"Empire of AI", by Karen Hao, tell the history of OpenAI, from its creation as a non-profit with the ambition of developing AI for the benefit of humanity, to the profit-making company behind ChatGPT; one of the most powerful actors in AI today.
Much of the book follows Sam Altman and the people around OpenAI: It describes their disagreements, changing business model and eventually the crisis in 2023, when Altman was fired by the board only to return a few days later. This history is interesting because OpenAI is not just another company: it is influencing how one of the most important technologies of our time is developing.
I felt that the book at times became too personal (some of the conflicts sound very similar to what can happen in any office…), and some of the biographical material unnecessary for the book’s point (in particular on Mr. Altman’s sister).
What I found much more interesting though, were the disagreements about what AI actually means and where it may take us. In fact, OpenAI itself was born out of these questions: If more powerful AI can transform society, who should control it? Should development be accelerated because of the enormous benefits it could bring, or slowed because we do not understand the risks? The disagreement between the AI “boomers” and “doomers” runs through much of the history of OpenAI and helps explain many of its internal conflicts. Thus, the best part of the book is when it looks beyond OpenAI itself; for instance when it discusses the enormous data centers needed to develop AI (including their consumption of energy and water, and the human labour behind AI, including workers in developing countries hired to label data and moderate disturbing content). The apparently immaterial world of AI actually depends on very physical infrastructure, resources and human labour, something that is too often lost in the wider discussion on the technology.
Ms. Hao’s main argument is that there is an emerging “Empire of AI”; a small number of powerful tech companies and individuals concentrating technology, capital and decision-making while exploiting resources, data and labour from across the world, just like any previous capitalist exploitation throughout history.
I was less convinced about this argument, as I think that it on the one hand ignores the complexity and speed of technological change, as well as simplifying the social dynamics of such change. I got a bit annoyed when she appeared to criticize people at OpenAI for not providing satisfactory answers about the consequences of AI, implying that they were avoiding difficult questions. I rather think there may be another explanation: they simply do not know the answers!
And that was indeed the most interesting conclusion I took from the book: The people developing AI are obviously intelligent, ambitious and visionary, but they are all still human beings! As Ms. Hao amply proves, they disagree, make mistakes, fight over office politics and interpret evidence differently, and in fact these people themselves do not really know how quickly it will develop, what its impact will be, what resources it will require, what it will do to employment and inequality, or what role it will eventually have in our societies.
I therefore think the book could have been much stronger if it had used the history of OpenAI more explicitly as a way to explore these questions.
Maybe my slightly uncomfortable conclusion from the book is that there appears to be nobody anywhere who fully understand what is happening with AI: Not OpenAI, not academics, not governments, not business... They have different ideas about what might happen, but nobody really knows where AI is going or how quickly we will get there...
Of course this does not mean that we should panic about AI, nor that we should simply leave its development to tech companies and billionaires. It rather means that we should talk much more seriously about what we want from this technology, what risks we are willing to accept, how its benefits and costs should be distributed, and what role we want it to have in our societies.
That is also why I would still recommend the book as one more in a canon of AI books; read together with other AI books it shows the very human organisation behind a technology that can otherwise seem almost abstract, and thus the questions we are all struggling with.
No comments:
Post a Comment