Once I read the "it will remind you to trade soda pop for sparkling water" parenthetical, my skepticism meter spiked, and it never went down throughout reading the article.
Most of us do not adhere to the optimal level of health management, because most of us want to do some things that aren't optimal. I know I have a few myself. (But not swapping out soda pop for sparkling water, I did that many years ago!) Some people do things that are very suboptimal to their own health. I think that that's regrettable, but ultimately it's their body and they're in charge of what they want to do to it.
Now, I don't want to completely downplay the opportunities that can be had with advances in informational medical technology. For those who want to make a concerted effort to improve their health, this could be a useful tool to help them out. But I think there will be plenty of people who just say, "Nah, I want to do this instead because I enjoy it.". I would think that a top tier model would account for this, and try to perhaps find a harm reduction avenue instead, but some still might just want to opt out of the technology altogether and not be bothered with bogging down their notifications further.
I think it did--your own skepticism was targeted at socioeconomic factors, which do matter and are a barrier. I just wanted to make sure to add the other barrier of people enjoying doing things that aren't optimal for their health.
I agree with you but my biggest health obstacles is that my brain is an unreliable narrator and if I can push back a little bit it gives me a bit more space to do the things that are optimal and once I have that foothold it can spin up a virtuous cycle.
I would not even say the brain is the issue but the predetermination to not accept discomfort. If you are responding to every time you get a hunger signal rather than accept that you should be hunger for sometime before a meal, then you will gain weight. So much of our food behaviors are habitual reactions rather than planned actions, so I don't how an app (AI or not) is going to help clean up a broken construction of dealing with food.
That's great that you recognize it for yourself, and if you think that it can help you, then by all means go for it. Others may end up disagreeing for themselves.
I agree people don’t leverage the information for their own gain. VO2 max is a good representation of one’s health (I get there’s other important factors too, but this seems under appreciated). If a person had a stationary bike or elliptical and went hard for 10 minutes a day they would significantly increase their physical wellbeing and fitness. These bikes can cost as little as 2-3 month of gym membership. Nike had it right, just do it.
I had the same reaction at first. But... is that true? Willpower is a limited resource. Sometimes it really is helpful to have someone else reminding you, constantly, of what you're supposed to do. It's not about knowledge, it's just having something external to help you. Personal trainers for exercise, for example, clearly help, and most of them aren't teaching arcane secrets of exercise. You can look up all the exercise advice on the internet for free. But it's still really helpful to have someone there with you in the gym, yelling "cmon! one more, one more!" Maybe AI can be like that for everyone, with diet.
I assume that the conversation with investors goes like this:
“So the user puts in their age and sex. Is that going to give them meaningful health advice?”
“No, no — but then we nudge them to input more personal information — diet habits, sleep, exercise. We offer *personalized* advice!”
“Okay, so does that make them healthier?”
“No, it’s still ineffective. So, we nudge them to put in more personal information— links to medical records, credit card data, dating profiles.”
“And that all makes them healthier?”
“No, no — totally ineffective. And the less it does, the more we nudge them to enter personal information. Look: we’re designing this to make you money. You’re not doing this for your health. “
"Social determinants of health" + diabetes really sets off my alarm bells because this is something where I think the left has too much faith in what is possible through government.
I know some young or middle-aged people with diabetes, including a 24 year old who died from it. And you could certainly point to social determinants being a factor - they're mostly lowish income and none have college degrees. But their root "social determinant" problems are that they are generally less capable, less conscientious people whose parents were similar, not that they didn't have access to opportunities.
To take the 24 yo - his parents are medical techs and hold down the job just fine. But their backyard has been completely overtaken by trash and neighbors because they've never cut the grass. When their kids were younger they used to poop back there. Only 1 of their 4 kids has a drivers license, and another one of them was jailed in a metal health facility for threatening to blow up a school.
Other examples are my cousin's husbands, the younger of which puts ranch dressing on nearly every single thing he eats, and the older of which is a landscaper who's been arrested several times into his 40s for getting in fights with strangers.
I feel like many educated liberals don't have these kinds of associations and imagine poor people with diabetes as black or brown people who've been discriminated against and turn to sugar to ease the pain or something. They could be "saved" with more government effort, or whatever.
And I don't think that's wholly wrong, government could do better or more, especially in the most low-income neighborhoods. But we've already grasped a lot of the low-hanging fruit. Convincing people not to put ranch dressing on their bagels is not a government or an AI problem!
> To take the 24 yo - his parents are medical techs and hold down the job just fine. But their backyard has been completely overtaken by trash and neighbors because they've never cut the grass.
This seems to get back to John from FL's comment about how best to evaluate an area--at the grocery store, do people return the dang carts to the corral?
lol I love this. I grew up in a trashy neighborhood. In retrospect, the families who mowed their lawns and did mildly pro-social things like pick up trash if it was blowing down the street - all of their kids made it out. The shitbags who set off fireworks at 3am, whose teenagers girdled neighbors' trees as "pranks" etc. are all still there. As an adult, I vote for opportunities for people who want to make the most of them, and have no interest in spending time or resources trying to convince those who don't.
If the problem is they keep intending to do things better and keep forgetting to, then an AI app might be able to work by reminding them just at the relevant moment about their plan.
If the problem is that they don’t intend to do things better, it can’t fix that.
I feel like there are too many confounders to say very much.
To be clear about some of the things I was trying to say above, culture, habits, genes, personal choices, government programs / policies and the environment must all influence health. But enough of the above are either outside of government control (like personal control, genes) or not low-hanging fruit (redesign all American cities to be more walkable).
So "just adopt European social / medical policies", in my view, would do far less than the left often believes.
I feel like my optimism is more like the people who currently use health apps could be better supported by one with more features is where my optimism is not we can totally rework people.
I use these a lot because a lot of depression and anxiety is just vigilance against habit collapse.
I forget how I ended up at a Michael Huffington campaign speech in high school. But his sticker ended up on the jewel case for a Duran Duran CD. Things got REAL different for both of them after that.
My health insurance keeps offering me a service where some person will email and call me about a diabetic diet. I know all the recommendations, and nagging will not make me more likely to follow it. The service will send me recipes as well. I Hate to cook and have undoubtedly see version of all the recipes.
And I absolutely hate healthy recipes. There is nothing so tedious and time-consuming to make on this earth because the target is health-hobbyist foodies. So they all require immense prep time rather than, like, all the ingredients purchasable frozen or in a can.
And this really is a thing. I've been on a whole kick of cooking classic casseroles and, well, you can get prep down to nearly zero with a recipe of "dump everything in a big bowl, mix, dump into a casserole dish, bake for 30 minutes, and you can feed six people for about $12".
Especially coming from Arianna Huffington, the obliviousness to the social determinants of health leaves a strong whiff Marie Antoinette: “The common people are suffering from chronic diseases? Let them have AI”
Got the Walz poll back from the field. Great numbers for him. You guys will see the full thing sometime next week but here are two fun tidbits to tide you over until then.
We polled a bunch of Democratic policies, some specific and some vague. On average, the specific ones had 47% support while the vague ones had 53% support.
We also did a battery of questions on how voters percieve Obama, Biden, Trump, and Harris on economic policy. Net approval for Obama's economic policy was +14; for Biden's policies, -4; for Trump's policies, -4; and for Harris' policies, +4. When asked how Obama's policies were for themselves and their families, by a +8 margin voters said they were good for them. By a +9 margin voters thought Biden's policies were bad for themselves and their families; for Trump it was +1 good; for Harris' proposed policies it was +4 bad. When asked how each person's policies were for the economy overall: Obama good +10, Biden bad +8, Trump good +1, Harris even (+0). Obama did better than Biden, Trump, and Harris with every single demographic on all the economic policy questions. And his net favorabilty was better with all groups too. GOAT.
Sounds like nostalgia is the biggest factor in attitudes—people have blissfully forgotten how bad their lives were when Obama was in office (not that it was his fault, he inherited the Great Recession after all). People are hyper aware of everything that is wrong right now, mainly because of inflation. Would people really rather have low prices and even lower wages (and a harder time finding work)? Houses were cheap thanks to the crash, but so many people were out of work not many could qualify for a mortgage. Of course it is true that the trend under Obama was one of continual improvement that was not halted by Trump taking office, until the pandemic. It’s really crazy, though, that there would be polling data on Harris’ economic policies, given that she has only begun to say anything about it in the last couple of days.
Agree on nostalgia being a factor. I do think the revealed preference is for lower prices and higher unemployment rather than the converse; just look at voter perceptions of the economy under Biden. Possibly because inflation hurts everyone whereas unemployment only hurts a few.
The exact wording was "the economic policies Kamala Harris has proposed." We're going to do another poll of the proposals she rolled out yesterday in North Carolina—I'm writing that one as we speak.
The Democratic and progressive elites who decided to turn on their back on Obama after 2016 did quite the own goal. It's not even "if it aint broke don't fix it" it's "If it works, break it, then try to fix it by making it much worse" (+14 to -4!!!!!)
Biden was a pretty unserious candidate in 2008 (I’m sure he took himself seriously, but no one else did). In 2020 he was not-Bernie, which was all that was needed to beat Trump. He was never really personally popular.
Does comparing people's views of Obama's and Harris's economic policies really mean anything?
In general, I would far prefer an interview method where you just ask a voter "Which [Trump/Harris] policy do you [most like/least like] and please tell us why" then use transcription and some smart analytical tools to quantify their reactions, throwing out all answers where it's clear the voter doesn't have a clue and is just making things up. (Or coding them as "don't have a clue" and keeping that in the final reported results.)
Not really how polling works. If you throw out the answers where voters are ill-informed about policy you’re not going to have very many answers to work with. And ill-informed views do inform vote choice so you want to include them. Also a huge pain (and very expensive) to do open quesitons and then hand-code ~1,000 responses.
As to whether it means anything, I think yes. I don’t think it’s a coincidence that Obama’s economic policies are rated much more highly than the rest and that he’s got the best favorability number, even when you’re comparing Democrats to Democrats.
I don't know how I would answer this question. Obama was so long ago that I don't really remember how his economic policies affected me, even though I am definitely old enough to remember the Obama years
Count me in the skeptical group! For many years I was a busy Professional single woman raising children and taking post graduate courses. During those years I tried every diet plan to keep weight down but then I’d be exhausted and “need” a treat or 2 or 3 and 😫. No monitor was going to help. Finally retired, remarried, kids educated, I have found the one plan that works: Eat Less Food. Starting with cutting normal portions to half and eventually eating about 1/3 of that, one can get to a healthy weight. Of course this means no highly processed foods, low salt, and plenty of vegetables in soups and salads. During my working and bringing up children time of life, no number of AI beeps would have helped!!
I think your diet conclusions here are spot on, but I think the advances in AI make it possible to tailor advice in ways that people are more responsive to, e.g. talking to you about how you feel hungry and guiding you towards foods that promote satisty at low calories. Talking to you about your personal reasons for wanting to improve your diet, etc.
I do think the article is right that low income etc make this all harder, but in my case, finding a meal kit service that provides tasty low calorie meals has been a huge benefit to my diet. It's a physical product, but I needed the information that it existed. In my case, FB ads to my wife were the critical link here, but it could have easily have been an AI health assistant.
And the bull case for AI here is: think how good this could be if you had an actual human personalizing advice to you. It's the entire premise behind talk therapy and coaching of all sorts.
I think this project is still obviously somewhat of a moonshot obviously, AI capabilities are often overhyped, etc, but R&D moonshots are a good use of private capital. We shouldn't assume that they will work, but we should be cheering on the people trying to achieve good ends.
When speculating about the near-future of AI, and especially GenAI, replacing "AI" with "extremely naive and impersonal, but knowledgable and hard-working team of interns" is the right way to test a thesis.
So yeah, your bull case makes sense if an extremely naive and impersonal but hardworking intern would solve any given problem. Sometimes the answer will be yes, other times no, and many other times "yes, if designed very well".
That's a reasonable approximation of the core models, depending on your view of interns, but as you alude to, we build systems around these core models that allow them to go beyond them, eg by feeding them information about the person they're talking to, either plumbed from other systems, or extracted from users through (potentially multimodal) conversation.
IMO, "Yes, if designed well" should probably round to Yes, since big enough opportunities will attract people who can design these systems well.
So while I agree with your assesments of the core capabilities, I think it's too pessimistic about the abilities of systems with these models at the core.
Worth pointing out that, with very little known about the Thrive app, it isn't guaranteed to be harmless: "and at the very least, it likely won't lead to worse behavioral habits." If it recommends, for example, additional cancer (or other disease) screening, it may be a net negative, due to the very real dangers of false positive test results, as I wrote here (and as I experienced first-hand):
"(A) busy professional with diabetes who is struggling to manage blood sugar levels."
Welp, this is me, although I'm not technically diabetic, at least not yet. And for me, there is no use case for this sort of thing. I know which foods are healthy - and I already eat them! I know that I should exercise - and I do! I wear a CGM (the Abbott one) so I know, literally minute-to-minute, what my glucose levels are, and if they get really out of whack in either direction, an alarm goes off. I certainly don't need push notifications to "remind" me to eat a healthy meal. I'm not an idiot!
What I *would* like is a better understanding of why I'm in this situation in the first place, being a healthy, active person with a BMI < 20 and very good eating habits. Yes apparently there is some family history, as I have belatedly learned. But why does half a sandwich send it higher than a little scoop of ice cream? Why does it crash so low when I'm sleeping? Why is my A1C high even though day-to-day the levels don't seem that high? If there were some kind of sophisticated analysis that could help me figure some of this out on offer, I would be interested.
Having spent a month in Barcelona, I realized the biggest difference with Americans is that the latter _drive_ everywhere. Is the app going to nudge you to walk more?
I'm not saying the physical environment makes it easier (except in major cities, even L.A.), but anecdotally seeing people trying to park closer to the health club rather than walking across the parking lot, or as reported elsewhere, moving their car from one side of the strip mall rather than walking to the other store they are going to visit, tells you a lot.
I am a person who walks between strip malls because it seems dumb to drive that distance. But I totally get why people drive. No one expects you to be there, even with crosswalks and other pedestrian infrastructure, and you have to be constantly alert.
"the prevalence of chronic disease continues to rise"
Related: our country keeps getting older and disease screening is getting more accurate. Not saying that we couldn't be healthier, but if we want to reduce prevalence _rates_ we should probably boost immigration of young people. Did I YglesiAce the test?
I share everyone's skepticism, but I think Ben is overlooking where an AI coach could be most helpful: actual medical management of chronic diseases. I agree that reminders to get up and walk or eat a healthy snack won't be very helpful. However, I could absolutely envision a scenario where push notifications remind patients to order a refill and to take their medications. Depending on how much money is pumped into this tech and what wearables (among other monitoring parameters like a person's home BP cuff or lab data) are able to be linked to the app, a more aspirational goal could be individualized treatment of common chronic diseases.
I think the easiest example, bc it already exists at some level, would be to pair a continuous glucose monitor with an APP that manages an insulin pump. Or for someone with HTN, folks are prescribed a "smart" BP cuff that automatically sends BP readings to the app which tells the patient what dose of their antihypertensive to take (and it could be personalized based on their prior responses).
Obviously some tech and logistic hurdles to overcome there, not to mention privacy concerns, but I think there is some strong face validity to the idea that AI applied in this realm could be highly valuable.
The described app seems potentially useful on the margin but not transformative.
Also, it’s pretty unimpressive relative not just to the potential AI capabilities that Sam Altman has projected in his other public appearances, but also relative to actually existing AI tools now. In fact, you could pretty easily build an app with the described functionality that operates just fine without any AI element; you just need a scheduler that reminds people to take their pills, exercise, go to bed, etc (and plenty of stuff like that already exists.)
"Guys, we're gonna make America skinny again, one app at a time..."
Every time digital health interventions are invented, I think of those studies showing FitBit usage correlated with worse health outcomes...since the people most likely to track steps or whatever were aleady in worse health to begin with, and not even heroic levels of Cass and Sunsteining them can cross that gap. It's just really hard to instill conscientiousness if the baseline isn't there, if the social circumstances undermine it. One wants to believe really hard that Ignorance Is The Enemy and we can vanquish intractable societal issues with the Power Of Democratized Information (alternative moralistic gloss: bad outcomes are the result of """misinformation""" and/or corporate subsidies). But the knowledge has been available for decades, yet the horse still often won't drink. Honestly, a working version of ThrAIve Health would just give users $5 every time they acknowledge a notification and forget the nudges. But tie notifications to prosocial behaviours like not texting while driving, that way it'll still be a miens-tested program and no one will complain about moral hazard.
(...maybe they only work for managing communicable diseases? I seem to remember the State of California wanting to install an "automated contact tracing" app on my phone during the recent viral brouhaha, which probably would have worked in theory if I at all trusted the state with such surveillance. Many Such Cases.)
Once I read the "it will remind you to trade soda pop for sparkling water" parenthetical, my skepticism meter spiked, and it never went down throughout reading the article.
Most of us do not adhere to the optimal level of health management, because most of us want to do some things that aren't optimal. I know I have a few myself. (But not swapping out soda pop for sparkling water, I did that many years ago!) Some people do things that are very suboptimal to their own health. I think that that's regrettable, but ultimately it's their body and they're in charge of what they want to do to it.
Now, I don't want to completely downplay the opportunities that can be had with advances in informational medical technology. For those who want to make a concerted effort to improve their health, this could be a useful tool to help them out. But I think there will be plenty of people who just say, "Nah, I want to do this instead because I enjoy it.". I would think that a top tier model would account for this, and try to perhaps find a harm reduction avenue instead, but some still might just want to opt out of the technology altogether and not be bothered with bogging down their notifications further.
Yep, I was skeptical as well. But really, who knows! Hope that came across in the piece.
I think it did--your own skepticism was targeted at socioeconomic factors, which do matter and are a barrier. I just wanted to make sure to add the other barrier of people enjoying doing things that aren't optimal for their health.
I agree with you but my biggest health obstacles is that my brain is an unreliable narrator and if I can push back a little bit it gives me a bit more space to do the things that are optimal and once I have that foothold it can spin up a virtuous cycle.
I would not even say the brain is the issue but the predetermination to not accept discomfort. If you are responding to every time you get a hunger signal rather than accept that you should be hunger for sometime before a meal, then you will gain weight. So much of our food behaviors are habitual reactions rather than planned actions, so I don't how an app (AI or not) is going to help clean up a broken construction of dealing with food.
That's great that you recognize it for yourself, and if you think that it can help you, then by all means go for it. Others may end up disagreeing for themselves.
I agree people don’t leverage the information for their own gain. VO2 max is a good representation of one’s health (I get there’s other important factors too, but this seems under appreciated). If a person had a stationary bike or elliptical and went hard for 10 minutes a day they would significantly increase their physical wellbeing and fitness. These bikes can cost as little as 2-3 month of gym membership. Nike had it right, just do it.
I had the same reaction at first. But... is that true? Willpower is a limited resource. Sometimes it really is helpful to have someone else reminding you, constantly, of what you're supposed to do. It's not about knowledge, it's just having something external to help you. Personal trainers for exercise, for example, clearly help, and most of them aren't teaching arcane secrets of exercise. You can look up all the exercise advice on the internet for free. But it's still really helpful to have someone there with you in the gym, yelling "cmon! one more, one more!" Maybe AI can be like that for everyone, with diet.
I think the hope for an AI powered app rather than a simple reminder app is that it might be able to learn appropriate harm reduction for you.
I assume that the conversation with investors goes like this:
“So the user puts in their age and sex. Is that going to give them meaningful health advice?”
“No, no — but then we nudge them to input more personal information — diet habits, sleep, exercise. We offer *personalized* advice!”
“Okay, so does that make them healthier?”
“No, it’s still ineffective. So, we nudge them to put in more personal information— links to medical records, credit card data, dating profiles.”
“And that all makes them healthier?”
“No, no — totally ineffective. And the less it does, the more we nudge them to enter personal information. Look: we’re designing this to make you money. You’re not doing this for your health. “
I don't have a Simpsons clip for this one, but it totally sounds like something the writers would come up with.
"Social determinants of health" + diabetes really sets off my alarm bells because this is something where I think the left has too much faith in what is possible through government.
I know some young or middle-aged people with diabetes, including a 24 year old who died from it. And you could certainly point to social determinants being a factor - they're mostly lowish income and none have college degrees. But their root "social determinant" problems are that they are generally less capable, less conscientious people whose parents were similar, not that they didn't have access to opportunities.
To take the 24 yo - his parents are medical techs and hold down the job just fine. But their backyard has been completely overtaken by trash and neighbors because they've never cut the grass. When their kids were younger they used to poop back there. Only 1 of their 4 kids has a drivers license, and another one of them was jailed in a metal health facility for threatening to blow up a school.
Other examples are my cousin's husbands, the younger of which puts ranch dressing on nearly every single thing he eats, and the older of which is a landscaper who's been arrested several times into his 40s for getting in fights with strangers.
I feel like many educated liberals don't have these kinds of associations and imagine poor people with diabetes as black or brown people who've been discriminated against and turn to sugar to ease the pain or something. They could be "saved" with more government effort, or whatever.
And I don't think that's wholly wrong, government could do better or more, especially in the most low-income neighborhoods. But we've already grasped a lot of the low-hanging fruit. Convincing people not to put ranch dressing on their bagels is not a government or an AI problem!
> To take the 24 yo - his parents are medical techs and hold down the job just fine. But their backyard has been completely overtaken by trash and neighbors because they've never cut the grass.
This seems to get back to John from FL's comment about how best to evaluate an area--at the grocery store, do people return the dang carts to the corral?
They can't return the carts because poverty!
lol I love this. I grew up in a trashy neighborhood. In retrospect, the families who mowed their lawns and did mildly pro-social things like pick up trash if it was blowing down the street - all of their kids made it out. The shitbags who set off fireworks at 3am, whose teenagers girdled neighbors' trees as "pranks" etc. are all still there. As an adult, I vote for opportunities for people who want to make the most of them, and have no interest in spending time or resources trying to convince those who don't.
If the problem is they keep intending to do things better and keep forgetting to, then an AI app might be able to work by reminding them just at the relevant moment about their plan.
If the problem is that they don’t intend to do things better, it can’t fix that.
So what’s the explanation for comparatively healthier countries in Europe + Japan? Is it a matter of culture eating strategy for breakfast?
I feel like there are too many confounders to say very much.
To be clear about some of the things I was trying to say above, culture, habits, genes, personal choices, government programs / policies and the environment must all influence health. But enough of the above are either outside of government control (like personal control, genes) or not low-hanging fruit (redesign all American cities to be more walkable).
So "just adopt European social / medical policies", in my view, would do far less than the left often believes.
I recall that MY has attributed some of the difference, ironically I’d say but believably, to a higher income in the US.
Yeah - that might come out the most wrt things like auto accidents and especially drugs and alcohol.
I was mainly thinking of weight-related issues but yes good point.
I think nuclear incidents are relatively rare but earthquakes are definitely in the “occasional” category!
I feel like my optimism is more like the people who currently use health apps could be better supported by one with more features is where my optimism is not we can totally rework people.
I use these a lot because a lot of depression and anxiety is just vigilance against habit collapse.
Makes sense.
Arianna Huffington…
*takes long drag off cigarette *
Now there’s a name I haven’t heard in a long time.
Michele Bachmann was the name that came up this week that I did an Obi-Wan Kenobi impression for.
Top-10 Crazy Eyes legend.
I forget how I ended up at a Michael Huffington campaign speech in high school. But his sticker ended up on the jewel case for a Duran Duran CD. Things got REAL different for both of them after that.
I don’t eat healthy with my wife nagging me. I don’t think I’ll change when it’s Siri.
(Because nearly everyone already knows the healthy activities they should do.)
My health insurance keeps offering me a service where some person will email and call me about a diabetic diet. I know all the recommendations, and nagging will not make me more likely to follow it. The service will send me recipes as well. I Hate to cook and have undoubtedly see version of all the recipes.
This is the biggest thing.
And I absolutely hate healthy recipes. There is nothing so tedious and time-consuming to make on this earth because the target is health-hobbyist foodies. So they all require immense prep time rather than, like, all the ingredients purchasable frozen or in a can.
And this really is a thing. I've been on a whole kick of cooking classic casseroles and, well, you can get prep down to nearly zero with a recipe of "dump everything in a big bowl, mix, dump into a casserole dish, bake for 30 minutes, and you can feed six people for about $12".
Especially coming from Arianna Huffington, the obliviousness to the social determinants of health leaves a strong whiff Marie Antoinette: “The common people are suffering from chronic diseases? Let them have AI”
This app sounds, frankly, annoying AF. Why not just pop up photos of your mother with a disapproving look?
Got the Walz poll back from the field. Great numbers for him. You guys will see the full thing sometime next week but here are two fun tidbits to tide you over until then.
We polled a bunch of Democratic policies, some specific and some vague. On average, the specific ones had 47% support while the vague ones had 53% support.
We also did a battery of questions on how voters percieve Obama, Biden, Trump, and Harris on economic policy. Net approval for Obama's economic policy was +14; for Biden's policies, -4; for Trump's policies, -4; and for Harris' policies, +4. When asked how Obama's policies were for themselves and their families, by a +8 margin voters said they were good for them. By a +9 margin voters thought Biden's policies were bad for themselves and their families; for Trump it was +1 good; for Harris' proposed policies it was +4 bad. When asked how each person's policies were for the economy overall: Obama good +10, Biden bad +8, Trump good +1, Harris even (+0). Obama did better than Biden, Trump, and Harris with every single demographic on all the economic policy questions. And his net favorabilty was better with all groups too. GOAT.
Sounds like nostalgia is the biggest factor in attitudes—people have blissfully forgotten how bad their lives were when Obama was in office (not that it was his fault, he inherited the Great Recession after all). People are hyper aware of everything that is wrong right now, mainly because of inflation. Would people really rather have low prices and even lower wages (and a harder time finding work)? Houses were cheap thanks to the crash, but so many people were out of work not many could qualify for a mortgage. Of course it is true that the trend under Obama was one of continual improvement that was not halted by Trump taking office, until the pandemic. It’s really crazy, though, that there would be polling data on Harris’ economic policies, given that she has only begun to say anything about it in the last couple of days.
Agree on nostalgia being a factor. I do think the revealed preference is for lower prices and higher unemployment rather than the converse; just look at voter perceptions of the economy under Biden. Possibly because inflation hurts everyone whereas unemployment only hurts a few.
The exact wording was "the economic policies Kamala Harris has proposed." We're going to do another poll of the proposals she rolled out yesterday in North Carolina—I'm writing that one as we speak.
It’s striking how, in the 80s, Mankiw considered giving up macroeconomics precisely because that plausible relationship didn’t exist in the data!
https://www.journals.uchicago.edu/doi/pdf/10.1086/654160
The Democratic and progressive elites who decided to turn on their back on Obama after 2016 did quite the own goal. It's not even "if it aint broke don't fix it" it's "If it works, break it, then try to fix it by making it much worse" (+14 to -4!!!!!)
Wonder what the polling data were like at the time? Nostalgia is powerful.
If you look at the 538 tracker Obama’s always been more popular than Biden at similar points in their presidencies.
Biden was a pretty unserious candidate in 2008 (I’m sure he took himself seriously, but no one else did). In 2020 he was not-Bernie, which was all that was needed to beat Trump. He was never really personally popular.
Does comparing people's views of Obama's and Harris's economic policies really mean anything?
In general, I would far prefer an interview method where you just ask a voter "Which [Trump/Harris] policy do you [most like/least like] and please tell us why" then use transcription and some smart analytical tools to quantify their reactions, throwing out all answers where it's clear the voter doesn't have a clue and is just making things up. (Or coding them as "don't have a clue" and keeping that in the final reported results.)
Not really how polling works. If you throw out the answers where voters are ill-informed about policy you’re not going to have very many answers to work with. And ill-informed views do inform vote choice so you want to include them. Also a huge pain (and very expensive) to do open quesitons and then hand-code ~1,000 responses.
As to whether it means anything, I think yes. I don’t think it’s a coincidence that Obama’s economic policies are rated much more highly than the rest and that he’s got the best favorability number, even when you’re comparing Democrats to Democrats.
I don't know how I would answer this question. Obama was so long ago that I don't really remember how his economic policies affected me, even though I am definitely old enough to remember the Obama years
Count me in the skeptical group! For many years I was a busy Professional single woman raising children and taking post graduate courses. During those years I tried every diet plan to keep weight down but then I’d be exhausted and “need” a treat or 2 or 3 and 😫. No monitor was going to help. Finally retired, remarried, kids educated, I have found the one plan that works: Eat Less Food. Starting with cutting normal portions to half and eventually eating about 1/3 of that, one can get to a healthy weight. Of course this means no highly processed foods, low salt, and plenty of vegetables in soups and salads. During my working and bringing up children time of life, no number of AI beeps would have helped!!
I think your diet conclusions here are spot on, but I think the advances in AI make it possible to tailor advice in ways that people are more responsive to, e.g. talking to you about how you feel hungry and guiding you towards foods that promote satisty at low calories. Talking to you about your personal reasons for wanting to improve your diet, etc.
I do think the article is right that low income etc make this all harder, but in my case, finding a meal kit service that provides tasty low calorie meals has been a huge benefit to my diet. It's a physical product, but I needed the information that it existed. In my case, FB ads to my wife were the critical link here, but it could have easily have been an AI health assistant.
And the bull case for AI here is: think how good this could be if you had an actual human personalizing advice to you. It's the entire premise behind talk therapy and coaching of all sorts.
I think this project is still obviously somewhat of a moonshot obviously, AI capabilities are often overhyped, etc, but R&D moonshots are a good use of private capital. We shouldn't assume that they will work, but we should be cheering on the people trying to achieve good ends.
When speculating about the near-future of AI, and especially GenAI, replacing "AI" with "extremely naive and impersonal, but knowledgable and hard-working team of interns" is the right way to test a thesis.
So yeah, your bull case makes sense if an extremely naive and impersonal but hardworking intern would solve any given problem. Sometimes the answer will be yes, other times no, and many other times "yes, if designed very well".
That's a reasonable approximation of the core models, depending on your view of interns, but as you alude to, we build systems around these core models that allow them to go beyond them, eg by feeding them information about the person they're talking to, either plumbed from other systems, or extracted from users through (potentially multimodal) conversation.
IMO, "Yes, if designed well" should probably round to Yes, since big enough opportunities will attract people who can design these systems well.
So while I agree with your assesments of the core capabilities, I think it's too pessimistic about the abilities of systems with these models at the core.
Worth pointing out that, with very little known about the Thrive app, it isn't guaranteed to be harmless: "and at the very least, it likely won't lead to worse behavioral habits." If it recommends, for example, additional cancer (or other disease) screening, it may be a net negative, due to the very real dangers of false positive test results, as I wrote here (and as I experienced first-hand):
https://www.linkedin.com/pulse/thrive-ljubomir-buturovic-jfwmc/
"(A) busy professional with diabetes who is struggling to manage blood sugar levels."
Welp, this is me, although I'm not technically diabetic, at least not yet. And for me, there is no use case for this sort of thing. I know which foods are healthy - and I already eat them! I know that I should exercise - and I do! I wear a CGM (the Abbott one) so I know, literally minute-to-minute, what my glucose levels are, and if they get really out of whack in either direction, an alarm goes off. I certainly don't need push notifications to "remind" me to eat a healthy meal. I'm not an idiot!
What I *would* like is a better understanding of why I'm in this situation in the first place, being a healthy, active person with a BMI < 20 and very good eating habits. Yes apparently there is some family history, as I have belatedly learned. But why does half a sandwich send it higher than a little scoop of ice cream? Why does it crash so low when I'm sleeping? Why is my A1C high even though day-to-day the levels don't seem that high? If there were some kind of sophisticated analysis that could help me figure some of this out on offer, I would be interested.
Yes, this is exactly why I have it. My doctor suggested it and I was kind of surprised it was covered. I pay about $45 with insurance.
Having spent a month in Barcelona, I realized the biggest difference with Americans is that the latter _drive_ everywhere. Is the app going to nudge you to walk more?
"But [program], there's no sidewalks or bike lanes on this route, how am I supposed to safely walk or bike?"
I'm not saying the physical environment makes it easier (except in major cities, even L.A.), but anecdotally seeing people trying to park closer to the health club rather than walking across the parking lot, or as reported elsewhere, moving their car from one side of the strip mall rather than walking to the other store they are going to visit, tells you a lot.
I am a person who walks between strip malls because it seems dumb to drive that distance. But I totally get why people drive. No one expects you to be there, even with crosswalks and other pedestrian infrastructure, and you have to be constantly alert.
"the prevalence of chronic disease continues to rise"
Related: our country keeps getting older and disease screening is getting more accurate. Not saying that we couldn't be healthier, but if we want to reduce prevalence _rates_ we should probably boost immigration of young people. Did I YglesiAce the test?
I share everyone's skepticism, but I think Ben is overlooking where an AI coach could be most helpful: actual medical management of chronic diseases. I agree that reminders to get up and walk or eat a healthy snack won't be very helpful. However, I could absolutely envision a scenario where push notifications remind patients to order a refill and to take their medications. Depending on how much money is pumped into this tech and what wearables (among other monitoring parameters like a person's home BP cuff or lab data) are able to be linked to the app, a more aspirational goal could be individualized treatment of common chronic diseases.
I think the easiest example, bc it already exists at some level, would be to pair a continuous glucose monitor with an APP that manages an insulin pump. Or for someone with HTN, folks are prescribed a "smart" BP cuff that automatically sends BP readings to the app which tells the patient what dose of their antihypertensive to take (and it could be personalized based on their prior responses).
Obviously some tech and logistic hurdles to overcome there, not to mention privacy concerns, but I think there is some strong face validity to the idea that AI applied in this realm could be highly valuable.
All of the things you described are problems that software can solve, and where an LLM adds basically nothing on top of standard software.
The described app seems potentially useful on the margin but not transformative.
Also, it’s pretty unimpressive relative not just to the potential AI capabilities that Sam Altman has projected in his other public appearances, but also relative to actually existing AI tools now. In fact, you could pretty easily build an app with the described functionality that operates just fine without any AI element; you just need a scheduler that reminds people to take their pills, exercise, go to bed, etc (and plenty of stuff like that already exists.)
"Guys, we're gonna make America skinny again, one app at a time..."
Every time digital health interventions are invented, I think of those studies showing FitBit usage correlated with worse health outcomes...since the people most likely to track steps or whatever were aleady in worse health to begin with, and not even heroic levels of Cass and Sunsteining them can cross that gap. It's just really hard to instill conscientiousness if the baseline isn't there, if the social circumstances undermine it. One wants to believe really hard that Ignorance Is The Enemy and we can vanquish intractable societal issues with the Power Of Democratized Information (alternative moralistic gloss: bad outcomes are the result of """misinformation""" and/or corporate subsidies). But the knowledge has been available for decades, yet the horse still often won't drink. Honestly, a working version of ThrAIve Health would just give users $5 every time they acknowledge a notification and forget the nudges. But tie notifications to prosocial behaviours like not texting while driving, that way it'll still be a miens-tested program and no one will complain about moral hazard.
(...maybe they only work for managing communicable diseases? I seem to remember the State of California wanting to install an "automated contact tracing" app on my phone during the recent viral brouhaha, which probably would have worked in theory if I at all trusted the state with such surveillance. Many Such Cases.)