future We Want” have some tongue in cheek renaming suggestions: ‘“mathy maths’, ‘a racist pile of linear algebra’, ‘stochastic parrots (referring to large language models specifically)’ or Systematic Approaches to Learning Algorithms and Machine Inferences (aka SALAMI)”. (Page 5) These witty writers are Dr. Emily Bender, professor of linguistics at the University of Washington, and Dr. Alex Hanna, director of research at the Distributed AI Research Institute and lecturer in the school of Information at the University of California, Berkeley.
While working in Silicon Valley in the 1990s, I wrote quite a bit of code automating friction testing in hard drives. The maximum forces occurred when drives started up. About fifty test drives with sensitive gauges were used to test 50,000 or more start-stops. Once the testers were setup, it was all automated with friction data being stored in files and when the test finished, the files would automatically be uploaded to a database which would graph the data and create a report. I thought it was really cool but the only intelligence involved was mine and the creators of the database. It was a set of algorithms and nothing more; that is all these “racist piles of linear algebra” are.
Large Language Models (LLMs)
The texts produced by LLMs are plausible on almost any subject, but this is highly misleading. The models consist solely of extensive information about what sets of words are similar and what words are likely to appear in what context. The outputs look just like something a person might have written and we humans naturally interpret it by imagining the mind behind the text. Unfortunately, there is no mind and it is important for us to let go of that imaginary mind we conceive.
The authors label LLMs like ChatGPT “synthetic text extruding machines”. (Page 31) Like plastic extrusion, LLMs force language collections through complicated computer algorithms to achieve a product that looks like language. However, there is no human thinking behind it.
LLMs and their cousins, synthetic image machine, are based on massive data theft and wanton energy use. The backbones of synthetic extruding machines are data centers which consume enormous amounts of energy. It is estimated that they will consume 1,580 terawatt-hours a year by 2034. A terawatt hour is the equivalent of a billion kilowatt hours. That amount of energy is the same as the total amount of energy predicted to be consumed by the world’s most populous country, India. (Page 159)
In 2016, the largest tech companies in the world signed on to the Paris climate accords. Google said they planned to be net-zero emissions by 2030 and Microsoft announced plans to be net-negative and remove all of the carbon it had produced since its founding in 1975. Now the companies are admitting they will dramatically miss their climate pledges because of these “racist piles of linear algebra.” (Page 160)
In addition to being energy gluttons, text and image extruding machines are water hogs. For every 5 to 50 prompts ChatGPT generates, about two cups of water are consumed. The large amounts of PFAS (“forever chemicals”) used to manufacture microchips is another environmental issue. “Synthetic text extruding machines” are an environmental disaster, but for billionaires it is all about profits. (Page 157)
Bender and Hana observe, “Today’s synthetic media extruding machines are all based on data theft and labor exploitation, and enable some of the worst, most perverse incentives of each of these attendant fields.” (Page 135)
Boosters and Doomers
At the 2023 eighth Insight Forum, which was closed to reporters and the public, Senator Chuck Schumer asked the participants what was the probability of doom. It is unknown what the precise answers were but Jared Kaplan, co-founder of the AI company Anthropic, and Aleksander Madry, head of Open AI preparedness, have both spoken about “catastrophic risks” if a model grew a mind of its own. Some of the participants spoke about the myriad benefits while others seemed to harbor existential fear. Bender and Hana label these groups Boosters and Doomers. (Page 1 and Page 138)
From the authors’ perspective, Booster and Doomers are on two sides of the same coin. One sees extrusion machines as leading to a world of abundance while the other fears a dystopian hellscape. “Neither depicts the real harms of actually existing automation, at best dismissing them as less important than the imaginary existential threats.” (Page 139)
Oddly, almost all AI Doomers think AI development is a good thing. Bender and Hana suspect a few Doomers are not being honest:
“But for some of them, it’s not really about trying to save humanity, but rather a running of the con: the supposed danger of the systems is a splashy way to hype their power, with the goal of scoring big investments in their own AI ventures (like Musk and Altman) or funding for their research centers (Like Bourgon). (Page 141)
Modern Eugenics
There are claims that machines will gain an advanced level of “general intelligence”. However, there is not an accepted definition of “artificial general intelligence” (AGI). Companies like OpenAI just avoid the question. Microsoft’s “Sparks” paper contains a preliminary definition of AGI. A prior version of the paper was published in a 1994 Wall Street Journal article signed by 52-psychologists. It proffered, “The consensus group defined intelligence as a very general mental capability that, among other things, involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience.”
This was written in defense of Richard Herrnstein and Charles Murray’s 1994 book, “The Bell Curve”, which argues that there are significant differences between the inborn intelligence of different racial groups, and that those differences are due to genetics. A bastardized use of Alfred Binet’s work on intelligence testing was employed by three eugenicists, Henry Goddard, Lewis Terman and Robert Yerkes. They created tests biased towards middle-class white Americans and that bias persist in IQ testing to this day. (Page 35)
Bender and Hana state, The paradigm of describing ‘AI’ systems as having ‘humanlike intelligence’ or achieving greater-than-human ‘superintelligence’ rests on this same conception of ‘intelligence’ as a measurable quantity by which people (and machines) can be ranked.” (Page 36)
Billionaires—among them Elon Musk and Marc Andreessen—are setting the agenda for creating AGI and financially backing a modern-day eugenics. Musk repeats common eugenicist ideals claiming that there are not enough people and that humans (particularly the right humans) need to be having children at higher rates. Marc Andreessen echoed Musk’s thoughts when suggesting that elites from developed countries should be having more children. (Page 38)
Musk and Andreessen believe we are on the cusp of AGI development or are they just selling the con? Most people working on extruding machines are aiming to make a system that achieves what looks like human intelligence “to get ahead in what is already a crowded market.” (Page 39 and 40)
Today, there is no AGI; moreover it is unlikely that machines will ever develop “intelligence”.
Some Final Observations
“The AI Con” is packed with important information that could enable people to see through this billionaire-financed scam. Read it and convince your friends and family to study it as well.
Text and image extruding machines are not worth their costs to the environment and they have many hidden inefficiencies. It is wonderful that my smart-phone can assist me with texting, but I hate the AI driven enshitification of Google’s search engine.
Extruding machines are bad for education but people are out there hyping AI’s use in classrooms. The British Government has done serious harm to their health care system by mindlessly installing AI as a point of contact.
AI is not capable of doing science. A salient feature of extruding machines is they were designed to make stuff up.
Bender and Hana correctly note, “The AI project has always been more fantasy that reality.”