I have recently, watched the YouTube video, “How tech CEOs are lying to you”, where Karen Hao is interviewed by Aaron Bastani on Novara Media’s channel. Ms Hao has just published a book, “Empire of AI” and the interview covers the topics of the book.
At the centre of her arguments are, that large language modules, and their use of resources, water, power, land and rare metals are a choice, and one that society cannot afford. She questions the business model of the AI industry, sees it as a threat to [US] wealth and notes its scofflaw approach to its own regulation and its oppression of poor and vulnerable communities. She is highly critical of the motivations of the oligarchs funding the AI bubble.
What follows is my notes of the interview. I have rearranged the order to try and tell the story without repetition or diversion.
She argues that at the centre, the industry’s self serving ideology , the messianic commitment of its protagonists, the questionable value of the industry, the public cost in land, water, electricity and talent, and the threat to democracy as they avoid and corrupt the regulators are all things we should be highly concerned about. She also argues the lack of business model and its failure to create wealth is a threat to the [US] economy. They talk about what is AI, how it came to be, OpenAI vs Google, the immense resources required to run the infrastructure and the conflicts with democratic accountability & legal regulation. She also highlights the export of oppression to countries with even weaker labour protection laws than the US, highlighting the maltreatment of workers in East Africa and Columbia who work on influencing the model training.
How we got here and the big choice
She argues that the term, AI was coined in 1956 to attract research funding. It now means many things partly because there is no consensus as to what human intelligence is. AI mostly means deep learning systems, which google gemini defines as, “A deep learning system is a type of artificial intelligence and machine learning that uses a complex architecture of artificial neural networks, inspired by the human brain, to learn from vast amounts of data and make decisions with minimal human intervention. It processes information through multiple layers to identify increasingly complex patterns, enabling tasks such as image and speech recognition, natural language processing, and making predictions in applications like self-driving cars and chatbots.” The Wikipedia page on Deep Learning, is also helpful.
10 years ago, people thought that incremental innovation of AI technology would be enough, Open AI was led by people who disagreed with this and decided that scale was the appropriate route to beating Google because Google had the talent and competing with that talent was too hard. Chat GPT was founded as a competitor to Google; they originally argued for the need for non-profit governance because of the existential danger to humanity if run by for-profits. The scale model was adopted as a competitive weapon and is not necessary. The choice of scale meant that the bottleneck ceased to be talent and was to become [working] capital. They were founded as a not for profit but needed to become a for-profit in order to attract the capital that scale required. There was also at this time, a choice between Altman & Musk as CEO; the talent in AI chose Altman.
Altman came from the world of venture capital (VC), and was seen as an investor not a CEO. He also cultivated politicians while at Y Combinator (YC). Bastani asks what he does, despite his perceived personal shortcomings. Hao says that he is great at talent acquisition and motivation. He is good at getting people to give him what he needs, talent, resources, capital, land, water and laws. He operates best in one-to-one meetings, which leads to a dichotomy of views on him.
They explored some alternatives, and some earlier competitors to Open AI. One of these is the Chinese developed Deepseek, which Hao explains is a simplification of the scaling models. She also states that Open AI and US Big Tech won’t adopt it because of the barriers to entry that the need for large scale funding creates together with the sunk investment i.e. what they’ve already spent. In passing she mentions, Stable Diffusion which came to market before Open AI and was invented in Europe. It has been eclipsed by several other players and I wonder if this is yet another example of proving that Europe can turn money into ideas, but not ideas into money. (British capitalism is lazy and not interested in innovation or British welfare; it doesn’t care where profit, interest or fees come from.)
Does it do anything worthwhile?
They turned to consider what the AI business model is and what it does in terms of participating in a productive economy. They ask how much AI do we need anyway and what are we using it for? It would seem that we are mostly generating memes or images. While I do use it for this purpose, I have to ask if it’s not really overly expensive for making decorations for my blog.
The nature of monopoly in within the internet services industries has led to what Cory Doctorow calls enshittification, see https://pluralistic.net/tag/enshittification/ where he eloquently shows how Amazon fails to serve its supply chain or its customers and is a notoriously bad employer. Others have documented both Facebook and Googles paths journeying to this end point as these companies prioritise profit over customer service.
Returning to the issue of the Open AI business model, Hao argues that, it’s not just the money. There is a belief in the need for a true AI. This creates pressure on the legacy players. As a hedge bet against the dream of creating Artificial General Intelligence (AGI), Open AI and its chatbot competitors are offering labour substitution as a value proposition. She is doubtful the extent to which it will succeed, but there is a bandwagon effect attracting capital. If everyone’s wrong, then the amount of capital invested in AI is large enough to drag the US economy down and we’ve seen the effect of such a bubble bursting.
The exorbitant cost
It is common knowledge that there is an immense consumption of resources involved, on power, Hao says, that McKinsey predict +50% to +120% of UK power capability needed globally. (I don’t find this very accessible, and I need to find the source. ) Finding hard data is very hard, but in 2024, Microsoft and & Google increased their carbon production by 30%, and as a result of Hao’s suggestions, I searched for and found two McKinsey reports The-role-of-power-in-unlocking-the-european-ai-revolution, and How-data-centers-and-the-energy-sector-can-sate-ais-hunger-for-power.
In order to meet this demand, the fossil fuel providers are extending the life of coal mines and coal burning plants. She says that, Musk’s XAI is powered by unlicensed methane burning powerplants. This becomes more than a threat to the climate, it is also a direct threat to public health. Furthermore, Data Centres need fresh water to cool the computers. This is usually acquired using current infrastructure which is designed to support consumer consumption and was In the case of the UK built 100 years ago. It should be noted that much of the US power and water infrastructure is also old and poorly maintained. The Data Centres, thus compete with people for drinking water. There are examples from the US where drinking water is mixed with lower quality water which causes public health problems. This is an issue that needs to be addressed in the UK now that the UK-US Technology deal has been announced.
The data centers are located in poor communities, and she illustrates the effect by quoting examples from Arizona where the Data Centre’s compete for water, but in the UK we are used to hosepipe bans, and Hao states that there is now a ban on housing construction in the M4 corridor because of the data centres’ competing need for electricity. ( So while we used to talk about people, schools and health services as the necessary infrastructure, we have entered a world in which electricity and water also require planning, and the M4 corridor is not a poor area in England.)
Hao raised the issue that some of the inventors working in the UK, are Brits, educated and trained in the UK so the taxpayer investment in that migrating human capital is lost to the sterling economy. Some of them may seek to come back given the Trump administration’s policies on visas and entry to the USA.
Hao mentions two engagements by Open AI with the global labour market taking advantage of weaker employee protection laws and lower wages/cost of living expenses. She mentions Kenya and content filtering, which can be a traumatising experience, leading to PTSD and she has a hwrrowing story about its impact on family life. She also quotes a Venezuelan refugee suffering from depression due to the pressures of working on a work sharig platform and competing for tasks with too many others. Workers were made to bid for the jobs, and rates were very low. Hao speaks of the racism that allows some to earn millions and others to earn pennies. Maybe we’ll make a Marxist of her yet.
Democracy, Regulation and Lobbying
Their scofflaw approach to even basic regulation constraints is a threat to democracy. Like Karen, I have met some of these people and understand their contempt for regulation, I remember the surprise that Monopoly is an illegal business model expessed by some of the most senior business executives and the shock expressed by US colleagues when we explained the TUPE laws to them.
The interference in public decision making occurs at all levels, from Parish to International level and it’s why anti-lobbying laws are so important. We can see the effect of this in the coming fallout from the UK US Tech Trade deal.
They explore the impact of the State/Enterprise osmosis and Hao considers the colonisation of India via the East India company which she sees as an example in terms of power and style, emphasising in particular the permeability between private and state power. Today we call it buying laws, but with Trump the threat of force is always there. She’s not clear how the new state monopoly capitalism/empire will come to be; in the 18th century the UK nationalised the EIC, who knows what’ll happen next time. Further understanding requires the exploration of the relationship between the [US] state and private big tech. This metaphor reinforces Hao’s statement that the political/big tech complex see foreign countries and their peoples as ‘resources’, as does the treatment of workers from global south. She notes that the “Big Beautiful Bill” prohibits state regulation of AI.
What to do?
Peoples and states need to organise for data & compute sovereignty. The EU are seeding an EU stack which not only articulates the need for open and maintainable software products, but includes physical resources too, systems, networks and storage. She emphasises that colossal data centres are not needed; it’s Open AI that has made this necessary. She also calls out the impact of the fossil fuel lobbying.
There are examples of places and communities that fight back, she quotes that in Chile, artists are fighting for ownership of the training inputs and this is also a fight in the US & Europe. She says that in the EU privacy laws are an important constraint, and that the role of AI in education is a democratic battleground. It would be good to see more people fighting for priority in the use of land and water but the opportunity is there and will become more acute as the the UK Government seeks to roll out the data centres envisaged by the US/UK Tech deal.
I hope this article has enough of me in it to make it a worth while read, even if only through its structure. The interview has its transcript published at singjupost, for those who prefer to read & skim. I ran the interview at1:25 speed, at least the secondtime. A much shorter and more focused review is at Empire of AI by Karen Hao – book review by Ali Caterall, with a sub title of “A powerful, troubling exposé revealing tech giant OpenAI’s secretive rise, enormous resource demands, and internal struggles, while debating its wider impact on society and the environment”.