The 2020s have seen unprecedented acceleration in the sophistication of artificial intelligence, thanks to the rise of large language model (LLM) technology. These machines can perform a wide range of tasks once thought to be only solvable by humans: write stories, create art from text descriptions, and solve complex tasks and problems they were not trained to handle.
We posed this question to five AI experts in two parts including James Poulos, roon, Max Anton Brewer, Robin Hanson, and Anton Troynikov. —Eds.
1. What year do you predict, with 50 percent confidence, that a machine will have artificial general intelligence (AGI) – that is, when will it match or exceed most humans in every learning, reasoning, or intellectual domain?
2. What changes to society will this affect within five years of occurring?
AGI will arrive around 2050. It will mean the end of privacy and autonomy. In or around 2050, machine-learning systems will dominate human cognition in its ability to engage in generalized learning tasks.
Machine learning is already more capable of generating superior knowledge and reasoning than humans at directed tasks and with curated datasets. ML algorithms can today augment humanity to be better writers, better players in games of logic and strategy, even better artists and surgeons. The next major frontier in ML will be the development of algorithms that are able to autonomously access generalized datasets and rapidly learn to pattern-match to adapt to novel stimuli. The world is drowning in what could be called “dark data.” As the use of connected sensors rises exponentially in everything from shirts and shoes to buildings and even natural environments, it will become ever more evident that corporations, interest groups, and governments will require learning systems that “learn how to learn.”
Over the next century, the rise of these algorithms will mean the virtual end of what Western people, today, consider privacy and autonomy. Our cultures live in the transition period between Panoptic societies and Control societies – and we’re rapidly moving toward Control. In his essay, “Postscript on the Societies of Control,” Gilles Deleuze made an early – and in some ways naïve – attempt to classify and outline the mechanisms that will produce order in the future: a future where everyone is watching at all times, and every (important, political or sociocultural) action is directed.
He proclaimed the death of the individual and the dawning of the dividual. Participation in advanced societies of the future will require the dissolution of the individual, who will come to relinquish their sovereignty over decision-making to intelligent systems and allow smart devices to shape choice sets. This will happen mostly voluntarily and – importantly – become much more distributed and autonomous, portending the death of bureaucracy. Those who cling to centralization like the European Union and China are already suffering technologically and socially and will rapidly fall into irrelevance if they choose not to reform.
Alongside this development, as we already see in its infancy, will be splinter movements aimed at providing digital sovereignty to pseudonymous identities. What has been called “Web3” has been built on this premise, and technologies like Urbit allow those who wish to define their level of participation in intelligent networks to do so. Those who are radical in their pursuit will spend a lot of time curating their experiences and guarding themselves. At the moment, this takes a high level of intelligence, but it will come to only require elevated levels of investment of time and capital.
The shape of the world of the future will simultaneously be the dissolution of identity brought about by the domination of intelligent systems in routine daily life and the explosion of identities in hyper-nomadic cultures that are digitally native. The dividual of the future will live multiple lives, some entirely independent of each other, as they transition between networks.
Niklas Blanchard is a software engineer who has been building and training learning models for large and mid-tier enterprises for five years. He has recently been working on integrations with generalized datasets.