For decision-makers, the key issue is not whether to โdo AIโ, but what kind of AI portfolio the organisation is funding and who is accountable for its wider costs. A strategy built around ever-larger general-purpose models can quietly concentrate vendor dependence, cloud spend, data exposure and reputational risk in a small number of external providers. That makes AI less a technology choice than a governance choice: which capabilities belong in-house, which should be bought, and where scale is simply an expensive default.
Portfolio discipline matters. Leaders should separate use cases that justify broad platform investment from those better served by narrower, task-specific models. The practical question is not only expected ROI, but also data sensitivity, model performance requirements, energy use, compliance burden and the cost of ongoing human review. In many organisations, a smaller set of targeted deployments may deliver more value with less operational drag than a flagship enterprise AI programme.
The labour and ethics dimension also has management consequences. If model performance depends on hidden human moderation, outsourced labelling or poorly governed third parties, then procurement, supplier due diligence and workforce standards become part of AI assurance. CIOs and transformation leaders should ask whether contracts cover working conditions, content handling, audit rights and escalation paths for harms. If they do not, the organisation may be externalising risk rather than reducing it.
The next questions are strategic, not technical: what decision rights govern AI model selection, who signs off on acceptable use and exception cases, how will success be measured beyond usage growth, and what portfolio trade-offs will be made when a model is costly but not mission-critical? Organisations that cannot answer these questions are likely to inherit the politics, cost structure and accountability gaps of the broader AI arms race.
Extraction, power and the dark side of the race for AI dominance.
Amid AIโs promise to improve efficiency, enable new innovations and supercharge corporate profits, there is a dark side to the technology that canโt be ignored. Water-guzzling data centres are reportedly causing pollution andย water shortages in rural communities. Workers in emerging economies are paid less than US$2 an hour toย sift through toxic content in order to make chatbots safe for public consumption. And ChatGPT has been shown to give vulnerable teenagers harmful advice, including instructions on how to compose a suicide letter.
In a recent INSEAD Lifelong Learning webinar, Professor of Technology and Businessย Theodoros Evgeniou and INSEAD alumnusย Tim Gordonย (MBA โ00D) from Best Practice AIย were joined by author and award-winning investigative journalistย Karen Hao for a provocative conversation on the power dynamics, ethical blind spots and structural inequalities shaping AIโs future.
The battle to build an AI empire
Hao was the first journalist to profile OpenAI in aย 2020 article for MIT Technology Review, and the company features heavily in her new bookย Empire of AI. From unbridled Silicon Valley ambition to global exploitation, the book explores how an original mission for safety turned into a high-stakes race for dominance โ fuelled by vast data, invisible labour and mounting environmental costs.
โWe really need to think of companies like OpenAI as new forms of empire,โ Hao said. โIf you consider the sheer economic and political power of companies like OpenAI, Google, Meta and other tech companies that have now pivoted to aggressively trying to dominate AI development, you could argue that thereโs pretty much no nation-state government that is more powerful than these companies right now.โย
Hao drew four parallels between the empires of old and the tech companies of today: They lay claim to resources that are not their own; they exploit an extraordinary amount of labour; they monopolise knowledge production; and they engage in a competition narrative of an existential race, positioning themselves as โgoodโ empires on a mission to bring progress and modernity to the world.
By using their financial resources to exert influence over everything from politics to the media, Hao argued that tech companies have wrestled a controlling influence on the mechanisms used to regulate AI and shape how it is perceived by the public. They have also snapped up most of the worldโs top AI researchers, to the point where the actual science has been distorted โ much like how climate research would be skewed if most of the worldโs climate scientists were bankrolled by fossil fuel companies.
An extreme form of capitalism
Hao discussed how the current AI arms race represents an extreme form of capitalism โ one underpinned by a deeper ideological drive that has parallels with colonialism. From her perspective, these tech companies โpray at the altar of scaleโ and are trying to expand their AI models in a way that is not only unnecessary for what they are trying to achieve, but also doesnโt make business or financial sense.
โThey are burning an extraordinary amount of money and have not, in any way, developed a business model to generate the comparable amount of revenue that would allow them to return a profit,โ she said. โBut because they see AI as a narrative cloak around the ability to continue expanding to ever further reaches, this continues to perpetuate their hunger for building these AI systems and pouring more and more resources into it.โ
Pointing out that every industrial revolution has been grim for the people involved, Gordon asked Hao if the ongoing use of labour in emerging economies to train AI models should be viewed as exploitation, or if it could be a potential development pathway for these countries.
In response, Hao spoke about her fieldwork in Kenya, where she met with workers performing content moderation for what would become ChatGPT. They were tasked with labelling reams of disturbing content โ from violent and sexual content to hate speech โ so that OpenAI could develop a filter to prevent it from reaching ChatGPTโs eventual users. Yet, companies that have spent billions on AI development canโt even pay all these workers US$2 an hour.
โThe workers advocate for it to be a real dignified economic opportunity that provides skill-building, a career ladder, good, stable pay and healthcare benefits,โ Hao said. โBut the problem is that this is almost never actually the case.โ
General-purpose vs. task-specific AI
Evgeniou then brought up a commonly cited argument by AI believers: What if we just invest in developing and scaling the technology now, and then use its capabilities to address these negative externalities in the future?
Haoโs criticism of this perspective is that we simply may not have the planetary or human resources to waste. Perhaps most importantly, she expressed scepticism as to whether going all-in on scaling general-purpose AI models will lead to a technology that can solve all these problems.
We must not sleepwalk into a future we don’t likeโฆ and definitely not a future shaped by just a few, let’s say, self-dealing AI emperors.
Rather than general-purpose models, Hao believes that task-specific AI development is a smarter way forward. This tailored approach is not only less costly in terms of resources but also uses more curated data to develop high-performance models that address specific needs.
โ[Big Tech] companies are trying to convince us that every single person needs a rocket for everything, so that you should be using a rocket to commute from Paris to Madrid. But you should actually be using much more efficient, tailor-made, task-specific transportation modes to achieve that purpose,โ Hao said. โIn the same way, we should be moving all of the capital weโre pouring into this nebulous, general-purpose tool towards task-specific tools that actually tackle medical advancements, drug discovery, healthcare applications and climate change mitigation, because we’ve already seen plenty of evidence that those solutions work, and theyโre significantly less costly than this other approach.โ
Individual agency still matters
Evgeniou wrapped up the conversation by stressing that while AI may be here to stay, how itโs developed is not a foregone conclusion. โWe must not sleepwalk into a future we don’t likeโฆ and definitely not a future shaped by just a few, let’s say, self-dealing AI emperors,โ he said. โWe need to collectively consider alternatives and what a more balanced future could look like.โ
Hao echoed these sentiments and emphasised that we all have an active role to play in shaping AI development. From investing in alternatives to putting consumer pressure on Big Tech firms, everyone has the power to assert their own preferences on how they think AI should be developed and deployed in their lives.
โThe reason why I think itโs so important to understand that there is no inevitabilityโฆ is because then it opens up much more space for anyone to get involved in shaping the future of technology,โ Hao said. โThe shape of a technology is never inevitable โ it is the many human choices that determine how it ultimately looks and works.โย
For more on the subject,ย listen to this episodeย of The Age of Intelligence podcast hosted by Theodoros Evgeniou and Tim Gordon and featuring Karen Hao.
Edited by:
Rachel Eva Lim
About the author(s)
is a Senior Editor at INSEAD Knowledge.
-
https://knowledge.insead.edu/responsibility/hidden-forces-driving-ai-boom
Related
