In just over a year since its launch in the USA, the book “The AI Empire”, by American journalist Karen Hao, has become one of the main references for those seeking to understand the era of generative artificial intelligence.
In the work, Hao tells the story of OpenAI’s rise based on hundreds of interviews and documents, which contain indecent details. The portrait that appears in the book is very different from the public image that the company strives to build.
Starting with the CEO of OpenAI himself, Sam Altman, who appears as someone who, in the eyes of important collaborators, would not be worthy of trust. Even the company’s business model that attracts billions in investment becomes a target. For Hao, the plans are not solid, and the company’s own employees admit this in a quiet manner.
The journalist’s conclusion is that AI companies are like the colonial empires of the past: driven by the concentration of power, exploitation of labor and natural resources, in addition to the search for political influence.
Visiting Brazil to launch the book in Portuguese, Hao says in this interview with the reporter that these imperial traits of AI companies have been deepening – including because of the alliance with the Donald Trump government, committed to promoting the military use of this technology.
The journalist defends forceful action by civil society, in the USA and the rest of the world, to pressure politicians and make it more difficult for these companies to do business.
*
QUESTION – The book shows how OpenAI went from a non-profit organization to a profit-driven structure. Now, the company will carry out its IPO. Is it the natural conclusion of the process you describe? With public capital, can we expect more transparency?
KAREN HAO – OpenAI is likely to become relatively more transparent than a privately held company. But does this mean it will be transparent in absolute terms? No.
The other day I was talking to a former Wall Street Journal colleague who covers the financial aspects of the AI industry. He commented that today there are many creative ways in which even publicly traded companies hide how they are really spending money on artificial intelligence.
And OpenAI is known precisely for creating a structure with multiple overlapping entities, which makes it difficult to understand how money circulates, who governs who and how the organization’s governance works. Therefore, I do not expect the IPO to suddenly solve any of these problems.
Q – Technology companies have deepened their alliance with the Donald Trump administration. What does this partnership represent?
KH – The main difference between the old empires and these companies was just the degree of violence they exercised. This difference is decreasing, because of the alliance with the State. AI companies are selling their technologies directly to the military for use in warfare.
Although there are small differences in the way each one approaches this issue, in the end they are all contributing to integrating their technologies into the state apparatus of violence.
The relationship between these companies and the Trump administration reminds me of that between the British East India Company and the British crown. They were two distinct entities, but they had a common agenda and sought to expand their power and influence throughout the world.
Today this alliance benefits Silicon Valley because it gives companies access to more spaces of power, more land and the possibility of exercising even more extreme forms of power.
This situation can be resolved, as long as there is very strong social mobilization and accountability mechanisms capable of putting pressure on this alliance between the State and companies.
Q – The war in the Gulf region has served as a laboratory for the use of AI applied to armed conflicts. What will be the effects of this war on this sector?
KH – The first consequence is that the war forced the industry to reveal its true position in relation to the military use of artificial intelligence much more clearly. It was clear that these companies are perfectly willing to support this type of operation.
The second consequence concerns financial risks. Because of the war, energy prices increased. All the projections they made – about when they would start generating profits – were based on certain assumptions about the price of energy and the cost of operating their data centers. These assumptions have completely changed and financial risk has increased significantly.
On the one hand, the war showed more clearly the damage that these companies can cause, both from a human and economic point of view. On the other hand, it revealed even more of its weaknesses.
Q – Opposition to AI companies has grown in the United States, with actions against data center projects in several communities. What will be the effects of this movement?
KH – The example I like to use is that OpenAI closed its video service. When Sora launched, the company touted it as its most important product since ChatGPT. The fact that the company ended this initiative is due, in my opinion, to three factors – all influenced by grassroots actions.
The first is the limitation of computational capacity. Protests against data centers are restricting the company’s ability to continue expanding its infrastructure and maintaining so many different products.
The second is uncertainty regarding the financial future. Wall Street is increasingly concerned about the political and social backlash against the AI industry and is beginning to doubt whether it will be able to deliver on the promises it made to investors.
As OpenAI prepares to go public, it also becomes more vulnerable to market sentiment.
The third factor is that consumer demand appears to have stagnated. People simply don’t want this type of technology as much as they thought. This is also a form of collective action.
We cannot count on public policymakers to do the right thing spontaneously. They will only act when there is a real threat to their positions. And this only happens through pressure exerted from below.
Q – You say that AI companies politically instrumentalize the concept of superintelligence and the supposed risk that China will develop it before the US. Is this narrative still convincing?
KH – I recently spoke to some policy makers in the United States, and they said that the argument “what about China?” lost a lot of steam in Washington. In part, this is because the AGI (artificial general intelligence) narrative is a myth. And myths thrive when there is an information void.
What happened is that this technology became part of people’s lives. When this happens, the myth loses strength, because there is now a concrete basis of facts and evidence about what this technology really is.
In the case of China, many policymakers came to feel manipulated. Over the last decade, companies have used the same argument during the expansion of social networks. They said: “Don’t regulate us, let’s build the world’s main platforms and strengthen democracies.” Exactly the opposite happened.
Then, with artificial intelligence, a new version of the same argument emerged: “Don’t regulate us. Instead, impose restrictions on China through export controls, so that American models dominate the market.”
But China responded by developing open source models precisely because of these restrictions. These models are more efficient, free and, in many cases, even American companies prefer to use them.
Q – You maintain that the current AI development model, based on unprecedented gains in scale, was not a technical decision. Can you explain this argument better?
KH – When I started covering artificial intelligence, in 2018, the area was heading in exactly the opposite direction. There was huge interest in what was called “tiny AI”. The focus was to develop systems capable of learning with very little data, using little computational capacity and consuming little energy.
I even wrote about systems that were trained directly on a cell phone. This alone shows that the bet on scale was a choice. There were already completely different ways of developing powerful models. The reason scale became dominant was the competitive dynamics between OpenAI, Anthropic, Google and other companies.
There are two ways to advance AI. One of them is to produce truly new research, make discoveries and develop new techniques. The other is to take techniques that already exist and simply apply more and more computational power to them.
Before OpenAI popularized this second strategy, most research followed the first path, because that’s where real scientific innovation happens.
The problem is that research takes time. When you run a company that competes with other companies, it is much more attractive to be able to increase the model’s capabilities in a predictable and constant way.
In this case, everything depends on how much money you have and how fast you can build bigger supercomputers. But this transfers the costs of developing AI to society as a whole.
Q – Today there is some skepticism regarding the business model of these companies. Will their growth be slowed by political opposition or economic limitations?
KH – Protests increase the risk of companies’ business. The more social mobilization makes it difficult for them to expand, the longer it takes them to execute their plans. And the longer they take, the more money they lose.
These companies do not have a business plan. I’m still in contact with people who work at OpenAI, and they acknowledge that they don’t have a business plan. They also know that the projections of the revenue needed to break even are extraordinary.
They themselves recognize that what they are trying to do is extremely ambitious. They think it could work, but they don’t say it’s likely. They just say that it is possible – as long as everything is executed practically perfectly.
And that is precisely the problem. With all this social mobilization, it has become much more difficult to imagine that everything will go perfectly.
*
X-RAY | KAREN HAO, 32
With a degree in mechanical engineering from the Massachusetts Institute of Technology, the journalist writes for publications such as The Atlantic and, last year, was named one of the 100 most influential people of the year by Time magazine. She was a former reporter at the Wall Street Journal and senior AI editor at MIT Technology Review. She also leads the Pulitzer Center’s AI Spotlight program, which trains journalists around the world to cover advances in AI.
THE AI EMERIUM
Price R$ 99.90 and R$ 47.40 (ebook)
Authored by Karen Hao
Rocco Publisher (440 pages)
Translated by Ananda Alves and André Sequeira
Source: www.noticiasaominuto.com.br
Source link
