This is a fabulous piece, thank you! I love "hypersuasion"--had not heard that. I just watched the first episode of Mrs. Davis because someone recommended it as the most possibly relevant vision of AI in current sci fi. They're trying to showcase a more rudimentary form of hypersuasion there... the AI in charge of civilization and sort of hypnotizes people to be its mouthpiece through earbuds, but also mysteriously knows everything about everyone and promises to fulfill wishes. I'd like to see a film exploring more subtle forms of hypersuasion and questions of human agency, autonomy, self-determination...
I love the idea of coining new terms. It would be fun to play around with Claude to help with that, too.
How would cognidoubt be different from skepticism? I wonder if there's a philosopher we could ressuscitate that would be relevant... I've been talking a lot about cultivating skeptical habits of mind in students in relation to AI.
I hadn't heard about Mrs. Davis! I will definitely check it out. I think 'Her' does a good job of showing the subtlety of an effective AI companion, though I am not sure we can say that what ScarJo's AI character is doing in that movie should be considered persuasion. And I am not sure that 'she' steals Theo's agency in that movie - just his heart...
Good question about skepticism. Cognidoubt aims to replace 'trust' in the lexicon but your question leads me to realize where it falls short. The problem with 'trust' in an AI 'relationship' is that it implies shared experiences and feelings. AI brings neither to the table. BUT, when we work with AI, we do have a certain level of 'confidence' (trust) that the system will produce something of value - that, to me, is inevitable and also "okay." So I think the new word in this case would have to capture the confidence and leave behind the feelings and shared experiences aspect of the word trust. Thoughts?
(Then again, the word confidence includes "confide" and also makes me think of "confidence man" or "con artist." Yikes....our language has a longgg way to go!)
If you want a good reminder that you are interacting with a machine, just repeat the exact same back and forth set of prompts with 3 different models. While the combination of words change somewhat and the structured outputs may be organized differently, the responses generally converge around the same basic ideas and feedback. I like the focus of this post. Humans have a tendency to anthropomorphize almost anything (i.e., cars) so it's really not all that surprising that we are going to gravitate to using this kind of language when we are literally interacting with something that speaks to us in the way chatbots do. It would actually be surprising if we didn't. But in many ways all the chatbot is doing is reconfiguring trillions of word combinations back at us in ways that we can make meaning from. And it does this far better and more clearly than many people can do with language. But it's not an alien intelligence and it certainly does not have it's own internal motivations. Thoughtful post.
This middle ground between tech and human is a concept I am obsessed with. Rob Nelson and I did a mini-series on it a few months ago - you might find the Wild Robot articles of interest.
Also, great point about cars. I think cars are probably the best analogy point to use when discussing AI use - it works on a wide variety of levels, including this one! Thanks.
I suggest LLMs should be viewed as a database of a great many documents which can be queried in natural language. The machinery both collects and confabulates text to return. It has no inner life even when clever prompting makes it look like that. But… because it uses words, our biases fall for it.
I always love telling High School Freshmen that Shakes invented more words than anyone else. The whole concept of inventing new words blows their minds enough to make them think I am some kind of genius...
This is a fabulous piece, thank you! I love "hypersuasion"--had not heard that. I just watched the first episode of Mrs. Davis because someone recommended it as the most possibly relevant vision of AI in current sci fi. They're trying to showcase a more rudimentary form of hypersuasion there... the AI in charge of civilization and sort of hypnotizes people to be its mouthpiece through earbuds, but also mysteriously knows everything about everyone and promises to fulfill wishes. I'd like to see a film exploring more subtle forms of hypersuasion and questions of human agency, autonomy, self-determination...
I love the idea of coining new terms. It would be fun to play around with Claude to help with that, too.
How would cognidoubt be different from skepticism? I wonder if there's a philosopher we could ressuscitate that would be relevant... I've been talking a lot about cultivating skeptical habits of mind in students in relation to AI.
I hadn't heard about Mrs. Davis! I will definitely check it out. I think 'Her' does a good job of showing the subtlety of an effective AI companion, though I am not sure we can say that what ScarJo's AI character is doing in that movie should be considered persuasion. And I am not sure that 'she' steals Theo's agency in that movie - just his heart...
Good question about skepticism. Cognidoubt aims to replace 'trust' in the lexicon but your question leads me to realize where it falls short. The problem with 'trust' in an AI 'relationship' is that it implies shared experiences and feelings. AI brings neither to the table. BUT, when we work with AI, we do have a certain level of 'confidence' (trust) that the system will produce something of value - that, to me, is inevitable and also "okay." So I think the new word in this case would have to capture the confidence and leave behind the feelings and shared experiences aspect of the word trust. Thoughts?
(Then again, the word confidence includes "confide" and also makes me think of "confidence man" or "con artist." Yikes....our language has a longgg way to go!)
If you want a good reminder that you are interacting with a machine, just repeat the exact same back and forth set of prompts with 3 different models. While the combination of words change somewhat and the structured outputs may be organized differently, the responses generally converge around the same basic ideas and feedback. I like the focus of this post. Humans have a tendency to anthropomorphize almost anything (i.e., cars) so it's really not all that surprising that we are going to gravitate to using this kind of language when we are literally interacting with something that speaks to us in the way chatbots do. It would actually be surprising if we didn't. But in many ways all the chatbot is doing is reconfiguring trillions of word combinations back at us in ways that we can make meaning from. And it does this far better and more clearly than many people can do with language. But it's not an alien intelligence and it certainly does not have it's own internal motivations. Thoughtful post.
Thank you sir!
This middle ground between tech and human is a concept I am obsessed with. Rob Nelson and I did a mini-series on it a few months ago - you might find the Wild Robot articles of interest.
Also, great point about cars. I think cars are probably the best analogy point to use when discussing AI use - it works on a wide variety of levels, including this one! Thanks.
I suggest LLMs should be viewed as a database of a great many documents which can be queried in natural language. The machinery both collects and confabulates text to return. It has no inner life even when clever prompting makes it look like that. But… because it uses words, our biases fall for it.
Amen, couldn't agree more!
Now we need a word that describes this approach -or words! Easier said than done.
Much easier said than done!
I love that you put this out here :)
Thanks! Hope we get some fun words in the Form.
If Shakespeare did it....
I should've included that, haha!
I always love telling High School Freshmen that Shakes invented more words than anyone else. The whole concept of inventing new words blows their minds enough to make them think I am some kind of genius...