“AI can strengthen digital giants, but also weaken them”

IWill artificial intelligence (AI) revolutionize digital cards or undermine the power of the sector’s major players? The success of software that can create, from a simple written command, texts – such as ChatGPT – or images – such as Midjourney – initially seemed to confirm the second hypothesis. If it were possible, these systems would be even more dominated by big technologies like Google, Microsoft, Meta (Facebook) or Amazon than by online research, social networks, software or e-commerce…

But in recent months, with the emergence of powerful open-source AI models, Therefore, an accessible, alternative thesis has been developed: AI can weaken the digital giants.

There is no shortage of elements to support the idea of ​​consolidating the power of dominant players: OpenAI, the start-up that created ChatGPT, has been firmly attached since 2019, in partnership with Microsoft, which was ready to invest 10 billion dollars (9.4 billion euros). Google has bet $300 million to take a stake in startup Anthropic. And, through their cloud hosting and services subsidiaries, Google is partnering with Cohere or C3 AI and Amazon with Stability AI. Digital giants bring here the computing power to train gigantic models (540 billion parameters for PaLM, from Google). This is a crucial and expensive resource, as the latest chips cost up to $40,000 each.

That stranglehold is already drawing criticism, especially from Elon Musk, for whom the creation of OpenAI was a way to keep AI out of Google’s hands. “Big tech’s dominance of the digital economy will close unless regulators step in.” warned Sarah Myers West of the NGO AI Now Institute Financial Times. The FTC, the US antitrust authority, said it was “vigilant”because “AI risks further consolidation of dominance of major digital players”.

future “hybrid”

Some fear that small businesses and public research will depend on a few big AI models because they depend on big platforms, social networks, mobile environments… Added to this is economic uncertainty – today these models cost fractions of pennies per query. but tomorrow? – as well as sovereignty issues…

But for others, the colossus of technology rested on feet of clay: “It’s an inconvenient truth, but we at Google are not in a position to win the AI ​​sprint race. And neither is OpenAI… open source is catching up with us, released a Google engineer in a leaked internal memo in early May semi-analysis. Our lead is melting at an incredible rate. Open source models are faster, more user-friendly, and more efficient. says the employee and points to it “No one will pay if there is a free alternative.” From now on, software, much smaller but powerful for specific tasks, can be trained on simple computers, even smartphones, at a fraction of the cost.