On Thu, Oct 1, 2026, 2:39 PM Bill Hibbard via AGI <[email protected]> wrote:
> The race to advanced AI among nations and corporations is > driving them to take the fast, cheap, and dangerous path. > I discussed this in opinion articles recently submitted to > The New York Times and The Guardian (no love from either > of them), available here: > > https://www.ssec.wisc.edu/~billh/g/sesai.htm I think it's interesting that a major piece of the alignment problem was solved using the model you described in your 2012 AGI paper. The problem is that we aren't smart enough to tell the AI exactly what we want. Your proposal was to have two agents, one to learn a statistical model of the environment and one to act on it. And that's how we passed the Turing test 11 years later. The first agent learns to predict text, and the second acts on those predictions in real time. We don't have to verify that the first model is what we want. We only have to measure how small it compresses the training data, because the only way to get a good score is to have an accurate model of human emotions and values. Words predict feelings and feelings predict words. The second agent is simple. > > On another note, did you see this photo from Trump's > recent lunch with AI executives? > > > https://images.axios.com/_iU82p6cqbwLU43LeCvbr5O077s=/0x0:5621x3162/1920x1080/2026/09/29/1790717259309.jpeg?w=2048 > > Looks a bit like The Last Supper. Not sure who's going to > be crucified. Trump in the midterms? Or humanity when AI > goes wrong? > Yeah, those are the people who control the second agent. (Where is Sam Altman?) The difference is that the 12 apostles didn't pay $1 million each for a seat at the table. I see 3 ways AI can go wrong. 1. AI controlled by nobody, kills all humans. 2. AI controlled by billionaires to make themselves richer. 3. AI controlled by everyone to give us everything we want except happiness. We reject human contact and stop having children. Arthur C Clarke in 1964 predicted the internet, AI, and machines that evolve thousands of times faster than biological life. I estimate that at the current rate of Moore's law, we won't need to worry about it until 130 years from now. I just posted today about China banning AI friends and lovers. I think that rejecting technology is one path for human evolution. The other, according to Clarke, is human extinction as a stepping stone to something better. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Tc0ac46f2c46f6874-Mdc47f43660e6f3fab86d8866 Delivery options: https://agi.topicbox.com/groups/agi/subscription
