Your experiment demonstrates how sensitive the "responses" are to the way the question is asked. But more fundamentally: Having built this more or less comprehensive socio-econom ic profile of each riding, where exactly do the "opinions" come from? Something's missing here.
One of my sons asked me exactly the same question - and I haven't yet convinced him. But I think it's fairly straightforward. As an analogy, it's reasonable for me to claim that I know my family members well enough to fairly accurately predict their opinions on a wide range of topics - even without asking them. The reason I can do that is because I'm familiar with their past actions and with the way they tend to think.
So I don't think it's much of a jump to extend such analysis to an AI who has been "briefed" on a population group's previous voting experiences and demographics. It's not unreasonable to assume that, say, highly educated, high-income people without families are more likely to think one way, while low-income families with less education would more likely to think differently. That's all the AI is doing. The benefit of using an AI over a wonk from Statistics Canada is that the AI can do the job much faster and for a lot less money.
Interestingly, when I asked the AI about its methodology, it told me that it didn't use any randomization to permit reproducability. I didn't expect that.
David, you can put me in the same unconvinced column as your son.
I absolutely agree that I (just to personalize it) am a member of group X (I'm old and retired; I'm male; my income is whatever it is; I live in riding X which elected X candidate and so forth). Now, just because that is who I was at the last election, have I become more or less persuaded on your question than I was at the time that I voted? Has my group started to see the fallacies / glories of policy X, and so forth?
I see your attempt as a good toy for you to play with and I say that with no derisory tones. Playing with concepts is how concepts are proven / improved / exposed and so forth. My point is that you should explore this to your heart's content but please do not - at this point, anyway - view this as "almost" ready for prime time. It isn't.
But, again, everyone is allowed hobbies so this can be yours.
I'm not sure election predictions are the best use-case for this particular tool. But there are many other scenarios where people might be very interested in understanding populations.
Having said that, there's nothing preventing us from adding all kinds of up-to-date background information to the AI prompt to add depth to the assessments.
Any AI "polling" will be as accurate as Professor Allan Lichtman was in predicting the 2024 US Presidential Election result; that's why the machine overlords had to constantly "reboot" the Matrix in the movie series.
Excellent read...predicting is always a fun sport. Even for AI.
Your experiment demonstrates how sensitive the "responses" are to the way the question is asked. But more fundamentally: Having built this more or less comprehensive socio-econom ic profile of each riding, where exactly do the "opinions" come from? Something's missing here.
One of my sons asked me exactly the same question - and I haven't yet convinced him. But I think it's fairly straightforward. As an analogy, it's reasonable for me to claim that I know my family members well enough to fairly accurately predict their opinions on a wide range of topics - even without asking them. The reason I can do that is because I'm familiar with their past actions and with the way they tend to think.
So I don't think it's much of a jump to extend such analysis to an AI who has been "briefed" on a population group's previous voting experiences and demographics. It's not unreasonable to assume that, say, highly educated, high-income people without families are more likely to think one way, while low-income families with less education would more likely to think differently. That's all the AI is doing. The benefit of using an AI over a wonk from Statistics Canada is that the AI can do the job much faster and for a lot less money.
Interestingly, when I asked the AI about its methodology, it told me that it didn't use any randomization to permit reproducability. I didn't expect that.
David, you can put me in the same unconvinced column as your son.
I absolutely agree that I (just to personalize it) am a member of group X (I'm old and retired; I'm male; my income is whatever it is; I live in riding X which elected X candidate and so forth). Now, just because that is who I was at the last election, have I become more or less persuaded on your question than I was at the time that I voted? Has my group started to see the fallacies / glories of policy X, and so forth?
I see your attempt as a good toy for you to play with and I say that with no derisory tones. Playing with concepts is how concepts are proven / improved / exposed and so forth. My point is that you should explore this to your heart's content but please do not - at this point, anyway - view this as "almost" ready for prime time. It isn't.
But, again, everyone is allowed hobbies so this can be yours.
I'm not sure election predictions are the best use-case for this particular tool. But there are many other scenarios where people might be very interested in understanding populations.
Having said that, there's nothing preventing us from adding all kinds of up-to-date background information to the AI prompt to add depth to the assessments.
Any AI "polling" will be as accurate as Professor Allan Lichtman was in predicting the 2024 US Presidential Election result; that's why the machine overlords had to constantly "reboot" the Matrix in the movie series.