The “mutual destruction guaranteed” is a well -known concept in nuclear. It is now available in AI. A noticed, titled note “Surprintelligence Strategy “recently exposed what a similar derailment of AI would look like. If a state was trying to obtain unilateral domination in the sector, the others would be legitimate to prevent it by espionage or sabotage. This publication might seem like thriller if it was not the quality of his authors: Eric Schmidt, ex-president of Google, Alexandr Wang, founder of Scale Ai, one of the leaders of the datasets for the training of models, and Dan Hendrycks, advisor to Elon Musk XAI.
These three specialists start from the postulate that superintendent or act (general artificial intelligence), a much better AI than humans in all cognitive tasks, will arrive in a horizon of twenty-four to thirty-six months. The hypothesis is based on scale laws, the idea that more computing power, more data and larger models will inevitably open the way to greater progress in the power of artificial intelligences.
This is what pushed researchers, from 2017, to feed their models with ever larger volumes of data. Like Bookcorpus, a collection of 7,000 unpublished books to train GPT-1 for example. There are many reasons for not believing men of art: they clearly have an interest in making daring predictions because they raise capital, increase the enhancement of actions and are convinced, perhaps, of their own historical importance.
Land laws draw AI
But if, last fall, doubts still surrounded the progress that these laws of scale could bring, they were swept. The models continue to progress. At the end of December, the O3 system of Openai obtained a score of 85 % in the Arc-Agi reference testwell above the previous best score of an AI (55 %) and tied with the average human score. It is now clear that with the data centers being delivery, which contain a tenfold calculation power, the new models will reach general artificial intelligence.
Silicon Valley investors finance new companies only on the condition that they are “AG-PROOF”, in other words resistant to AC. The ex-artificial intelligence of the Biden administration, Ben Buchanan, explained Recently how the American government – the previous one but also the current – is preparing for this upheaval, especially in its national security mission, with the conviction that Cybernetic operations will revolutionize, intelligence and even the conduct of war with drones or autonomous robots.
Beyond national security, which rightly interests governments, what would be the impact for our societies? And more specifically, what would be the impact on the labor market and therefore the distribution of income and wealth? The risk of a lack of material resources is limited. The replacement of intelligent work will result in a relative increase in the value of physical capital. Factory, infrastructure, power plants will gain importance. The owners of these equipment will have an undeniable advantage in the face of those who are devoid of them. Public opinions will react to the mass disappearance of so -called qualified jobs, either by the refusal of these technologies – with the possibility that they are supplied by another more open country – or by the establishment of safety nets, the famous universal income. The wealth then residing in more easily taxable physical capital, these safety nets will find their funding. The real question posed by this model of society is due to the fact that things will remain frozen, without hope, in countries which will not master Act. The chances of progressing and growing will be extremely tenuous.
Measuring the capacity of the act to innovate is therefore crucial. If it accelerates the pace of technological progress and therefore improving the level of material satisfaction with fewer resources, it will propel human societies into an economy of abundance, succeeding that of the shortage (or post-scare). Concepts that passed recently for science fiction but which must now be combined with the present.
.