A worker can receive a payment from an automated economy without owning any of it. They can be financially protected and still have little say over the systems deciding what work exists. That distinction sits near the center of the DeepMind Institute’s emerging debate about life after artificial general intelligence.
The institute’s introduction, by Shane Legg, James Manyika and Demis Hassabis, presents a platform for discussing the social consequences of AGI: AI with broadly general capabilities, rather than skill at one bounded task. It calls for perspectives beyond the people building the technology. Its essays are explicitly not Google’s official views.
The institution is worth examining as an attempt to shape the questions society asks, not as proof that the future it discusses has arrived. Its leaders’ expectations about approaching AGI are attributed judgments. An essay collection does not establish a timetable, and a scenario is not a forecast.
Three futures, not one prediction
In Economic Policy for AGI, Julian Jacobs and Alex Imas compare eleven interventions across three scenarios: relatively familiar disruption, broader displacement, and a much deeper separation of economic growth from human work.
The proposed response changes with the scenario. Existing wage supplements and unemployment support play a role in the mild case. More durable income support becomes relevant if displacement persists. Broad ownership of capital becomes a backstop if wages cease to distribute much of the economy’s gains. These are policy arguments, not demonstrated effects of an AGI economy.
Importantly, the essay proposes watching economic conditions rather than treating an AGI announcement as the switch. The stronger lesson is procedural: prepare institutions before a crisis, but connect their activation to observed harm.
Read the panel’s small print
The policy-rating tables require particular care. Their 51 economist personas are AI simulations informed by economist surveys, not 51 people independently voting on each displayed score. That makes the rankings an exploratory modeling exercise, not an expert consensus or a real-world policy trial.
A simulated panel may help organize competing considerations. It cannot settle whose values should prevail. A decimal score can make a judgment look more measured than it is; the useful questions remain what assumptions produced it and how the conclusion changes when those assumptions change.
Having a stake is not the same as setting the rules
Our reading is that the institute’s most important challenge is public agency. Imagine a fund that pays everyone a share of investment returns. That might spread financial gains. It would not, by itself, decide whether workers can challenge an automated dismissal, whether communities can refuse a facility, or who determines a system’s acceptable risks.
Another institute contributor, Cambridge’s Stephen Cave, argues for humility and pluralism in Principles for a New Utopianism. His point is not that one perfect society can be engineered, but that communities need room for different ideas of a good life. That offers a useful counterweight to treating technological capacity as a destination in itself.
These are separate decisions about institutions and rights. They cannot be inferred from a productivity chart. An open debate needs room to question the expected benefits, the timetable and the proposed distribution of power—not only how to adapt to a future already chosen.
For readers, that makes the institute a source to engage with critically, rather than a roadmap to accept wholesale. Its connection to a leading AI developer makes the agenda influential; the disclaimer makes clear that individual essays should not be mistaken for corporate commitments.
Three existing Vastkind questions help make this debate concrete: who sets the speed of AI development; who can refuse the infrastructure; and who carries its costs. None requires agreement on an AGI arrival date before public choices matter.
The test for this new forum is therefore concrete: will people outside the AI industry help choose the destination, or only be invited to discuss the journey?
