Doom or Bloom
问题 1
你认为AI对我们的未来意味着什么？为什么？
I expect AI to be broadly beneficial, though not remotely harmless. It can become a “cognitive prosthesis”: making intellectual work more accessible, helping people write software, design products, learn faster, and pursue discoveries that would otherwise require scarce expertise. Wider access matters. I would rather see many people, companies, and countries gain powerful tools than have one corporation, government, or supposedly perfect AI control them.
But useful progress is not the same as an inevitable intelligence explosion. AI can help engineers improve AI without creating a self-sustaining runaway loop. The key question is how much validated research progress we get for the resources invested. Generating more code or plausible ideas is not enough if testing them is expensive, judgment remains unreliable, or each improvement delivers diminishing returns. Physical science adds another constraint: discoveries still require observations, instruments, and experiments, although automating laboratory work could help enormously.
There will also be accidents, malicious uses, and deployment failures. Openness and competition distribute benefits, but they do not abolish risk. We need defense in depth: better instruction following, monitoring, sandboxing, red teaming, cyber defenses, and accountability for negligent providers. Concentrating power to avoid every possible misuse creates profound risks of its own.
So my default future is neither effortless utopia nor inevitable doom. It is consequential progress, substantial benefits, serious harms, and an ongoing contest over who gets access and how well we manage the consequences. Dramatic forecasts deserve scrutiny: evidence matters more than hunches.
