Preface. When it comes to artificial intelligence, most articles assume it will happen, so discussions range around when and how it will happen. Often speculation that a general AI may use its ability to find patterns in data will allow it to bootstrap itself to consciousness.

And then it will take over! After all, Elon Musk, Stephen Hawking and 1.5 million google hits on “Artificial Intelligence existential threat” say so! Elon Musk says adopting AI is like “summoning the demon” because AI will start a world war and they will dominate the world as deathless authoritarians.  Prof Stephen Hawking said thinking machines pose a threat to our very existence because we may not be able to control them, and spell the end of the human race (Cellan-Jones 2014). Putin predicts that the leader in AI technology will be come the ruler of the world, and 51 other experts predict AI will go out of control in various ways (CBINSIGHTS 2019).

Nick Bostrom, at the University of Oxford has written a book about how this new super-intelligence could become extremely powerful, perhaps beyond our control, and consequently our own fate as a species might depend on what decisions the machine super intelligence makes (Bostrom 2014).

Computer scientist Stuart Russel at the University of California, Berkeley, circulated a letter signed by AI researchers at google, Facebook, Microsoft and scientists around the world demanding that only research keeping AI beneficial be funded lest it get out of control and threaten us (Wolchover 2015).

So many sci-fi movies and TV have AI robots like us that it seems as if AI could happen any day now. After all, we’ve been familiar with the idea since the early 1950s when the first robot stories appeared.

AI sells products at a higher price, but it is not intelligent, it is just software code

Artificial Intelligence will never be intelligent. It is just software. There is no Wizard of Oz, just a bunch of programmers behind the curtain. And programmers write bad code. It happens for many reasons: Because everyone makes mistakes, some programmers aren’t very good at it, it is impossible to test code fully, new code breaks older code, or the specifications themselves were incorrect or missing business rules. I know that AI is just software code because for my 25-year career I designed and added new features to computer systems for health care, banking, and transportation, coding in assembly language, COBOL, C, C++, java, Powerbuilder, Model204 and other languages. Now and then I had to back my code out, sometimes I’d made a mistake, or my test data wasn’t extensive enough to find the few exceptions that would break the code. Often a bug was not my fault, but I’d be called at 2 am anyhow, only to discover it was a new upgrade of Oracle, Unix, and other systems of software and hardware.  Like pirates, once you buy another companies products, you pay tribute every year to them as they upgrade it, yet can’t shut it down lest they stop supporting older versions.

AI is just code. Oh sure, it can find patterns. If you give it a million images of dogs, and it was programmed to find dog patterns, it will usually identify a dog in a photo after many hours of computer time and electricity.  But not a cat, a house or truck. That will require many more hours of computer time and electric energy.

Many promised AI miracles may never appear.  For example, it is highly unlikely we will transition to self-driving cars (Friedemann 2020).  I’m all for the drivers assistance with lane changes and emergency braking.  By self-driving I mean cars that let you read a book, sleep — pay no attention at all.   And what a disastrous waste of energy. Studies have shown that people would drive even more with cars far less energy efficient than mass transit, clogging roads with traffic (Mervis 2017, Taiebat et al 2019).

AI will never be intelligent because it can’t match the human brain

AI can never come close to the human brain, because the coding would take hundreds of trillions of lines of code inevitably riddled with trillions of errors, because the human cortex is 600 billion times more complicated than any artificial network (Kasan 2011).

Or even an insect brain. A factory robot is no smarter than the cockroach running around on the floor below. “Today’s state-of-the-art computers process roughly as many instructions per second as an insect brain,” and they lack the ability to effectively scale.” (Kendall 2020).

AI proponents insist that AI can catch up to the human brain, because our brains are also digital.

Not true. The latest science reveals that the human brain is highly analog, with dynamic synapses that “speak in a range of whispers and shouts.” Electric spikes are delivered as analog signals whose shape impacts the magnitude of chemical neurotransmitter released across the synapses, similar to a light dimmer with variable settings. For many years these spikes were thought to be delivered digitally, like an on and off light switch.  This gives our brains tremendous supercomputer level capabilities using the energy equivalent of a refrigerator light bulb (Chao et al 2020).   The brain is so powerful and compact it can fit on your shoulders, while a modern supercomputer can take up space the size of three tennis courts (Schranghamer et al 2020)

AI and other digital systems use a tremendous amount of electricity

Just one bitcoin requires 9 years’ worth of the $12,500 electricity used in a typical home. A year of bitcoins uses more electricity than Finland, a nation of 5.5 million consumers — half a percent of all electricity consumed in the world (Huang et al 2021).

Similarly, training an AI model generates as much carbon emissions as it takes to build and drive five cars over their lifetimes (Saenko 2020). The MegatronLM language model used as much energy as three homes in a year and other AI systems even more energy (Labbe 2021).  AlphaZero, Google’s Go- and chess-playing AI system, generated 192,000 pounds of CO2 during training. John Cohn, IBM Fellow and research scientist with the MIT-IBM Watson AI Lab said that when you look at how fast AI is growing, you can see we are heading in an unsustainable direction (Dickson 2020).

To program a robotic hand to manipulate a Rubik’s cube required 1,000 desktop computers plus a dozen machines with special graphics chips for several months, consuming about 2.8 gigawatt hours of electricity, the output of three nuclear power plants for an hour.  Machine-learning algorithms consume more and more energy and data while training longer and longer (Knight 2020).

AI learning requires massive amounts of data. AlphaZero used an exabyte of data. It would take 1.5 billion CD-ROM discs to contain an exabyte, which could store nearly 11 million movies in 4K format (Fisher 2021). AI is constantly trolling Big Data to analyze how businesses can make more money and there’s lots of data to crunch through –Walmart collects 2.5 petabytes of data from 1 million customers every hour.

Image recognition training requires huge amounts of data, it took 1.2 million images to train AI to recognize 1,000 objects, while a child can learn to recognize a new kind of object or animal with just one example (Simonite 2016).

And tremendous amounts of energy to crunch through data and images.

More good news: oil and coal are essential for making robots and AI and they’re declining

In my books “When Trucks Stop Running” (Friedemann 2016) and “Life After Fossil Fuels” (2021) I use peer-reviewed citations to explain why transportation and manufacturing are showstoppers for so-called renewables.  Basically essential transportation, the trucks, locomotives, and ships that run on the diesel fraction of a crude barrel of oil (about 15% of it) can’t be electrified or run on hydrogen or anything else.  And manufacturing also requires the very high heat of fossil fuels, there are no electric or hydrogen commercial processes now to make iron, steel (arc-furnaces melt existing steel), ceramics, glass, silicon chips, bricks and more.  Over 90% of the petroleum we use is conventional, and that peaked in 2008, and all world oil production including unconventional oil probably in 2018.  And I make the case that the electric grid itself can’t stay up without natural gas.

If I’m right, then robots and AI cannot make themselves or repair themselves. They cannot reproduce. The electric grid will fail for good when natural gas is scarce or wars destroy NG power plants. AI, robots, the electric grid, wind turbines, solar panels, and anything with cement or steel require fossil fuels for every single step of their life cycle, from mining to manufacturing to transportation of their parts to an assembly factory from all over the globe and to their final destination. iPhones require 75 of the 118 elements in the periodic table, many of them rare, many of them sourced only from China (Stone 2019).

Meanwhile the cost and time to create neural networks is calling into question whether AI can continue to scale up. It costs millions of dollars to train just one model, and supercomputers aided by dozens of expensive servers and graphical processing units. After that, each query requires dozens of these expensive machines (Sparkes 2021).

How AI could harm us

A very likely way AI will harm us is taking down the electric grid.  Algorithms in artificial intelligence, are doubling their power use every two months. Another application of conventional semiconductors, Bitcoin mining, saw a tenfold increase in semiconductor energy use in 10 years and as of August 2021, its estimated annual electricity used (91 TWh/yr) is more than the annual energy use of Finland. Without a strong energy efficiency focus, conventional semiconductors’ energy use may continue to double every three years or faster while energy production only increases at 2-3% per year. Computational energy demand is rising exponentially while the world’s energy production is increasing linearly (DOE 2022).

Alice Friedemann  www.energyskeptic.com  Author of Life After Fossil Fuels: A Reality Check on Alternative Energy; When Trucks Stop Running: Energy and the Future of Transportation”, Barriers to Making Algal Biofuels, & “Crunch! Whole Grain Artisan Chips and Crackers”.  Women in ecology  Podcasts: WGBH, Financial Sense, Jore, Planet: Critical, Crazy Town, Collapse Chronicles, Derrick Jensen, Practical Prepping, Kunstler 253 &278, Peak Prosperity,  Index of best energyskeptic posts

***

References

Bostrom N (2014) Superintelligence: Paths, Dangers, Strategies. Oxford University Press.

CBINSIGHTS (2019) How AI will go out of control according to 52 experts. Research Briefs, CBINSIGHTS. https://www.cbinsights.com/research/ai-threatens-humanity-expert-quotes/

Cellan-Jones R (2014) Stephen Hawking warns artificial intelligence could end mankind. BBC.  https://www.bbc.com/news/technology-30290540

Chao IH et al (2020) The potassium channel subunit Kvβ1 serves as a major control point for synaptic facilitation. Proceedings of the National Academy of Sciences;  DOI: 10.1073/pnas.2000790117

Dickson B (2020) AI Could Save the World, If It Doesn’t Ruin the Environment First. As AI usage grows, its energy consumption and carbon emissions are becoming an environmental concern. PcMag. https://www.pcmag.com/news/ai-could-save-the-world-if-it-doesnt-ruin-the-environment-first

DOE (2022) Semiconductor Supply Chain deep dive assessment. U.S. Department of Energy Response to Executive Order 14017, “America’s Supply Chains”.

Fisher T (2021) Terabytes, Gigabytes, & Petabytes: How Big Are They? Lifewire.com

Friedemann A (2016) When Trucks Stop Running: Energy and the Future of Transportation. Springer

Friedemann A (2020) Why self-driving cars may not be in your future. Energyskeptic.com

Friedemann A (2021) Life After Fossil Fuels: A Reality Check on Alternative Energy. Springer

Hall CAS, Klitgaard K (2018) Energy and the Wealth of Nations: An Introduction to Biophysical Economics. Springer.

Huang J et al (2021) Bitcoin Uses More Electricity Than Many Countries. How Is That Possible?. New York Times.  https://www.nytimes.com/interactive/2021/09/03/climate/bitcoin-carbon-footprint-electricity.html

Kasan P (2011) A.I. Gone awry: the future quest for artificial intelligence. Skeptic.

Kendall JD, Kumar S (2020) The building blocks of a brain-inspired computer. Applied Physics Reviews, DOI: 10.1063/1.5129306

Knight W (2020) AI Can Do Great Things—if It Doesn’t Burn the Planet. Wired.  https://www.wired.com/story/ai-great-things-burn-planet/

Labbe M (2021) Energy consumption of AI poses environmental problems. SearchEnterpriseAI. https://searchenterpriseai.techtarget.com/feature/Energy-consumption-of-AI-poses-environmental-problems

Mervis, J. December 15, 2017. Not so fast. We can’t even agree on what autonomous, much less how they will affect our lives. Science.

Murphy TW (2021) Energy and Human Ambitions on a Finite Planet. eScholarship. https://open.umn.edu/opentextbooks/textbooks/980

Saenko K (2020) It takes a lot of energy for machines to learn – here’s why AI is so power-hungry. The Conversation. https://theconversation.com/it-takes-a-lot-of-energy-for-machines-to-learn-heres-why-ai-is-so-power-hungry-151825

Schranghamer T.F. et al (2020) Graphene memristive synapses for high precision neuromorphic computing. Nature Communications, 2020; 11 (1) DOI: 10.1038/s41467-020-19203-z

Simonite T (2016) Algorithms That Learn with Less Data Could Expand AI’s Power. Technologyreview.com

Sparkes M (2021) Largest ever AI suggests limits to scaling up. New Scientist.

Stone M (2019) Behind the Hype of Apple’s Plan to End Mining. Gizmodo.com

Taiebat, M., et al. 2019. Forecasting the Impact of Connected and Automated Vehicles on Energy Use: A Microeconomic Study of Induced Travel and Energy Rebound. Applied Energy247: 297

Wolchover N (2015) This Artificial Intelligence Pioneer Has a Few Concerns. Increasingly rapid advances in AI have given Stuart Russell’s concerns heightened urgency. Wired. https://www.wired.com/2015/05/artificial-intelligence-pioneer-concerns/