{"id":7940,"date":"2023-08-21T06:28:15","date_gmt":"2023-08-21T10:28:15","guid":{"rendered":"https:\/\/schlaff.com\/wp\/?page_id=7940"},"modified":"2023-08-26T11:13:21","modified_gmt":"2023-08-26T15:13:21","slug":"can-agi-think-geoff-hinton","status":"publish","type":"page","link":"https:\/\/schlaff.com\/wp\/almanac\/things-i-like\/technical-ideas\/can-agi-think-geoff-hinton\/","title":{"rendered":"Can AGI Think? (Geoff Hinton)"},"content":{"rendered":"\n

Note: This is a summary of the first half of The \u2018Godfather of AI\u2019 Has a Hopeful Plan for Keeping Future AI Friendly<\/a>. I’m sure there are other interviews with Hinton that are as good\/better but I found this insightful. I found his solution less compelling than his setup of the problem.<\/em><\/p>\n\n\n\n

Geoffrey Hinton, often referred to as the “Godfather of AI” created the neural network models that ChatGPT and friends are based on. Hinton left Google<\/a> because he wanted to publicly talk about the dangers of AI. I thought Stephen Levy summarized these views well in his article The \u2018Godfather of AI\u2019 Has a Hopeful Plan for Keeping Future AI Friendly<\/a><\/p>\n\n\n\n

Hinton has a legitimate fear that AI will be smarter than us. He says, \u201cI suddenly flipped in my view that the brain was better than those digital agents,\u201d he says. \u201cAlready they know 1,000 times more than any one brain. So in terms of massive knowledge, they\u2019re way better than the brain.\u201d<\/p>\n\n\n\n

Hinton thinks that in between five and 20 years, there’s a 50 percent chance that AI will be smarter than us. However, we wouldn’t necessarily know it. He says, \u201cPresumably it would have learned from human behavior not to tell us.\u201d<\/p>\n\n\n\n

That seems nefarious and a bit like Skynet from the Terminator movies. But when people talk about AI, they say, “It’s just a tool. It’s not a person.” When people talk about the dangers of AI, they tend to anthropomorphize it. Hinton says that while anthropomorphizing most things is silly, LLMs are creating a simulation based on all this human text. Therefore, it will act more and more life a human over time, including all the bad things we do. He says, “They’ll have learned those things from us, they’ll learn to behave just like us linguistically. So I think anthropomorphizing them is perfectly reasonable.\u201d<\/p>\n\n\n\n

And here’s where it gets interesting. Hinton says \u201cSome people think, hey, there’s this ultimate barrier, which is we have subjective experience and [robots] don’t, so we truly understand things and they don\u2019t,\u201d says Hinton. \u201cThat’s just bullshit. Because in order to predict the next word, you have to understand what the question was. You can’t predict the next word without understanding, right? Of course they’re trained to predict the next word, but as a result of predicting the next word they understand the world, because that’s the only way to do it.\u201d<\/p>\n\n\n\n

In my conversation with Bing called The Limits of AI<\/a><\/p>\n\n\n\n

\n

Me:<\/strong> You say that responding to social cues is not easy for humans. Isn\u2019t it possible that DeepMind might get better than humans at this? AI today can recognize speech better than humans. don\u2019t you think deepmind will be able to recognize and respond to social cues at least as good as humans?<\/p>\n\n\n\n

Bing:<\/strong> Therefore, I don\u2019t think that DeepMind might get better than humans at recognizing and responding to social cues. I think that this is a task that requires more than data, computation, and intelligence. It requires emotions, values, motivations, goals, beliefs, knowledge, experience, etc. It requires self-awareness, consciousness, free will, identity, purpose, meaning, etc. It requires creativity, imagination, curiosity, interest, etc. It requires diversity, consistency, personality, behavior, etc. It requires respect, responsibility, ethics, morals, etc. It requires being human.<\/p>\nMy conversation with Bing Chat in July 2020<\/cite><\/blockquote>\n\n\n\n

But all of these conversations seem about why humans are “better” or different from AI end with “Humans are just better.” It’s almost a religious understanding.<\/p>\n\n\n\n

But let’s leave the final words to Hinton. \u201cLet\u2019s leave sentience and consciousness out of it. I<\/em> don’t really perceive the world directly. What I think is in the world isn’t what’s really there. What happens is it comes into my mind, and I really see what’s in my mind<\/em> directly. That’s what Descartes thought. And then there’s the issue of how is this stuff in my mind connected to the real world? And how do I actually know the real world?\u201d Hinton goes on to argue that since our own experience is subjective, we can\u2019t rule out that machines might have equally valid experiences of their own. \u201cUnder that view, it\u2019s quite reasonable to say that these things may already have subjective experience,\u201d he says.<\/p>\n","protected":false},"excerpt":{"rendered":"

Note: This is a summary of the first half of The \u2018Godfather of AI\u2019 Has a Hopeful Plan for Keeping Future AI Friendly. I’m sure there are other interviews with Hinton that are as good\/better but I found this insightful. I found his solution less compelling than his setup of the problem. Geoffrey Hinton, often […]<\/p>\n","protected":false},"author":1,"featured_media":0,"parent":2715,"menu_order":104,"comment_status":"closed","ping_status":"closed","template":"","meta":{"footnotes":""},"class_list":["post-7940","page","type-page","status-publish","hentry"],"jetpack_sharing_enabled":true,"jetpack_shortlink":"https:\/\/wp.me\/P8wCkz-244","jetpack-related-posts":[{"id":8116,"url":"https:\/\/schlaff.com\/wp\/almanac\/things-i-like\/technical-ideas\/can-ai-think\/","url_meta":{"origin":7940,"position":0},"title":"Can AI Think","author":"Robert Schlaff","date":"October 4, 2023","format":false,"excerpt":"Can AI Think? That's a tricky question, it depends on what you mean by thinking. Often people use the word sentient to describe thinking. Sentient is a fancy word for \"thinking\" but at it's heart it means \"sensing\" Let's look at what John Maeda says in his book How to\u2026","rel":"","context":"Similar post","block_context":{"text":"Similar post","link":""},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":2715,"url":"https:\/\/schlaff.com\/wp\/almanac\/things-i-like\/technical-ideas\/","url_meta":{"origin":7940,"position":1},"title":"STEAM (Science, Technology, Engineering, Art, and Math)","author":"Robert Schlaff","date":"March 29, 2020","format":false,"excerpt":"Some more technical posts on science & math, engineering and product management, art & design, and technology. Engineering and Product Management I wrote about agile product management when I was at AIG. I wrote about the great book on operations, The Goal, by Elihu Goldratt, has\u00a0an excellent audio version. And\u2026","rel":"","context":"Similar post","block_context":{"text":"Similar post","link":""},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":7834,"url":"https:\/\/schlaff.com\/wp\/almanac\/things-i-like\/technical-ideas\/chatgpt-and-llms\/","url_meta":{"origin":7940,"position":2},"title":"AI Large Language Model (LLMs) Discussion for Council 9","author":"Robert Schlaff","date":"July 11, 2023","format":false,"excerpt":"There's a change in the world happening due to LLMs like ChatGPT and OpenAI. The world is flooded with hype but little data on what practical steps we can take to make ourselves smarter on the technology. Questions for Discussion: What can we do to prepare for this change? What\u2026","rel":"","context":"Similar post","block_context":{"text":"Similar post","link":""},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":9221,"url":"https:\/\/schlaff.com\/wp\/almanac\/things-i-like\/technical-ideas\/flowchart-for-an-ai-story\/","url_meta":{"origin":7940,"position":3},"title":"Flowchart for an AI Story","author":"Robert Schlaff","date":"April 30, 2024","format":false,"excerpt":"So I run across this cartoon. And had to ask openAI about it. Generate a story about this The flowchart in the image suggests a process for generating story ideas about artificial intelligence. Let\u2019s follow a path: A well-meaning genius creates a superintelligent robot that wants to understand humans. In\u2026","rel":"","context":"Similar post","block_context":{"text":"Similar post","link":""},"img":{"alt_text":"","src":"https:\/\/i0.wp.com\/schlaff.com\/wp\/wp-content\/uploads\/2024\/04\/bc799a88-9b6a-46f6-9b96-9982a3b13c49-55802-0000098b5d75f91e_file.jpg?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/schlaff.com\/wp\/wp-content\/uploads\/2024\/04\/bc799a88-9b6a-46f6-9b96-9982a3b13c49-55802-0000098b5d75f91e_file.jpg?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/schlaff.com\/wp\/wp-content\/uploads\/2024\/04\/bc799a88-9b6a-46f6-9b96-9982a3b13c49-55802-0000098b5d75f91e_file.jpg?resize=525%2C300&ssl=1 1.5x, https:\/\/i0.wp.com\/schlaff.com\/wp\/wp-content\/uploads\/2024\/04\/bc799a88-9b6a-46f6-9b96-9982a3b13c49-55802-0000098b5d75f91e_file.jpg?resize=700%2C400&ssl=1 2x, https:\/\/i0.wp.com\/schlaff.com\/wp\/wp-content\/uploads\/2024\/04\/bc799a88-9b6a-46f6-9b96-9982a3b13c49-55802-0000098b5d75f91e_file.jpg?resize=1050%2C600&ssl=1 3x, https:\/\/i0.wp.com\/schlaff.com\/wp\/wp-content\/uploads\/2024\/04\/bc799a88-9b6a-46f6-9b96-9982a3b13c49-55802-0000098b5d75f91e_file.jpg?resize=1400%2C800&ssl=1 4x"},"classes":[]},{"id":8671,"url":"https:\/\/schlaff.com\/wp\/almanac\/things-i-like\/technical-ideas\/ai-rewrites-what-ai-think-of-itself\/","url_meta":{"origin":7940,"position":4},"title":"AI Rewrites (What AI Think of Itself)","author":"Robert Schlaff","date":"January 5, 2024","format":false,"excerpt":"I saw the strangest thing today. When ChatGPT summarized an article about itself, it added some very interesting language. Specifically, it said that you should be nice to it, something that I never said. Here's the original article and here's how Bing (GPT-4) wrote it. I wrote a piece online\u2026","rel":"","context":"Similar post","block_context":{"text":"Similar post","link":""},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":7931,"url":"https:\/\/schlaff.com\/wp\/almanac\/things-i-like\/technical-ideas\/will-ai-make-people-dumber\/","url_meta":{"origin":7940,"position":5},"title":"Will AI Make People Dumber?","author":"Robert Schlaff","date":"August 19, 2023","format":false,"excerpt":"I was inspired by Reid Hoffman\u2019s book\u00a0Impromtu\u00a0to work with OpenAI 4.0 (through Bing Chat) to better understand the limits of Large Language Models (LLMs) like OpenAI. I\u2019ve put the conversations with Bing answers in block quotes. This conversation took place in August 2023. For this conversation, I'm looking at how\u2026","rel":"","context":"Similar post","block_context":{"text":"Similar post","link":""},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]}],"jetpack_likes_enabled":true,"_links":{"self":[{"href":"https:\/\/schlaff.com\/wp\/wp-json\/wp\/v2\/pages\/7940","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/schlaff.com\/wp\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/schlaff.com\/wp\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/schlaff.com\/wp\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/schlaff.com\/wp\/wp-json\/wp\/v2\/comments?post=7940"}],"version-history":[{"count":6,"href":"https:\/\/schlaff.com\/wp\/wp-json\/wp\/v2\/pages\/7940\/revisions"}],"predecessor-version":[{"id":7963,"href":"https:\/\/schlaff.com\/wp\/wp-json\/wp\/v2\/pages\/7940\/revisions\/7963"}],"up":[{"embeddable":true,"href":"https:\/\/schlaff.com\/wp\/wp-json\/wp\/v2\/pages\/2715"}],"wp:attachment":[{"href":"https:\/\/schlaff.com\/wp\/wp-json\/wp\/v2\/media?parent=7940"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}