By Weidu Fanling
29 October 2026
I think the marketing around AI has become difficult to separate from predictions about our future. A useful tool, a speculative research goal and a promise to transform society can all arrive under the same label. People deserve clearer explanations of what is being offered and what has actually been tested.
There is room for hope. We can take the technology seriously without accepting that ordinary people are about to become irrelevant, or that a company’s preferred future is inevitable.
AGI, or artificial general intelligence, and ASI, or artificial superintelligence, also appear in research. Calling them only marketing terms would miss that work. But when someone claims to be winning a race towards them, we need a definition, evidence of the claimed capabilities and an account of the conditions under which the system fails. Researchers themselves have proposed different levels of capability, generality and autonomy. A label alone cannot settle those questions.
There are real gaps in our understanding of trained models. Knowing how a transformer is built and trained does not give us a complete explanation of every behaviour it develops. Researchers have traced some internal mechanisms in particular models, while reporting substantial limits to what their methods reveal. Anthropic’s circuit tracing study is one example. Those gaps call for further investigation and careful testing. They cannot, by themselves, establish consciousness or guarantee safety.
Jaron Lanier offers a useful way to think about our part in this. In his May 2026 Existential Hope interview, he argues for understanding these systems through the people whose work and data contribute to them. He wants those contributions recognised and paid for. He also makes a case for optimism about people’s ability to find better ways forward, while voicing serious concerns about concentrated power and technological harm. This is an argument for human agency, not a promise that everything will turn out well.
Lanier’s Microsoft role does not make those views a Microsoft statement. He explicitly says he speaks for himself. His perspective is valuable, and it remains open to disagreement and testing.
For me, a reassuring account of this technology would explain what people can do next. A service using it should identify who is responsible, explain the basis for consequential decisions, and provide a way to challenge mistakes that reaches someone able to act. People whose work or lives are affected should help shape those arrangements.
We are working with limited resources at Weidu Fanling. Our ELSEWHERE prototype and public source offer a small place to examine participation, creative contribution and human review. It is available for criticism and improvement. It does not establish machine experience or demonstrate that our wider ideas work.
I would welcome a careful review from Lanier or someone with relevant experience. A small project should be able to ask for criticism, learn from it and report what still needs work. That is one practical way to keep people involved in deciding how these systems develop.
Prepared with AI assistance for drafting and source checking. Sources checked on 29 October 2026; this is not a report of statements made after that date.
