The writer is a former senior adviser on AI at the Federal Trade Commission and managing director of the AI Now Institute. Amba Kak also contributed to this article
If AI is poised to occupy an increasingly central place in our digital infrastructure, it’s time to think long and hard about who will control it.
At present, Big Tech companies such as Microsoft, Google and Amazon are positioned to strengthen their foothold on the digital economy, consolidating their power by dominating both the commercial AI industry and the horizon for future AI research. Without the robust enforcement of competition laws, generative AI could irreversibly cement Big Tech’s advantage, giving a handful of companies power over technology that mediates much of our lives.
There are several reasons why, as things stand, there is no AI without Big Tech. The biggest technology companies have significant first-mover advantages in this market. Most notably, they have access to the resources large-scale AI is dependent on, from massive data sets to the computational power to process them, to the skills and expertise needed to build these AI systems.
These resource dependencies are a chokepoint even for companies such as Microsoft and Alphabet, Google’s parent company. For example, Alphabet recently combined its AI teams, forcing them to overcome intense internal rivalries, while Microsoft is limiting internal access to AI hardware to keep Bing’s GPT-4 chatbot and its new Office 365 tools up and running. Sam Altman, chief executive of ChatGPT-maker OpenAI, described his company’s computational costs as “eye-watering”. While new start-ups are appearing, OpenAI, Anthropic, Cohere and even the open-source company Hugging Face all have contracts with the big three hyperscalers.
Given this resource intensity, there will be significant pressure to leverage generative AI systems for profit. Here too, Big Tech companies are best positioned. They already operate digital ecosystems across which generative AI systems can be applied, and can maximise their dominance over platforms and markets.
OpenAI’s release of an app marketplace is an indicator that it intends to operate from the same playbook, by offering a product and operating a marketplace in which to compete with other companies. Amazon’s launch of its generative AI cloud service Bedrock is also a case in point: Amazon will both offer its own Titan generative AI models and operate a platform, tied to Amazon Web Services, on which companies can access other generative AI services. This structure means Amazon is well placed to secure its dominant position in the cloud computing market.
If anti-competitive conduct by Big Tech companies was a problem in the past, the introduction of generative AI is set to make things far worse.
That is why we need early and sure-footed enforcement of competition law to shape the direction of generative AI. This is an opportune moment for intervention: there is already a push for more muscular enforcement of the laws to address the concentration of power in Big Tech.
The US Federal Trade Commission has demonstrated an appetite for early intervention through its challenge to Meta’s takeover of the VR studio Within, indicating that it will be more aggressive in targeting future harms to competition before they materialise.
FTC chair Lina Khan has expressed concern about the lack of competition in AI, noting that in transitional moments like this one, incumbent companies often “panic” and attempt to block new entrants through unlawful tactics to protect their dominance. Support for this stance is reinforced by the White House, through its executive order outlining its intention of curbing industry consolidation.
Intervention is needed on several fronts. For one, companies must be held accountable for attempting to stave off competition — starting with Microsoft’s recent move to limit access to data for competitor chatbot-search engines. The resource dependencies in AI must also be addressed: regulators in the UK, Japan, the Netherlands, France and most recently the US have all identified concerns with the concentration in the cloud market. The emerging consensus among regulators about the dangers of cloud monopolies should galvanise structural interventions that anticipate future attempts at consolidation by these companies.
Generative AI could irreversibly cement the Big Tech advantage. But concentration in the tech industry emerged partly because lax regulators missed many opportunities to intervene. This time around, we should learn from past mistakes and act before the market is cornered. It’s now regulators, not companies, who need to move fast and break things up.