It’s so sad that AI developers are creating affordable, powerful mental assistance for everyone—and yet are afraid this is somehow going to destroy the world.
AI is amplified intelligence
“AI” is a new type of computer. It doesn’t replace the human mind, it amplifies the human mind. AI should be called amplified intelligence not “artificial intelligence,” because it doesn’t replace the human mind, it amplifies the human mind. An artificial heart can replace the role your heart plays in your life. AI cannot replace the role your mind plays in your life. If I could only use three words to describe large language models, I would describe them as “symbol-guessing savants.” They aren’t conscious and they don’t have goals. But they are extraordinarily good at taking the symbols we give them and generating symbols we find useful.
We need to be careful how we connect it to high-stakes systems.
That’s it.
Premises behind “AI Going Rogue”
“AI went rogue” amounts to:
1. Someone took a powerful, open-ended computer that they don’t perfectly control (i.e., one running LLMs).
2. They deliberately reduced their control over its ability to produce potentially harmful outputs.
3. They connected the deliberately-badly-controlled computer to a high-stakes system.
Whether this is done irresponsibly or as a legitimate exploration of potential risks, it does not justify doomerism. The proper conclusion to draw is that we need to responsibly develop LLM computers and be very responsible about when and how we connect them to high-stakes systems.
Not that AI is some interested, conscious being that has a 10% or even 0.0000001% chance of killing us all. “AI” is a bunch of computers running an amazing new kind of software that we can and should use to be of enormous assistance to our lives. AI computers are no more self-interested or conscious than a calculator or an abacus—even though their outputs do a great job of mimicking interests and consciousness.
Solution to AI threats
While there is a real threat of people accidentally misusing AI assistance to do a lot of damage, that threat is always going to be limited by the humans who ultimately direct AI computers and can literally unplug them. We already have laws against configuring and connecting computers in ways that lead to felonies. AI companies should develop technologies and practices so their AI agents follow those laws, not ask the government to dictatorially limit AI progress as such.
The biggest AI threat by far is bad actors deliberately using AI assistance to do a lot of damage. To prevent this threat, as well as to reduce the damage of accidents, we need good actors, above all in the US, to proudly lead in this technology. Not to indulge in fear-mongering based on metaphysical nonsense.
The Industrial Revolution gave us machine physical assistance. The AI revolution is giving us machine mental assistance. This is an enormous expansion of human capability—and we’re only just beginning to discover what we can do with it.