In the race to evolve AI, tech giants are copying each other and seem to have joined an unpredictable race which could jeopardise their core businesses, says Jacques-Aurélien Marcireau, co-head of equities at Edmond de Rothschild.
The race to multimodal generative AI and AI agents capable of carrying out complex tasks using interfaces similar to human intelligence is like a Prisoner’s Dilemma exercise which has gone wrong. By copying each other, tech giants now seem to have joined an unpredictable race which could jeopardise their core businesses while forcing them to invest eye-watering amounts. The big winners from this clash of titans will no doubt be other economic actors, whatever their category.
The ongoing battle between tech giants to decide who will win the generative AI contest over the long term is still begging several questions. Who was responsible for escalating hostilities? Will it be a Blitzkrieg or a war of attrition? And will there be a clear winner in the end?
If we are right in thinking this will be a war of attrition with no clear winner, the current situation looks in many ways like a Prisoner’s Dilemma exercise which has gone wrong.
Prisoner’s Dilemma
The Prisoner’s Dilemma is a famous game theory which shows why two people might not cooperate even if it is in their best interests to do so. Imagine two suspects who are being questioned after being arrested. If one testifies against the other, he/she will go free while the other suspect will get ten years in prison.
If each suspect testifies against the other, they both get five years. If neither speaks out, they only get two years each for a small offence. As they cannot communicate, the dilemma is the natural tendency for one suspect to betray the other in an attempt to minimise his/her own sentence. This results in an outcome that is worse for both and the paradigm explains situations where confidence and cooperation are crucial but difficult to attain, with an impact in areas like the economy, politics and sociology.
Google has adopted a much more measured approach to generative AI than its main rivals.
Google’s approach
At the inception of these new kinds of AI, seminal research at Google on Transformers – the fundamental technology behind generative AI’s development – was initially open source. The group boasted exceptional open-source experts and still does. Google has adopted a much more measured approach to generative AI than its main rivals. The question is: why? There are two possible reasons.
Firstly, the models’ training and running costs; and secondly, caution over the technology’s unforeseen short and long-term consequences (short term consequences include “hallucinations”, or incorrect and/or misleading results, and the emergence of “agents” is a longer-term issue).
Tech giants aim to go beyond the simple generation of images and text to multimodal AI capable of carrying out complex tasks using a continuum of applications (agents), just like human brains do every day. We should not underestimate the consequences from these new AI capacities: if, for example, we look at Google’s search engine, advertisers are willing to pay up to feature on the first page because they know that people will not persevere up to, say, page 15 to optimise their search. In contrast, an “agent type” AI should be able to scrutinise every page and find the best result.
In other words, internet giants are experts in making the most of human cognitive bias but will face new challenges when dealing with a machine.
As well as advertising, social networks are also in the firing line. So too is software as it all too often relies on human inertia and habits. The simple fact is that these new AI capacities have called the economic models of numerous tech companies into question.
Gigantic financial sinkhole
At the same time, nobody knows if and when such AI technology will emerge and how much investment will be needed along the way. The whole experiment could turn out to be a gigantic financial sinkhole.
Hence Google’s “reasonable” approach to the Prisoner’s Dilemma. But this tactic can only work if the other groups follow suit. Instead, Microsoft, Open.AI and now Meta and Amazon have opted for an unreasonable approach. As a result, we are now seeing an unprecedented wave of investment in graphics processing units and experts. Tech giant capex is now flirting with $100 billion a year. And all to develop a technology that could in part backfire on them.
If the analogy between electricity and AI turns out to be right, AI producers will end up benefiting less than AI users, while also having to deal with the stiff competition they themselves created:
| Google is reasonable | Google is unreasonable | |
| Meta is reasonable | The Nash Equilibrium (optimum) low investment & maximum margins | In time, Google has the means to dominate Facebook and reduce its competitive edge |
| Meta is unreasonable | In time, Meta manages to dominate Facebook and reduce its competitive edge | Maximum investment and margin destruction over the long term due to the agent impact on each group’s core business |
The main factor missing in this schema is the size of the potential market for the group which successfully develops this technology, thereby possibly offsetting the loss of the legacy business. For reasons of space, we have left out any political interference which could cap the winner’s gains, notably through a clause to make the business fall into the public domain after a certain lapse of time.
* Jacques-Aurélien Marcireau is co-head of equities at Edmond de Rothschild.
