The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • EA - OPTIC [Forecasting Comp] — Pilot Postmortem by OPTIC
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: OPTIC [Forecasting Comp] — Pilot Postmortem, published by OPTIC on May 19, 2023 on The Effective Altruism Forum.
    OPTIC is an in-person, intercollegiate forecasting competition where undergraduate forecasters compete to make accurate predictions about the future. Think olympiad/debate tournament/hackathon, but for forecasting — teams compete for thousands of dollars in cash prizes on question topics ranging from geopolitics to celebrity twitter patterns to financial asset prices.
    We ran the pilot event on Saturday, April 22 in Boston and are scaling up to an academic league/olympiad. See our website at opticforecasting.com, and contact us at [email protected] (or by dropping a comment below)!
    What happened at the competition?
    Attendance
    114 competitors from 5 different countries and 13 different US states initially registered interest. A significant proportion indicated that they wouldn’t be able to compete in this iteration (logistical/scheduling concerns), but expressed interest to compete in the next one. 39 competitors RSVP’d “yes,” though a few didn’t end up attending and a couple unregistered competitors did show up. At the competition, the total attendance was 31 competitors in 8 teams of 3-4, with 2 spectators.
    Schedule
    1 hour check-in time/lunch/socialization
    10 min introduction speech
    1 hour speech by Seth Blumberg on the future of forecasting (Seth is a behavioral economist and head of Google’s internal prediction market, speaking in his individual capacity) — you can watch the speech here
    Questions released; 3 hours for forecasting (“forecasting period”)
    10 min conclusion speech, merch distribution
    20 min retrospective feedback form
    Forecasting (teams, platform, scoring, prizes, etc)
    Competitors were split up into teams of 3-4. They submitted one forecast per team on each of 30 questions through a private tournament on Metaculus. Teams’ forecasts were not made visible to other teams until after the forecasting period closed. Questions were a mix of binary and continuous, all with a resolution timeframe of weeks-months; all will have resolved by August 15. At that point, we’ll score the forecasts using log scoring.
    We will have awarded $3000 in cash prizes, to be distributed after the scoring is completed:
    1st place — $1500
    2nd place — $800
    3rd place — $400
    Other prizes — $300
    Note that prizes for 1st-3rd place are given to the team and split between the members of the team.
    Funding
    We received $4000 USD from the ACX Forecasting Mini-Grants on Manifund, and $2000 USD from the Long Term Future Fund.
    Organizers
    Our organizing team comprises:
    Jingyi Wang (Brandeis University EA organizer)
    Saul Munn (Brandeis University EA organizer)
    Tom Shlomi (Harvard University EA organizer)
    Also, Saul and Jingyi will be attending EAG London — please reach out if you want to be involved with OPTIC, have questions/comments/concerns, or just want to chat!
    The following is a postmortem we wrote based on the recording of a verbal postmortem our team held after the event.
    Summary
    Overall, the pilot went really well. We did especially well with setting ourselves up for future iterations, with flexibility/adaptability, and resource use. We could have improved time management and communication, as well as some other minor issues. We’re excited about the future of OPTIC!
    What went well
    Strong pilot/good setup
    As a pilot, the April 22 event definitely has set us up for future iterations of OPTIC. We now have a network of previous team captains and competitors from schools all around the Boston area (and beyond) who have indicated that they’d be excited to compete again. We have people set up at a few schools around the country who are going to start forecasting clubs which will compete as teams in forecasting tournaments. We have undergraduate interest (and associated emails)...
    11 min
  • LW - Job Opening: SWE to help build signature vetting system for AI-related petitions by Ethan Ashkie
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Job Opening: SWE to help build signature vetting system for AI-related petitions, published by Ethan Ashkie on May 20, 2023 on LessWrong.
    Full-Stack Software Engineer Position - AI Policy Petition Platform Development
    Survival and Flourishing .Com (SFC) seeks a competent Full-Stack Software Engineer to join us in creating a novel public opinion aggregation platform centered around AI policy issues.
    Annual compensation: $200,000 - $300,000
    Application here.
    Your Role
    Develop our "V-process" project, a system that ranks signatories of AI-policy petitions based on signature authenticity, credentials, and relevance to the subject matter. This role offers a unique opportunity to contribute to a platform that addresses the need for legible public expression, independent of existing social media platforms and power structures, regarding emerging technologies, particularly AI.
    Our Ideal Candidate
    Strong understanding of system demands and scalability considerations.
    Proficient in full-stack TypeScript, with experience building and maintaining large-scale web applications.
    Comfortable in a remote work environment, with occasional in-person meetings in the Bay Area.
    Excellent communication skills, with the ability to effectively convey progress, implementation trade-offs, and the impact of design decisions.
    Capable of efficiently updating UI based on design requests and contributing to feature decisions.
    Job Responsibilities
    Collaborate with SFC's Product Manager Ethan Ashkie and CEO Andrew Critch, as well as a diverse team of developers and contractors.
    Design and prototype user-friendly UIs for capturing and displaying signatory data.
    Implement business logic for signature validation and integrate with external APIs.
    Occasionally assist with other philanthropic and grant-making projects, time permitting.
    Benefits
    Competitive annual salary of $200,000 - $300,000, plus health benefits.
    Join as the first full-time hire in a flexible, predominantly remote team.
    We are open to hiring remote contractors living outside the US and can collaborate with institutions who may be capable of providing work visas.
    If you're ready to contribute to democratizing the conversation around AI safety and ethics, apply here. We look forward to working together to shape the future of AI policy.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min
  • EA - EAG talks are underrated IMO by Chi
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: EAG talks are underrated IMO, published by Chi on May 20, 2023 on The Effective Altruism Forum.
    Underrated is relative. My position is something like "most people should consider going to >1 EAG talk" and not "most people should spend most of their EAG in talks." This probably most applies to people who are kind of like me. (Been involved for a while, already have a strong network, don't need to do 1-1s for their job.)
    There's a meme that 1-1s are clearly the most valuable part of EAG(x) and that you should not really go to talks. (See e.g. this, this, this, they don't say exactly this but I think push in the direction of the meme.)
    I think EAG talks can be really interesting and are underrated. It's true that most of them are recorded and you could watch them later but I'm guessing most people don't actually do that. It also takes a while for them to be uploaded.
    I still think 1-1s are pretty great, especially if you're
    new and don't know many people yet (or otherwise mostly want to increase the number of people you know),
    have a very specific thing you're trying to get out of EAG and talking to lots of people seems to be the right thing to achieve it.
    I'm mostly writing this post because I think the meme is really strong in some parts of the EA community. I can imagine that some people in the EA community would feel bad for attending talks because it doesn't feel "optimal." If you feel like you need permission, I want to give you permission to go to talks without feeling bad. Another motivation is that I recently attended my first set of EAG talks in years (I was doing lots of 1-1s for my job before) and was really surprised by how great they were. (That said, it was a bit hit or miss.) I previously accidentally assumed that talks and other prepared sessions would give me ~nothing.
    See also the rule of equal and opposite advice (1, 2) although I haven't actually read the posts I linked.
    My best guess is that people in EA are more biased towards taking actions that are part of a collectively "optimal" plan for [generic human with willpower and without any other properties] than taking actions that are good given realistic counterfactuals.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min
  • LW - Twiblings, four-parent babies, and other interesting consequences of advancements in reproductive technology by GeneSmith
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Twiblings, four-parent babies, and other interesting consequences of advancements in reproductive technology, published by GeneSmith on May 20, 2023 on LessWrong.
    Two weeks ago I finally published my year-long research project into how to have polygenically screened children. In that post, I explained all the practicalities of embryo screening and how parents can use it right now to make their children smarter, happier, and less prone to disease.
    This week I'm going to talk about some of the more off-the-wall methods that have been proposed for increasing genetic gain across a variety of traits, along with their implications for family structure, society, and the human experience.
    What are Twiblings?
    A twibling is somewhere between a sibling and a twin. If identical twins share 100% of their DNA and siblings share about 50%, twiblings share 75%.
    To the best of my knowledge, twiblings don’t exist in nature. To get one, you’d need two different genetically identical eggs to be fertilized by different sperm. But why would we create twiblings?
    Anyone who has gone through IVF knows that the main limiting factor in all cycles is the number of eggs the female partner can produce. In an average menstrual cycle, a woman will produce exactly one egg. In an average ejaculation, a man produces about 100 million sperm. Using a specialized set of medications, we can increase the number of eggs harvested per menstrual cycle to between 5 and 100 (there's a lot of variance depending on maternal age and other factors). But even 100 is a lot less than 100 million.
    Older women in particular have a very hard time producing eggs. Women over the age of 42 often produce 5 or fewer per retrieval, and many of those have chromosomal issues that prevent them from turning into a baby. So the few eggs without such issues are very precious.
    It sure would be nice if there was a way to duplicate the few viable eggs these women can produce.
    As it turns out, this is possible! And we’ve already done it in the lab. You can make a haploid stem cell line by tricking an egg into thinking it’s an embryo; you let a sperm fertilize it, then you pull out the sperm’s pronucleus before it fuses with that of the egg.
    Success! The egg now thinks it’s an embryo and will start dividing.
    So now you have a bunch of haploid embryonic stem cells, each of which is fairly epigenetically similar to an egg.But there’s still one tricky step left; how do you derive an egg from these haploid embryonic stem cells such that it can be fertilized again?
    The most obvious answer is to use nuclear transfer; pull the haploid nucleus out of these embryonic stem cells and stick it in another egg. So you do this for a bunch of the haploid embryonic stem cells using a bunch of other eggs and now you have many eggs from the same mother!
    Of course this necessitates HAVING other eggs, which we already established are in short supply. But thankfully, those other eggs don't need to come from the same woman. You can get donor eggs without too much trouble. And if you don’t care much about the donors' DNA you can get them for quite a bit less money.
    But if you can clear that hurdle you now have a very interesting situation: you’ll have multiple genetically identical eggs, each of which can be fertilized by a different sperm.
    If the mother decides to have multiple children, they will be more genetically similar than any siblings, but less similar than identical twins. Twiblings!
    Twins of different ages
    At the 2, 4 or 8 cell stage, the cells in an embryo can turn into any tissue in the body or placenta. This leads to a natural question: if we split the embryo in half at that stage, will it form two embryos?
    It would! Or at least it did in other mammals we’ve studied. For some reason, no one seems to have tried this in humans yet. I don’t fully und...
    9 min
  • EA - Former Israeli Prime Minister Speaks About AI X-Risk by Yonatan Cale
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Former Israeli Prime Minister Speaks About AI X-Risk, published by Yonatan Cale on May 20, 2023 on The Effective Altruism Forum.
    Watch here:
    / (it's in English)
    From the video:
    just like nuclear tech is an amazing invention for humanity but can also risk the destruction of humanity, AI is the same. The world needs to get together NOW, to form the equivalent of the IAEA [...]
    Who is this?
    This is Naftali Bennett (wikipedia), an Israeli politician who was prime minister between June 2021 and June 2022.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    1 min
  • LW - Do Deadlines Make Us Less Creative? by lynettebye
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Do Deadlines Make Us Less Creative?, published by lynettebye on May 19, 2023 on LessWrong.
    This post is crossposted from my blog. If you liked this post, subscribe to Lynette's blog to read more -- I only crosspost about half my content to other platforms.
    Occasionally, my clients struggle to get things done, but worry that setting themselves deadlines will make them less creative.
    Is this a reasonable worry?
    To find out, let’s look at the psychology literature on pressure and creativity.
    There’s a classic psychology experiment called the “candle problem”. Participants are shown matches, a box of thumbtacks, and a candle as in the picture below. The experimenter then instructs the participants to mount the candle on the wall using the available materials. "The problem is considered solved when the candle can be firmly affixed to the wall, burn properly, and does not drip wax on the table or floor."
    If you’ve heard of this problem before, you probably know the answer. However, if this is new to you, take thirty seconds to try solving it before reading on.
    Done?
    If it’s still difficult, just imagine the thumbtacks on the table next to the box as in the picture below. Once you do that, suddenly it’s easy to guess that you should tack the empty box to the wall and put the candle inside it.
    When the box is filled with tacks, our brain writes it off as a “tack holder”, instead of seeing it as a possible “candle holder.” This is called “functional fixedness.” It takes creativity to see an object being used in one way, and then break that default association to work out what other uses it could be put to.
    This experiment is designed to study creativity, particularly the ability to find unusual or “out of the box” solutions to a problem. The experimenter can easily randomize whether the participants see the box full of tacks or empty (i.e. hard or easy creative thinking), plus add whatever other interactions they want to test.
    Which brings us to the point of this whole experiment—how long does it take the participants to find the solution under different conditions?
    In one study, the experimenters used a simple 2x2 design: participants were randomized so that half saw a picture of tacks in a box (which requires more creativity), and the rest saw tacks on the table and an empty box (which makes the puzzle easier). Half were told they’d receive a $20 bonus if they were the fastest in the group ($5 if they were in the top 25%); the other half heard no mention of a bonus.
    Now, adjusted for inflation since 1962, $20 is almost $175 dollars. So, participants had a strong motive to complete the task faster when offered that bonus.
    Did they?
    Only when they saw the empty box. When the picture showed an empty box, participants solved the problem about a minute faster if they were offered a reward (taking on average 3.67 minutes, compared to 4.99 minutes for the non-rewarded group). If instead the participant saw the box full of tacks and was offered a reward, they took over three minutes longer than those who saw a full box but never heard about a reward (11.08 vs 7.41 minutes respectively).
    What’s happening here?
    The basic theory is that when you add pressure, people get better at tasks they already know how to do, but worse at doing novel tasks. This finding has been repeated in studies that use financial rewards, performance evaluation, and even self-evaluation as the source of pressure.
    Think of it as tunnel vision. When you are particularly focused on one problem or motivated to get it done quickly, you get better at doing what you already know exactly how to do. But you get worse at looking around for novel solutions, because you get stuck thinking about the problem in one narrow way.
    Are these findings sound?
    I didn’t find any red flags: googling one of the papers plus the keywor...
    7 min
  • EA - “The Race to the End of Humanity” – Structural Uncertainty Analysis in AI Risk Models by Froolow
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: “The Race to the End of Humanity” – Structural Uncertainty Analysis in AI Risk Models, published by Froolow on May 19, 2023 on The Effective Altruism Forum.
    Summary
    This is an entry into the Open Philanthropy AI Worldview Contest. It investigates the risk of Catastrophe due to an Out-of-Control AI. It makes the case that that model structure is a significant blindspot in AI risk analysis, and hence there is more theoretical work that needs to be done on model structure before this question can be answered with a high degree of confidence.
    The bulk of the essay is a ‘proof by example’ of this claim – I identify a structural assumption which I think would have been challenged in a different field with a more established tradition of structural criticism, and demonstrate that surfacing this assumption reduce the risk of Catastrophe due to Out-of-Control (OOC) AI by around a third. Specifically, in this essay I look at what happens if we are uncertain about the timelines of AI Catastrophe and Alignment, allowing them to occur in any order.
    There is currently only an inconsistent culture of peer reviewing structural assumptions in the AI Risk community, especially in comparison to the culture of critiquing parameter estimates. Since models can only be as accurate as the least accurate of these elements, I conclude that this disproportionate focus on refining parameter estimates places an avoidable upper limit on how accurate estimates of AI Risk can be. However, it also suggests some high value next steps to address the inconsistency, so there is a straightforward blueprint for addressing the issues raised in this essay.
    The analysis underpinning this result is available in this spreadsheet. The results themselves are displayed below. They show that introducing time dependency into the model reduces the risk of OOC AI Catastrophe from 9.8% to 6.7%:
    My general approach is that I found a partially-complete piece of structural criticism on the forums here and then implemented it into a de novo predictive model based on a well-regarded existing model of AI Risk articulated by Carlsmith (2021). If results change dramatically between the two approaches then I will have found a ‘free lunch’ – value that can be added to the frontier of the AI Risk discussion without me actually having to do any intellectual work to push that frontier forward. Since the results above demonstraste quite clearly that the results have changed, I conclude that work on refining parameters has outpaced work on refining structure, and that ideally there would be a rebalancing of effort to prevent such ‘free lunches’ from going unnoticed in the future.
    I perform some sensitivity analysis to show that this effect is plausible given what we know about community beliefs about AI Risk. I conclude that my amended model is probably more suitable than the standard approach taken towards AI Risk analysis, especially when there are specific time-bound elements of the decision problem that need to be investigated (such as a restriction that AI should be invented before 2070). Therefore, I conclude that hunting for other such structural assumptions is likely to be an extremely valuable use of time, since there is probably additional low-hanging fruit in the structural analysis space.
    I offer some conclusions for how to take this work forwards:
    There are multiple weaknesses of my model which could be addressed by someone with better knowledge of the issues in AI Alignment. For example, I assume that Alignment is solved in one discrete step which is probably not a good model of how Aligning AIs will actually play out in practice.
    There are also many other opportunities for analysis in the AI Risk space where more sophisticated structure can likely resolve disagreement. For example, a live discussion in AI Risk at the ...
    37 min
  • LW - GPT as an “Intelligence Forklift.” by boazbarak
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: GPT as an “Intelligence Forklift.”, published by boazbarak on May 19, 2023 on LessWrong.
    [See my post with Edelman on AI takeover and Aaronson on AI scenarios. This is a rough, with various fine print, caveats, and other discussions missing. Cross-posted on Windows on Theory.]
    One challenge for considering the implications of “artificial intelligence,” especially of the “general” variety, is that we don’t have a consensus definition of intelligence. The Oxford Companion to the Mind states that “there seem to be almost as many definitions of intelligence as experts asked to define it.” Indeed, in a recent discussion, Yann LeCun and Yuval Noah Harari offered two different definitions. However, it seems many people agree that:
    Whatever intelligence is, more computational power or cognitive capacity (e.g., a more complex or larger neural network, a species with a larger brain) leads to more of it.
    Whatever intelligence is, the more of it one has, the more one can impact one's environment.
    1 and 2 together can already lead to growing concerns now that we are building artificial systems that every year are more powerful than the last. Yudkowski presents potential progress on intelligence with something like the following chart (taken from Muehlhauser):
    Given that recent progress on AI was achieved by scaling ever larger amounts of computation and data, we might expect a cartoon that looks more like the following:
    (Don’t take this cartoon or numbers too seriously. It is obtained by superimposing a hypothetical 1000T param model on the figure from Bolhuis, Tattersall, Chomsky and Berwick. 100T connections in Homo sapiens brain is a rough estimate. Axes implicitly assume synaptic density scales with volume.)
    Whether the first or the second cartoon is more accurate, the idea of constructing intelligence that surpasses ours to an increasing degree and on a growing number of dimensions is understandably unsettling to many people. (Especially given that none of the other species of the genus Homo in the chart above survived.) This post is not to say that we should not worry about this. Instead I suggest a different metaphor for how we could think of future powerful models.
    Whose intelligence is it?
    In our own species’ evolution, as we have become more intelligent, we have become more able to act as agents that do not follow pre-ordained goals but rather choose our own. So we might imagine that there is some monotone “agency vs. intelligence” curve along the following:
    (Once again, don’t take the cartoon too seriously; whether it is a step function, sigmoid-like, or some other monotone curve can be debatable and also depends on what one's definitions of “agency” and “intelligence” are.)
    But perhaps intelligence does not have to go hand-in-hand with agency. Consider the property of physical strength. Like intelligence, this is a capability that an individual can use to shape their environment. I am (much) weaker than Olga Liashchuk, who can lift a 300kg Yoke and walk 24 meters with it in under 20 seconds. However, if I were to drive a forklift, the combination of me and the forklift would be stronger than her. Thus, if we measure strength in functional terms (what we can do with it) instead of by artificial competitions, it makes sense to consider strength as a property of a system rather than an individual. Strength can be aggregated to combine several systems into a stronger one or split up to use different parts of the capacity for different tasks.
    Is there an “intelligence forklift”? It is hard to imagine a system that is more intelligent than humans but lacks agency. More accurately, up until recently, it would have been hard to imagine such a system. However, with generative pretrained transformers (GPTs), we have systems that have the potential to be just that. Even though r...
    6 min
  • LW - Trust develops gradually via making bids and setting boundaries by Richard Ngo
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Trust develops gradually via making bids and setting boundaries, published by Richard Ngo on May 19, 2023 on LessWrong.
    In my previous post, I talked through the process of identifying the fears underlying internal conflicts. In some cases, just listening to and understanding those scared parts is enough to make them feel better—just as, when venting to friends or partners, we often primarily want to be heard rather than helped. In other cases, though, parts may have more persistent worries—in particular, about being coerced by other parts. The opposite of coercion is trust: letting another agent do as they wish, without trying to control their behavior, because you believe that they’ll take your interests into account. How can we build trust between different parts of ourselves?
    I’ll start by talking about how to cultivate trust between different people, since we already have many intuitions about how that works; and then apply those ideas to the task of cultivating self-trust. Although it's tempting to think of trust in terms of grand gestures and big sacrifices, it typically requires many small interactions over time to build trust in a way that all the different parts of both people are comfortable with. I’ll focus on two types of interactions: making bids and setting boundaries.
    By “making bids” I mean doing something which invites a response from the other person, where a positive response would bring you closer together. Sometimes bids are explicit, like asking somebody out on a date. But far more often they’re implicit—perhaps greeting someone more warmly than usual, or dropping a hint that your birthday is coming up. One reason that people make their bids subtle and ambiguous is because they’re scared of the bid being rejected, and subtle bids can be rejected gently by pretending to not notice them. Another is that many outcomes (e.g. being given a birthday present) feel more meaningful when you haven't asked for them directly. Some more examples of bids that are optimized for ambiguity:
    Teenagers on a first date, with one subtly pressing their arm against the other person’s, trying to gauge if they press back.
    Telling your parents about your latest achievements, in the hope that they’ll express pride.
    Making snarky comments about your spouse being messy, in the hope that they’ll start being more proactive in taking care of your preferences.
    Asking to “grab a coffee” with someone, but trying to leave ambiguous whether you’re thinking of it as a date.
    Of course, the downside of making ambiguous bids is that the other person often doesn't notice that you're making a bid for connection—or, worse, interprets the bid itself as a rejection. As in the example above, a complaint about messiness is a kind of bid for care, but one which often creates anger rather than connection. So overcoming the fear of expressing bids directly is a crucial skill. Even when making a bid explicit renders the response less meaningful (like directly asking your parents whether they're proud of you), you can often get the best of both worlds by telling them explicitly about the emotion underneath the bid (e.g. that you often feel judged by them), rather than the bid itself .
    There's a third major reason we make ambiguous bids, though. The more direct our bids, the more pressure recipients feel to accept them—and it’s scary to think that they might accept but resent us for asking. The best way to avoid that is to be genuinely unafraid of the bid being turned down, in a way that the recipient can read from your voice and demeanor. Of course, you can’t just decide to not be scared—but the more explicit bids you make, the easier it is to learn that rejection isn’t the end of the world. In the meantime, you can give the other person alternative options when making the bid, or tell them expli...
    7 min
  • LW - Idea: medical hypotheses app for mysterious chronic illnesses by riceissa
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Idea: medical hypotheses app for mysterious chronic illnesses, published by riceissa on May 19, 2023 on LessWrong.
    (This post is not up to my usual standards but I was encouraged to publish it anyway to get feedback on the idea.)
    I have been ill for the past four years with a mysterious chronic illness. One of the things that I keep thinking would be nice to have (but as far as I know, doesn't exist) is some sort of symptom-tracking app that would allow me to test various hypotheses for my various symptoms.
    I already do a simple version of this just in my head, for example, I can notice that if I stand for too long at a time it gets kind of aversive and sitting down makes me feel better. But it's hard to do this in my head for anything more complicated. What if eating cheese subtly helps with my breathing problem? It is not so easy to figure this out on my own.
    The things that make it hard to detect connections between "things I do" and "how I feel":
    delayed effects (maybe eating cheese helps with my breathing, but only on the following day, or maybe the effect builds up over time)
    conjunctions/combinations of things (maybe I need to both eat cheese and exercise on the same day and avoid other dairy, in order to feel better in some way)
    small effects, which are still good to notice because sometimes you just need to do more of it
    it's also hard to come up with hypotheses yourself
    Also, if the thing is something you've already tried in the past, it's good to make use of the data you already do have instead of running every single experiment after you generate the hypothesis.
    As you live through chronic illness, you keep generating new data every day. You also, occasionally, come across new hypotheses. It would be nice if you can test each new hypothesis against all the data you already have.
    The closest thing I know is the app Bearable, but the only "hypothesis" it seems to have is in the form of correlations of things that happen in a single day (e.g. "there is a such and such correlation between sleep duration and mood"), and the interface was so bad I couldn't keep using it.
    Currently I just track my diet, medications, and symptoms in a bulleted list in Roam. This makes it possible to check for more hypotheses than by my intuition alone, but it's too difficult to search for all the relevant days and track effects. For instance if the hypothesis is "if I eat oatmeal then seven days later I will have breathing problems", I will have to find every single instance of when I ate oatmeal, and then by hand check how I was doing seven days later. So the symptoms and the interventions being machine-specifiable is important.
    I suspect that this sort of thing may be useful for healthy people and in domains other than health, though I think the barrier is that most people who are not seriously ill do not have the motivation to collect so much data and track how they are feeling each day.
    Won't this "fit" a lot of spurious hypotheses, especially if the hypotheses are auto-generated? (Or in other words, won't p-hacking/"correlation but not causation" type things will be a problem?) I think it might, especially if the number of times an intervention was tried is too few. But I don't think this will be a big problem for this app because as long as there are only a smallish number of viable hypotheses, I can just run the experiment later to test them. ("The app says I felt way better 3 days after eating watermelons, and this happened on two occasions? Well then, I guess I'll start eating watermelons for a while.") It may help to iteratively make the hypotheses more complicated, until one reaches the point of having just the right number of plausible hypotheses to play around with.
    Some hypotheses are expensive to test so I won't have a way to check it with the existing data, e.g. r...
    5 min