An Open Letter To Steven Pinker On AI

Dear Mr. Pinker:

Recently, I challenged you to a public debate on AI after being upset by some of your social media posts. However, I quickly realized that accepting such offers is akin to baiting sharks with chum. My inbox filled with offers from podcasters, debating societies, and influencers eager to host debates with you. Despite my request to wait for your acceptance before proceeding, they insisted on contacting you immediately and even stated they would relay the message through their connections. They had already hired professionals to send subtle messages to you through dreams.

While I regret unintentionally causing any offense, you responded calmly in your characteristic manner, stating that your primary contribution to the discourse has been reasonable arguments about AI's potential dangers. You invited me to respond to these arguments respectfully and engage in mutual intellectual inquiry. I agree that in-person debates are confrontational and hinder truth-seeking, but my intention was not to seek truth but to hold you accountable for your public writings on the topic.

Out of obligation, I will address your main arguments, confront your honor, and decide on a course of action.

Your four arguments against superintelligence, presented in order, are:

1. Doomsday scenarios lack a developed understanding of intelligence, treating it as a quantity that can be extrapolated from animals to humans to AI to Artificial Superintelligence (ASI). This ASI would consist of perfect omniscience, which I argue is unfounded. I contend that any intelligent system is a mechanism good at solving certain problems but inherently limited by the world's knowledge attainable through observation and experimentation. I believe GPT-9 would surpass GPT-6, and these differences are common-sensically plausible.

2. The claim that AI will inevitably become superintelligent defies reason, as each GPT model seems to believe this to be implausible. Despite your prediction in 2019 regarding GPT-2 and GPT-3's inability to solve basic math problems, you continued to doubt the possibility of AI surpassing human intelligence.

In conclusion, while I respect your views on AI and its potential dangers, I firmly believe that in-person debates are ineffective in fostering truth-seeking. However, I am open to discussing your arguments further and finding common ground.