Eli Dourado AMA
Hi everyone. I’m Eli Dourado, a senior research fellow at the Center for Growth and Opportunity at Utah State University. I work on policies to advance hard technologies that generate economic growth. I have published in top newspapers, negotiated supersonic standards, derailed an international treaty, and spend too much of my time on Twitter. I got my PhD in economics from George Mason University.
Ask me anything! I will be on here from Monday, January 30 for a few days, and will endeavor to get to your questions each day during that time.
Use the comments below to add questions, and upvote any questions you’d particularly like me to answer.
- The Roots of Progress 2023 in review by (LessWrong; 31 Dec 2023 18:16 UTC; 22 points)
- Eli Dourado AMA on the Progress Forum by (LessWrong; 28 Jan 2023 22:18 UTC; 19 points)
- Matt Clancy AMA on the Progress Forum by (LessWrong; 12 Feb 2023 20:23 UTC; 17 points)
- Progress links and tweets, 2023-02-01 by (LessWrong; 2 Feb 2023 2:25 UTC; 10 points)
- Progress links and tweets, 2023-02-08 by (LessWrong; 8 Feb 2023 15:52 UTC; 10 points)
- Jason’s links and tweets, 2023-02-08 by (8 Feb 2023 15:52 UTC; 5 points)
- Jason’s links and tweets, 2023-02-01 by (2 Feb 2023 2:24 UTC; 4 points)
So many of the regulatory/policy barriers to progress seem so daunting. Using the “Important, Tractable, Neglected” heuristic, what are the top opportunities to unblock progress? Put another way perhaps, if you were writing a priority list for an organization like the Institute for Progress or Balsa Research, what would you go after?
I kind of did this analysis in 2019 on “how to move the needle on progress” and landed on health, housing, energy, and transportation as important sectors to fix.
If you think about it in productivity terms, in general equilibrium, low-productivity-growth sectors will tend to get bloated as a percent of GDP, while high-productivity-growth sectors will tend to shrink.
I still think the 2019 analysis is basically right, although I would emphasize one particular aspect of tractability, which is having a specific solution in mind. Tom Kalil talks about this as a test of policy maturity: suppose you have a 15-minute meeting with the President of the United States, and after the meeting the President is willing to call somebody and tell them what to do. Who do you have him call and what do you have him tell them to do? Until you have an answer to that question, your policy solution isn’t mature.
I think there’s a division of labor in the policy world between the more researchy and more activist groups. The researchy people should be working to discover mature policy ideas (in the Tom Kalil sense) and then the more activist groups should be working to get them implemented.
So for the research side, who are starting out without mature policy ideas and trying to generate them, tractability isn’t really a concern, it’s more about importance. The goal is to generate something tractable. The more activist people need to think more about taking the mature policy idea and running with it, and for them, tractability (political viability, etc.) is more important starting out.
Progress is so hard to come by in the policy world that I don’t think we should disqualify anything for not being neglected. Even housing/YIMBY stuff, I’m happy for more people to go into it if it gets us over the line.
So policy researchers should work on big industrial sectors like health, housing, energy, and transportation (and major cross-cutting issues like immigration and permitting), and try to come up with mature policy ideas that increase productivity. Then the more activist groups should take the tractable ideas and amplify them and try to get them over the line.
If I’m writing a priority list for another org, the first question I’d ask is whether you want to be a more researchy org or a more activist org.
Thanks, Eli! This is a super helpful framing to me as I think about our role here at The Roots of Progress.
Follow-on question: when you say “researchy” do you mean academia—or do you mean groups in the more public intellectual policy space (think tanks) that take on more of an explainer rather than activist bend?
The latter, although sometimes they overlap with academia. For example, CGO and Mercatus publish a lot of academics and are situated within universities.
There are two magic buttons, as follows, but you can only press one. Which would be better for progress and why?
We instantly get the ideal legal/regulatory/policy environment for progress, across the board (this button does not affect science or R&D)
We instantly get huge scientific/R&D breakthroughs: cure for cancer and aging, nanotech that works, fusion that works, benevolent AI (this button does not affect anything social, so all these things would face today’s regulatory environment)
Given the trade you’ve laid out, I’d take the scientific breakthroughs.
I think there is no agency to regulate nanotech, so it would be a “born free” industry, and we’d see a lot of rapid progress. Benevolent AI too. On the cancer and aging cures, yes, FDA is broken, but they’d get through approval in several years, and then we’d have them.
I do think, however, that the policy environment is worth many years of R&D breakthroughs, perhaps 10 or more. We’d get a revitalized transportation and energy industry, dirt cheap housing, better consumer health tech, and a faster rate of R&D development going forward. It wouldn’t take much unbalancing of the scales to make me flip the answer.
Such a great question, excited to see Eli’s answer.
In your post from 2019 on moving the needle of progress, you mention health (or better, wellness) as one of four key levers toward progress, and you highlight patient empowerment and using data from wearable devices as potentially big opportunity.
Do you have any thoughts on what is stopping this from happening? It seems that using data to empower people to live healthier is a win all around: better quality of life/more energy/less pain for the individual, lower medical cost for the insurance and employer, and higher worker productivity. Why aren’t we, for example, seeing health insurance plans that incentivize healthy behavior that can be tracked via wearables and provide meaningful financial incentives (similar to safe driver programs)?
I think the biggest obstacle is FDA clearance for these devices.
The FDA seems to be concerned about people using consumer-grade products to make medical decisions. Let’s say Apple or Google release a smart watch with non-invasive blood glucose capability. Maybe it’s not perfectly accurate, but still useful information for non-diabetics to see how their blood glucose spikes after eating and to monitor the speed at which the body clears out the glucose.
If a diabetic customer starts giving themselves insulin shots based on the watch instead of a measurement from a medical device, that could be bad. Therefore, FDA is very restrictive about devices that report medically-relevant facts, even if there are disclaimers that they should not be used for medical purposes.
So it’s a slog to get the new non-invasive tech to be as accurate as medical-grade tech, to prove that they are equally accurate, and/or to get FDA to sign off on disclaimers that say the data shouldn’t be used to administer medication, etc.
I still think we’ll get there. There’s some info online about the Apple/Rockley Photonics partnership. You can expect a future Apple Watch to have not only its current sensor suite but also measurements for blood pressure, blood alcohol, lactate, and glucose. Blood pressure in the next 2 years, the rest maybe a couple of years later.
Why aren’t insurance companies paying for it yet? I think the current device sensor suite isn’t high enough on cost-benefit for them yet. As prices come down and the new capabilities are added, it seems like a no-brainer. Like in 2035 a device with all the capabilities I described above might be $100. Probably worth it then.
Thanks for the quick response, Eli. My follow-up is similar to Jasons: I’m wondering not so much why insurance companies don’t pay for these devices right now, but more why there isn’t a push to use them to financially reward or incentivize healthy behavior or outcomes.
For example, if it costs an insurance company $10K more to care for someone with diabetes than someone without, what if the insurer offered the patient a deal: if via non-medication means you reverse your diabetes (as measured by insulin and hemoglobin A1C or HbA1c test), we’ll pay you $5,000 (or, if they pay privately for their insurance, we’ll give you a year-end cost refund of $$).
Or, different approach: if you do specific behaviors that we know will lead to your diabetes improving, we’ll pay you a certain amount per month each month you meet the targets consistently (e.g., wear a fitness tracker and exercise 5x per week for 30 min at vigorous intensity and wear a continuous glucose monitor and stick to a 10-hour feeding window by practicing intermittent fasting for at least 20 days/month).
Of course, the details would need to be worked out and there are lots of questions (e.g., how does this work for people who are already metabolically healthy vs. those who aren’t). But there are companies playing this space at a small scale, using biofeedback and clear lifestyle incentives to improve health outcomes (albeit without insurance pay-back), such as VirtaHealth, Levels, and HealthyWage, and commercial supportive counseling programs like Noom.
Given how immense the cost of metabolic disease is for individuals and society I’m surprised that there isn’t a larger effort to use these trackers to fundamentally change how the incentives work. Wouldn’t it be better if insurance companies actually helped people get healthy, rather than pay for ongoing medication for multiple chronic diseases?
I’m really curious if anyone has done any digging on the regulatory barriers that would make it hard to make this happen. (Or maybe this is an untapped business opportunity for someone, if there aren’t any major regulatory hurdles!)
It looks like certain wellness programs that are allowed, but there are limits on what you can do if you make it outcome-contingent. https://www.dol.gov/sites/dolgov/files/ebsa/about-ebsa/our-activities/resource-center/publications/caghipaaandaca.pdf
Thanks for that link, Eli. This is exactly the type of context I was looking for. It woulds like there is a regulatory hurdle here with significant potential liability if the program were to get challenged as not meeting those requirements both on the actual program design and on the documentation.
Are there any restrictions on what insurance companies are allowed to do with this kind of info? Health insurance is highly regulated too.
I’m not sure what the privacy implications are, but they can definitely give you the devices for free if it’s cost-effective for them to do that.
I meant more on the question of financial incentives for metrics. Basically, charging healthy people less / charging more for risk factors. Are you allowed to do this? I think some amount of this is allowed in some jurisdictions, but are there crucial limitations on it?
The ability to charge people more and less based on observed (but not demographic) characteristics got pretty limited by the Affordable Care Act. I’m not sure of the details, however.
If you were to draft a set of cause areas for the progress studies movement, what would be high on the list?
1. Deregulate land use (YIMBY stuff)
2. Make transportation insanely great: eVTOL, supersonics, small airports with minimal screening, autonomous dynamic bus service
3. Lower the cost of clinical trials and expand freedom to go around the FDA through informed consent
4. Reform permitting/abolish NEPA/end vetocracy
5. Energy abundance/fix the NRC/fix the nuclear industry/expand geothermal/deploy solar
6. Make government that works and is run by grown-ups (I am a big fan of ranked choice voting for this)
7. Big increases in immigration, with concessions to the xenophobes that immigrants probably need to speak English and get deported if they commit serious crimes
8. End make-work policies that are embedded in almost every sector
9. Make sure safety rules are at least actually adding safety instead of safety theater
Is there good writing somewhere on how to lower the cost of clinical trials 10x? If we focus on the actual cost-lowering, rather than pure deregulation, it’s a rare area where I’ve never even seen someone who seems to know what to do.
Yes, from what I hear, it seems very hard. I’d point you to some recent pieces:
https://www.statnews.com/2022/11/03/why-were-not-prepared-for-next-wave-of-biotech-innovation/
https://milkyeggs.com/biology/why-are-clinical-trials-so-expensive-tales-from-the-beasts-belly/
https://www.nytimes.com/2023/01/27/business/paradigm-startup-clinical-trials.html
I’m also curious to know if you think that centralization, the way that other movements (I’m thinking of effective altruism) have specific cause area prioritizations that proponents tend to follow, is a good or a bad thing. In your opinion, what qualities make an effective modern movement that can actually get things done?
Great question.
I am glad the progress movement is still decentralized and organic. It’s more a community of fellow-travelers than a centralized organization setting priorities and allocating funding. I feel like I gain a lot from people in the community who are pursuing very different approaches than I am, and I don’t want that to stop.
I think being organic is better for influencing the culture in the long run. For getting specific things done, if we ever agree on what is to be done, we may need to think about some light centralization at some point.
Given that the supreme court frequently ruled against NEPA, why aren’t there more cases involving NEPA in front of the supreme court to curb its excesses?
It’s true, the Supreme Court has ruled against vetocracy with NEPA every time, usually unanimously. I think for potential litigants, there isn’t much value in going all the way to the Supreme Court. It’s possible the court won’t hear your case, so you have to take steps to comply with the lower court’s ruling anyway. Once you’re doing that, spending more on litigation isn’t going to get you anything; you can always just fix the EIS and move forward.
In other words, taking a case to the Supreme Court for the purpose of setting a new precedent is a public good, and a lot of people don’t supply public goods all by themselves.
But there are organizations that bring cases as a public good and they pay for the litigation. On the left, ACLU. For libertarians, the Institute for Justice (IJ). IJ identifies laws or regulations they want to dispute and then find a sympathetic plaintiff to represent. Someone like you would probably do a lot of good by convincing IJ to find some NEPA cases to bring.
The party of record is a federal agency, though. I’m not sure IJ can defend them.
Individuals don’t have standing to sue when NEPA holds up their projects for years?
The NEPA lawsuit is brought by an environmental org against an agency. I could be wrong but I don’t think a different party can appeal the decision.
This list the existing NEPA lawsuits before the supreme court and while some of them have government agencies as the plaintiff many don’t.
Monsanto Co. v. Geertson Seed Farms is one example.
Good catch. Looks like in the district court Geertson sued federal officials as per usual but Monsanto filed a motion to intervene and join the case.
That still seems surprising to me. The Koch’s for example seem to be very willing to spend a lot of money think tanks to advance the public good as they see it.
Is there a reason why billionaires like the Koch’s who are willing to spend money on policy changes don’t care about NEPA?
This is great to understand.
Hi Eli! You told me a little about this the last time we chatted in person, and I’d love to hear more: Could ould you elaborate a little more on your “theory of change”? That is, how do you see your day to day work resulting in the kind of improvements you want to see?
And for experienced working professionals looking to maybe make a more impactful career switch: any underrated or not well known career paths that you wished more people embarked on that could help? And any particular industry backgrounds/expertise that you wish were more involved in the policy conversations?
As a writer: influencing other people, building consensus on what the problems are, building a network of people that are aligned.
As a researcher: coming up with highly-specific policy solutions to an important problem. Ideally, this would be a small, non-controversial provision that someone could slip into a bill unnoticed.
So for example, I have written a lot about the problems with NEPA and permitting, and I think there’s been a consensus developed among a big chunk of the political spectrum that it’s a real problem and we need to fix it. At the same time, I have been trying to push a specific fix for geothermal permitting, which is to give it the same categorical exclusion that oil and gas has.
Two different kinds of change, and I try to do both, but succeeding at the latter is rare and extremely valuable such that if you do it only a few times in your life that is a successful career.
On career switching: I would advise people to look less at creating a coherent career, where there is a logical progression from one step to the next, and to instead just find a job that interests and inspires you to do your very best work. My own career has been pretty haphazard: I was going to be a professor, then no just kidding I’m going to do nonprofit policy research, then oh no I am going to work at a startup, then back to policy research. None of this was part of a deliberate plan.
One heuristic that I think works well:
1. What do you think is the single most interesting thing going on right now?
2. How can you put yourself at the center of that thing?
Do you have any views about wireless power beaming, both shorter-distance charging and longer-distance transmission? It seems to me that this is a potentially underrated technology, if it can unlock the potential of other technologies that currently face energy storage or transportation constraints (EVs, robotic exoskeletons, perpetual-flight drones & aerial platforms, remote wind/solar etc).
I got interested in wireless transmission for space-based solar. A lot of people have had doubts for a long time about whether the math works for space-based solar, but both panels and launch prices have plummeted, so people are giving it a second look.
One of the things about wireless transmission that could add value to space-based solar is being able to shift output on the fly from one receiver on Earth to another on a millisecond-to-millisecond basis. I thought that was pretty cool.
I haven’t really looked at it for terrestrial applications, though.
Policy barriers aside, speaking strictly from considerations of technology and economics, what is the ideal near-term future for energy? Nuclear, geothermal, solar? Maybe even solar-powered fuel synthesis like Terraform Industries is doing? Or what combination of the above?
A very important question is how long solar prices can continue to drop.
Assuming it continues a while, I have questions about whether it makes sense to transmit electricity long distances in such a world. A lot of smart people think transmission is very important to the clean energy buildout, but I don’t know. Transmission adds a fair bit of cost, and if solar gets cheap then it might make sense to pay the rooftop premium rather than the transmission premium.
So if solar keeps dropping in price, it may make sense to have rooftop solar everywhere + off-grid solar to power industrial applications.
Gigawatt-scale nuclear I think we could do for LCOE of 2¢/kWh if the industry and regulations were not so dysfunctional. Modular reactors will always be more expensive than that (maybe 4¢ best-case scenario), but the advantage of modular is that you reach some level of scale in manufacturing and deployment, which is where gigawatt-scale has really sucked (every gigawatt plant is bespoke). Modular is also better because you don’t have to do as much transmission as in a GW-scale plant.
If we get good at drilling holes in the ground, I think 3¢/kWh almost anywhere on the planet would be a good target for advanced geothermal. Also comes with the advantage of not having to worry about spent fuels and nuclear proliferation. Geothermal is also fantastic for low-grade heat needed for certain industrial processes like paper mills.
Wind is already pretty cheap, but it relies heavily on long-distance transmission, which as I’ve noted is a headache.
For mobile applications, high-density batteries are definitely possible. Batteries that have near the energy density of liquid hydrocarbon fuels have already been made in the lab, the challenge is switching over the manufacturing system and reaching scale.
Synthesizing liquid hydrocarbons is a great solution, especially until really high-density batteries arrive. I believe I bought the first quantity of zero-carbon jet fuel in the world when I was at Boom.
For a center hosted by a university, what cut of funding does the university normally take?
I am blessedly exempt from having to deal with any financial or management issues at the CGO, so I don’t know. We do get along with the USU administration really well, though.
What are your thoughts on the Inflation Reduction Act passed last year? I have read that it is protectionist in nature and could divert investment from Europe and other places.
Also, what do you think about the potential of small modular nuclear reactors?
IRA:
- Very expensive, and we should take fiscal responsibility more seriously than we do.
- Has some good stuff in it.
- Will only translate into significant change in the real world if it is paired with permitting reform and other policies focused on deployment. Right now we are basically subsidizing companies to push through the headaches associated with getting to market, but it makes more sense to reduce the headaches.
- It is more protectionist than I would like, but Europe is the most mercantilist place on Earth, so I don’t have a lot of sympathy for them on this.
SMRs:
- The lowest possible LCOE that you can get with SMRs is higher than the lowest possible LCOE that you can get with gigawatt-scale plants.
- That said, SMRs could theoretically solve the biggest problem that we have in nuclear, which is that we don’t churn out identical plants in high numbers.
- Also, SMRs allow lower-scale plants, which means less reliance on transmission infrastructure, which is important.
- I’d like to see ultra-small reactors become a thing. Kilowatt-scale. Generator replacements. Portable.
- To really make nuclear portability work, it would be good to have solid-state thermal conversion, like thermoelectric generators or thermophotovoltaics. These would be more compact than turbines, and could come down in cost faster.
What formative things between the ages of 14 and 24 made you who you are today?
Getting online in the mid-90s was huge. The web was tiny back then, but it was still such a window to the world. I tinkered with everything, taught myself HTML, played with hacker tools, read The Anarchist Cookbook, made myself a Geocities page, etc.
The other formative thing was in college, discovering economics, which was a way of thinking that comes completely naturally to me. Finally, people are making some sense, I thought. In my early 20s, the Econ blogging scene was crucial. These are my people, I thought, and I ended up putting myself at the center of that group by going to GMU for a PhD.
Referencing your recent AI article (which is great!):
How much of the problem of digital technology being hard to implement productively because of social/legal/policy stuff:
A path dependency issue: digital technology just has to exist substantively before the social/legal/policy environment is generated, improved and optimized to accommodate it?
To the extent this path dependency exists, do you think we could be doing more to prime the social/legal/policy environment for new technologies preemptively? Is better anticipation of the social/legal/policy needs of digital technology feasible? Or are the main gains (or least reckless approaches) to be found in speeding up the accommodation process for digital technology once it exists, and its practical applications become clearer.
For 1 & 2, how do you see the answers varying across different sectors of the economy?
I don’t think path dependency is the right way of looking at it. I’d frame it rather differently:
We are doing a bunch of clown stuff that is holding back productivity improvements all the time. There is nothing about that is unique to AI. However, it’s possible that it will become especially apparent that we are erecting all these obstacles ourselves as we observe AI getting very productive in unregulated or otherwise functional sectors.
Absolutely, we should be dismantling the clown policies proactively, but it isn’t proactive with respect to AI particularly, it’s just that we should not have clown policies in the first place.
Your latest article was something else! I would love to learn about your process of writing such essays from how you find the graphs to how you create the entire narrative. I’m especially interested in this section, “Are cargo airships startupable?” I enjoyed how you broke everything down using data, math, and other assumptions.
Unrelated: let me know if you’d be interested in a discussion/debate with a professor (in front of the entire class) with whom I’m taking a class about the future of energy. My professor thinks we should avoid growth, not use electricity and make it really expensive.
Thank you!
The key to the process on this one was first spending several years thinking about cargo airships and how you could make a business around them. I made my first phone call asking if I could buy a cargo airship over three years ago.
All charts other than the one I cribbed from the Review of Maritime Transport are original, either to me or to the engineer that did the trade study.
Once I sat down to write, the narrative came pretty easily. I just wrote what I thought, trying to explain why I thought airships are interesting and why they could be profitable and why it’s hard to get there from here (i.e., why I’m not starting an airship startup).
On cargo airships: has anyone analyzed designs that remain towed/tethered to surface routes/vessels?
For example, could an electric locomotive powered by ground lines tow far larger amounts of cargo via a tethered airship? Or, an oceangoing container ship tow more tonnage in its air-trailer than in its holds?
Alternatively, could such tethers-along-routes supply electricity to self-propelled airships, minimizing onboard fuel/generation weight?
Or, could airships receive beamed power from elsewhere? For example, tightly-planned routes might receive beamed power from surface stations – possibly even mid-ocean wind/solar/tide platforms. Or could arbitrary routes receive beamed power from solar-power satellites?
Might short-range drones, or even smaller airship lift platforms, enable loading/unloading smaller cargo onto larger airships without explicit landing stops- making the largest platforms permanent ‘conveyor belts’ in the skies?
Similar to your map showing where high-elevation mountains may present ‘no-go’ areas, are there regions where the constancy or unpredictability of high winds create no-go or highly-variable service limits? When an envisioned large ships hits unexpected/unpredicted high winds, what does it do to adapt? (Can it usually wait it out or ride the winds for mere delays, or are things like emergency landings or even scuttling the ship within the likely scenarios, under large scale operations?
If I understand the tethering question, I don’t think that would help. A train or container ship can already carry many times more cargo than an airship can.
When I looked at space-based solar power, I was struck that the wireless transmission was not much denser than solar (otherwise, they would pose a danger to the ground if they missed the receiver). I think putting thin-film solar could buy its way onto the top of the hull if it was cheap and light enough. I have heard about very high-powered laser propulsion systems for high-speed aircraft, but those seem a long way off.
If you’re talking about drones using rotary lift to carry 1 container at a time, those might need to be pretty powerful to work! I am not sure it calcs out. Would be cool if it worked though.
I am not aware of any no-go areas for wind. Would assume Antarctica might be bad, but it’s already no-go for altitude and few people live there anyway. I think you would want to design the mission to route around storms and just get there a day late if necessary (with a contract structure that reflects this possibility). You probably never want to scuttle the ship since it’s likely expensive compared to the cargo value.
So theoretical airship-cargo is still way more expensive per container or ton than either rail or container-ships – it just wins by speed or having more endpoints (like trucking)? Hence, ‘airship trailers’ provide no chance of incremental capacity boosts for surface vessels, if still limited to their speed/endpoints?
Capping laser/microwave/etc beamed power density to what’s safe for failure situations where it’s “missed the receiver” seems prematurely restrictive. What if misses are so rare, & so easily detected/ended-instantly, that such a limit is an inefficient way to increase safety compared to other tactics? If the big market is over unpopulated oceans, is the concern for brief rare misses that high? (And is the “long way off” for things like laser-propulsion or power-delivery really longer than the other engineering/regulatory hurdles involved?)
Drones (including airship-bouyant drones, not just multicopters) to ferry full containers to passing-by megaships would be interesting – but I was thinking as small as individual packages, dropping and rising from households & individual businesses. (At some margin, can automated megaships be warehouses/fulfillment-centers?)
By my intutions, I find wind issues underdiscussed in these next-generation airship visions. It amazes me the wind conditions in which winged flight remains tractable – but those craft seem to rely heavily on actively avoiding the worst conditions, & their own momentum/strong-propulsion. Airships feel at much greater mercy of winds, for both predictable-service/efficiency & safety. Deeper analyses of how frequently there’d be delays, emergency groundings, service outages, etc from wind conditions would better help sell the vision.
I think airships could in principle approach rail costs but it would add a lot of complexity relative to just running another train on the same track. Big container ships are always going to be cheaper, I think.
FAA and pilots get mad about people pointing laser pointers into the sky.
I agree winds are super important and must be designed for and planned for in routing. Using them for sailing (added propulsion) is also promising.
Stably-positioned, heavily-capitalized, professionally-managed, presumably-licensed beamed-power infrastructure, such as terrestrial power stations or solar power satellites, aren’t really like random malicious or careless people with portable lasers. So if beamed-power otherwise becomes technically practical/beneficial, this seems a safety/perception/regulation challenge not especially larger than the many others involved here.
Yes, sounds plausible to me.
Hi Eli, I read your piece on the regulatory barriers to AI progress having material impacts on society. For me this pushes things in the direction of “we’ll have more AI automation of AI R&D before big societal trends in job automation”, which could imply faster AI progress generally if labs are focused more on their own AI → research automation → better AI feedback loop. I do think that an AI that could perform basically any jobs (not requiring hands) as well as a human for pennies on the dollar would radically transform society, but maybe we don’t see as much change in AI systems until then. This Metaculus question (https://www.metaculus.com/questions/3698/when-will-an-ai-achieve-a-98th-percentile-score-or-higher-in-a-mensa-admission-test/) on when an AI will get a Mensa-worthy iq score (current prediction: April 2028) suggests to me that we’re not far away from AGI. What do you think? My sense is that you’re much less bullish on AI progress than e.g. the LessWrong or EA communities.
I think the kinds of tests that prove that a human is intelligent or sentient or whatever are not the same as the kinds of tests that prove a computer program is sentient.
For example, imagine a test where we timed the test-taker on how long it takes to multiply two 8-digit numbers together. For most humans, this would take several minutes. For even a dollar-store calculator, it would take under a second.
For many decades, Alan Turing’s proposal that a computer that could converse indistinguishably from humans would be a sign of human-level sentience and intelligence was widely accepted. I myself thought, “Sure, sounds good,” when I first heard of it.
But actually, it turns out that carrying out a conversation for machines is easier than we thought. There is no real cognition going on inside ChatGPT. It is spitting out answers based on a statistical function trained on encoded inputs and outputs.
I think it is quite possible that an AI will achieve a 98th percentile score on a Mensa test by 2028 (maybe earlier). What I don’t think is that that will be a sign of human-level sentience or intelligence. It’s a sign of being able to mimic a few salient aspects of human intelligence.
To get to parity with human brain experience, we need several orders of magnitude higher computational efficiency to match neurons. We don’t need to get there all the way on efficiency; we can do some by burning more energy. Even so, it will take a couple of decades in my estimation.
And even so, there is still the possibility that we don’t really understand how the neurons work and we could be way off base! Michael Levin has pointed out that a caterpillar essentially disassociates its brain to become a butterfly, and yet somehow it retains at least some memories. I think we are far from really grokking it.
Huh, it’s hard for me to imagine reaching a 98th-percentile IQ score without the ability to do lots of cognitive work (I’m not talking about some model fine-tuned on IQ tests or whatever, just a general language model that happens to score well on the test). I have different intuitions about the calculator example: the point I take away from it is...we use calculators all the time! I’m perfectly content calling calculators a transformative innovation, though these language models are already much more general than the calculator.
Re: “There is no real cognition going on inside ChatGPT. It is spitting out answers based on a statistical function trained on encoded inputs and outputs.” This seems like a No True Scotsman that will keep you from noticing how their capabilities are improving. SSC’s take on GPT-2 was good for this, and imo got extremely vindicated when the GPT family went from an interesting toy to being able to create real economic value.
Re: Have you read Gwern’s stuff on machine learning scaling? All of the “we don’t really understand it” takes on a very different tone when you read his “deep learning wants to work” take. A technique that AI researchers disdain because it doesn’t match their love for theory, that works anyway, that then the whole SV community realizes is really promising...strikes me as something real and useful we accidentally discovered in the world. That we don’t understand it doesn’t stop it from working, that every basic little trick we try yields more fruit suggests that the fruit is really extremely low-hanging. For me, it’s worrying because I think we need good theory to learn how to control it, but the basic case for this being a thing doesn’t seem in question.
What are some projects you have plan to work on in the near-term? In the long term?
What are some projects you wish someone else would work on?
In the short term, I am doing some geothermal projects, and then possibly writing a book.
Longer term, I have a lot I’d like to do. If I ever found myself in a position where I could seize real power, I would take it and use it to promote progressy things.
I really enjoy the part of my job where I talk to people working on promising hard tech startups; often they need a bit of advice or some introductions to people in my network. I’d like to do more to help them, as they are often brought into contact with the barriers and obstacles I spend so much time thinking about.
I do wish someone else besides me would start a cargo airship company and do it right, with iterative design and a focus on getting to 500+ tons of cargo as quickly as possible.
How would you restructure the production of education and research? Universities are clearly pretty inefficient. How to improve them? Are there any politically feasible plans? Will AIs just obviate internal reform by creating an entirely new form of education and research?
The biggest problem that I see in college education is that most people don’t actually want to learn very much. College social life is undeniably fun, and although most people find a few classes they enjoy, they’re there for the experience + the credential.
I don’t know how to fix it because I think there is demand for the current system, but there should be at least one college with unlimited enrollment that is rigorous enough that it weeds out the people who aren’t giving it their best effort. Maybe it should be self-paced, with a massive total learning requirement so that it takes the best students four years and others longer. The degree would be worth more than other degrees in the end because it is so rigorous.
I think recorded lectures would be a part of it, but you’d probably still need/want human tutors and performance coaches. Motivation is often the scarce factor; if it wasn’t you could learn just about anything with a library card.
AI can help with both content and motivational scripts but I don’t think it’s a radical difference from what we can do now with non-AI methods.
If Rigorous U took off this would separate researchers further from undergraduate students. Research labs could be separate institutions even. I’d like to see researchers spend more time exposed to industry. At least once in their career they should take a basic science breakthrough and try to take it all the way to commercialization. Yes, there are gains from specialization, but there are also gains from a broader range of experience and contacts.
I’m not optimistic that any of this is socially or politically feasible.
Peter Thiel recently argued that a slowdown in progress is overdetermined, and due in part to a widespread fear of progress itself. Should we be focusing on the mass psychology needed to support progress? What might help?
I think it’s true to some extent that the masses exert some demand for stagnation.
The way I’ve been thinking about it is that laws and norms are ways of solving iterated prisoner’s dilemmas. But because of loss aversion, there isn’t symmetry in the kinds of PDs that get solved this way. The “prevent something bad from happening” PDs get solved more than the “make something great happen” PDs do. (This is essentially the Nietzschean distinction between slave morality and master morality, applied to laws as well as morals.)
I don’t think the masses are ever going to change. Rather, I think elites need to compensate and be advocates for great things happening. There needs to be an elite conspiracy to elevate humanity far above where it would otherwise be willing to go.
A lot of policy change can happen with only elite consensus. In my work I focus a lot on small changes that need not concern most people, like a categorical exclusion for geothermal energy. Or changing how the Department of Energy does contracting for demonstration projects. I think a promising way to increase progress is to subtly remove a lot of small obstacles like this.
Maybe if we can get a few great, visible achievements it will soften mass opposition to some degree.
What’s your theory of political authority? Do citizens have a moral responsibility to obey government more than other organizations? What are the proper limits to government authority? How does this figure into your policy recommendations?
I don’t think there is any account of political authority that isn’t defeated by the standard objections.
For purely prudential reasons, I think people should give some deference to governments as long as the government is mostly functional and aligned with the population. Living in a state where the government is ineffective is not generally pleasant, and we should all in some sense be rooting for the government to succeed at least at its basic functions.
I don’t think there is a set of given-from-on-high proper limits to what the government should do, but I prefer modest aims executed with competence and focus compared to what we have now.
My recommendations don’t generally speak to the overall size or role of government. For the most part, I am trying to help the government succeed by its own lights—often by helping it get out of its own way. I think this approach gets me in with both Democrats and Republicans and makes me more effective than if I founded my ideas in a more explicit ideology.
So well put, the “succeed by its own lights” thing is such an important idea, and probably not articulated enough.
Why do you believe that deregulating housing will increase productivity (TFP)?
In theory, could you build a lobbying super-army against NEPA, housing regulation, etc. a la big oil?
Given that the US now sucks at building public transit, would deregulating housing be a disaster? It’s hard to fit on NYC trains as it is.
Is the price of solar truly going down, or does it just seem that way because the gov is paying for it via incentives and waivers?
If you could have your pick of any federal agency to run for four years, which one would you choose?
1. Housing is, depending on the year, 15-18% of GDP, and if we could get that for free, it would tautologically increase productivity. Also, high housing costs limit agglomeration effects by pricing some people out of the most productive markets. There are a bunch of other negative effects of high housing prices. I’d refer you to “the housing theory of everything” for a discussion.
2. I think the “lobbying super-army” we need is elite consensus. If we convinced all the smart people that vetocracy is a bad way to achieve environmental goals, that would basically do it.
3. If we deregulated housing, people in general would not have to commute as far! But yes, transit construction in the US is often a mess.
4. The price of solar is truly going down. It’s not just because incentives are offsetting the cost. However, I do think it is an open question how far the costs can keep falling.
5. Department level? Department of Energy. Agency level? FAA.
What is the relationship between public policy and the imagination?
That’s a broad question, but as it relates to progressy things, I think imagination about what the future could hold is certainly a factor in the kind of social ambitions that we aspire to.
It’s a common belief among some economic historians, for example, that we have already picked the low-hanging fruit. There are no new inventions in their mind that could match the inventions of the 19th and 20th centuries in terms of providing explosive growth. Maybe they’re right, but I can certainly imagine new inventions that could change everything.
As I argued previously on Progress Forum, futurism is important for producing a concrete vision that can inform our goals.
I loved your idea that Congress should have incentive pay based on growth in John Fernald’s Total Factor Productivity series. I would also argue that it’s important for that incentive to be smooth and linear from 0% to 7%, and that the ideal amount is large—at least $100,000 for each 1% (annually).
My big question: do you think there’s a chance of getting this done?
I still like that idea too, but it’s pretty weird and unlikely to pass any time soon. A more likely reform that I also like is ranked choice voting.
Dear Mr. Dourado,
I’ve recently been watching many clips of the TV Show, The Last of Us. While Mycologists insist that there are no zombie fungi, the loose talk about a new pandemic and the failure to create a universal covid vaccine has taught many people (including me) a deep dread. For example, the declining nutritional value of food could be causing Colon Cancer in young people. Medical science’s slow pace also fails to find treatments for scary diseases like ALS. How do you remain in a positive headspace when there’s so much negative news?
I like this Alan Watts quote:
“Things are as they are. Looking out into the universe at night, we make no comparisons between right and wrong stars, nor between well and badly arranged constellations.”
And Nietzsche’s new year’s resolution is words to live by:
“I want to learn to see more and more as beautiful what is necessary in things; then I shall be one of those who makes things beautiful. Amor fati: let that be my love henceforth! I do not want to wage war against what is ugly. I do not want to accuse. Looking away shall be my only negation. And all and all and on the whole: someday I wish to be only a Yes-sayer.”
I’m amazed that existence exists at all. Every moment is a gift.