AI 2027 - if you're not terrified, you're not paying attention
Many AI experts believe it likely that AI will kill every human, perhaps in the next decade.
These are not conspiracy theorists. Many tech bosses choose not to have children because they don't believe the world will be around much longer!
"AI 2027" is a beautiful, frightening description of what could happen. Our world will utterly transform. Some outcomes may be good, others may be very, very bad.
Today, the people in charge of this are utterly selfish and untrustworthy - not just Trump and Xi, but also Sam Altman, Jeff Bezos, Elon Musk, Mark Zuckerberg ...
But it's not just them.
The rest of us have become greedy and short-sighted. We've lost the ability to resist temptation, to delay immediate gratification (like AI offers), to consider long-term consequences. We willingly "trust" persuasive voices who allow us to pretend that our fears are unjustified. We know it's not true, but it's easier.(Climate change is tragic, living proof of this)
With AI this will be even harder. Because, while of course, there will be cynical AI proponents, there will also be well-meaning people who will argue against slowing down AI, because of its benefits, like medical advances that could cure real people. "How can we just let them die when AI might be able to save them?"
It won't be easy. But letting AI advance without guardrails is unfathomably irresponsible. We may destroy life as we know it. We may all die.
Mr. Trump eliminates guardrails just because he can. He emphasises the need to move faster than China, rather than move carefully and align with China. It is a tragedy that this moron is in charge at such a critical time, but that's the situation we have to deal with. We cannot wish him away.
The AI 2027 essay is very long. The people who write about AI typically write for an audience of people who have nothing better to do than read long essays about AI. (I've been told that I write too much, but this is another level). There is one upside to this, though: each point is supported by very detailed, data-based logic. You cannot read it and pretend it's just some luddite crackpot.
But in case you don't have time, I can share the key outcomes (spoiler alert) :
- There is an entirely conceivable outcome in which AI replaces (i.e. kills) all of humanity in 2030. Not 2070 or 2140 ... in 5 years. Many experts believe this will happen if we don't slow down. (does that make it worth an hour of your time??)
- People who work on AI have a short-hand terminology they call "p(doom)" - their estimate of how likely it is that AI will eliminate humanity. Most have a value in the range of 5-50%. Yes, 50% !! Nobody says 0%. This isn't about taking our jobs, this means killing us.
- Positive scenarios exist, but require AI powers (both AI companies and nations) to aggressively SLOW DOWN the rate of AI development, which is currently vastly outpacing our ability to evaluate its safety.
This is not happening today, instead we have a race dynamic where US and China each see the (real) danger of letting the other develop dominant AI first.
We need to fix this!
[this was originally a linkedin post. There, I linked to another linkedin post from Caroline Jeanmaire which gives more perspective and perhaps even more credibility].
