Transcript 0:00 Hey folks, it's Quinn. Since we recorded this conversation, we've renamed the podcast. It used to be called Important Not Important, which is what you'll hear in the episode and is still the name of the company. 0:12 But now the podcast is called The Most Important Question. That is, what can I do? 0:18 It's the question we get the most from all of you, and it turns out it's the question I've really been asking guests for almost two hundred episodes now. So here we are, the most important question. 0:29 Please enjoy this old conversation. Thanks for listening, and thanks for giving a shit. [intro music] This week, what happens to us when AI comes? And later, the news. 0:47 The Willow Project, a failure. Blueberries, honeybee vaccines, and climate disclosure rules. But first, I'm Quinn Emmett, and this is Important Not Important, science for people like you who give a shit. 1:02 The newsletter features the most important science news, but most importantly, how to think about it and what the hell you can do about it. 1:09 Hit subscribe right now to get this newsletter and my conversations with the world's smartest people every single week. 1:17 You can find the email version and links to everything at importantnotimportant.com/newsletter or in your show notes. It's March seventeenth, two thousand and twenty. Here's your weekly action steps. 1:29 Here's what we can do. Just a few bucks, buy some life-saving bed nets with Against Malaria, maybe the most effective NGO on the entire planet. Second, the only thing dumber than cancer is rare cancers. 1:43 Good news, you can help fund research against them and work up a sweat at the same time with our friends at Cycle for Survival. 1:50 Number three, super easy, get educated and follow the Black Maternal Health Caucus on Twitter. Number four, understand your home's exposure to flooding, fires, heat, and wind with Risk Factor. 2:03 And now, today's big question. Welcome to the unknown unknowns. Take a look at your calendar. Note the date. Today is the day the world changed forever. 2:17 Now let's take a step back into the human past for a moment to nineteen ninety-two. Even if you've never seen Jurassic Park, you know the quote, the one by the soon-to-be gloriously shirtless Dr. Ian Malcolm. 2:29 Your scientists were so preoccupied with whether or not they could, they didn't stop to think if they should. That's by David Koepp, who's the screenwriter of Jurassic Park, the movie. Dr. 2:40 Ian Malcolm's quote has stayed relevant because it is applied to many of the technological advancements we've made since nineteen ninety-two. And it applies very much to what's happening in artificial intelligence. 2:52 But as I'm going to explain today, not in the way that you think. Now, to be clear, I'm not here to bash innovation, not by a long shot. Look at what we do here. 3:02 The world is better in almost every measurable way since nineteen ninety-two, not just because of environmental protections, lawsuits, anti-smoking campaigns, the UN Sustainable Development Goals, but also because of advancements in genome science and big data, medical devices, targeted cancer therapies, pharmacology, ART for HIV, heart disease, and so much other stuff. 3:27 And to answer Dr. Malcolm, we absolutely should have done those because we could. 3:33 We were finally, marvelously technologically able, and millions of lives could have been, and were subsequently, improved by us doing those things. 3:44 Today, we can do many, many fruitful things, and we should do those as well because we understand them, because they are necessary. 3:53 For example, we have virtually every technology we need to build a world powered by renewable energy sources. 4:01 And with climate change here, of course, we should build these things to bring meaningful relief to billions of people, animals, and ecosystems. 4:11 The only can holding us back now is the political will to overcome trillions in fossil fuel subsidies and industry lobbying to build what the hell we need to build. 4:22 Once we overcome those, we can, should, and we will spend the next few decades building this abundant, incredible world, relieving the devastating burdens we've put upon the planet's ecosystems and most marginalized people, and honestly accounting and paying for the new trade-offs along the way. 4:40 And we can do those things now and simultaneously because we have the information we need to understand them and a little time to most ethically execute them. Those are what we call the known knowns. 4:55 With artificial intelligence, or AI, can is no longer a question of technical ability, really, but a much more urgent question of how much change we can possibly absorb as a people. 5:06 It requires us to shift Malcolm's question from should we do X to can we do X, all things considered. 5:13 This new AI era supersedes everything before it, truly, where the abundance from its utter ubiquity will rapidly compound into known unknowns, which we have some experience with and we'll get into, but we don't handle very well, and soon unknown unknowns. 5:32 And that's where fundamental assumptions of how society works evaporate and future shock becomes the status quo. So this is why time matters, to give it to some context. 5:44 One of the driving mechanisms of both the pandemic and the climate crisis has been a refusal to calculate, much less pay for, the costs of a couple hundred years of progress, of building what we want to build, where we want to build it, and of whatever materials we please.So after decades of lies and lobbying and of companies, governments, and rich people finally doing the math on their daily emissions, and they're-- now they're throwing millions of dollars at carbon offsets that aren't real in a dishonest attempt to forestall the future. 6:17 Now, side note, it shouldn't surprise you that I don't support carbon offsets. In, in almost every version we've seen so far, they aren't real. 6:25 And worse, their existence sets decarbonization back even further because it provides those same companies, governments, and the wealthy with a get out of jail free card for continued emissions. 6:36 So anyways, this isn't that essay, but I do support the continued testing and scaling of carbon removal, if only it's additional and if only executed in concert with all available efforts to get to real zero new emissions. 6:51 Back to the story. The point is, the industrial era built the world we take for granted every minute of every day. 6:58 It lifted billions of people out of poverty, but with perspective, with time, we understand just how much it cost us, how much damage we've done along the way. 7:10 We understand now what carbon does to the atmosphere, and that we can remove carbon, and that our ocean and trees have been doing it for us all along. 7:18 But they can't do it much longer, and there's a lot of species we can't bring back, and we can't put sea level rise back in the box as far as we can tell. 7:29 It's gonna march on for the rest of our lives and for our descendants too. So progress very clearly comes with trade-offs, but it can take time to understand what they are and the first and second effects of those. 7:43 We've built some mechanisms to speed these up. For instance, in health, it's massive randomized clinical trials, for example. But we can't see how long it takes for something to affect the body over time. 7:54 As humans, we can only know so much and project so far into the future, much less to travel to it. You know, there's a reason we pull ice cores from the Antarctic and Greenland. 8:05 They give us a better understanding of what happened the last time things were like this. But with AI, there's no ice cores to pull. There's no precedent in the geological record. 8:19 The only factor that's relevant about the past here is an understanding of how we as humans have made decisions in the past, given enormous change, given time to adjust. 8:31 But with AI, we have no time to adjust, to assess what is necessary to who we are and how the pieces of our society and economy fit together, what's worth preserving and what is not, and what our descendants lose when we take options away. 8:48 Now, look, to be clear, it's not like AI overlords are gonna tell people they can't write fiction anymore. In the best case scenario, we'll have even more time to write fiction. The question is how we tell it apart. 9:00 The question is who's gonna pay us for it? With AI, we can do already wonderful, imaginative, and soon impossible to imagine things. 9:13 But the biodiversity of human contributions to society and economies as we know it is at risk. We don't really know what will happen when those go away. 9:25 In the book Thinking in Systems, uh, Donella Williams and, and the other con-contributors described what happens when we willingly sacrifice biodiversity as we tradi-traditionally define it. 9:36 They said, "When you understand the power of system self-organization, you begin to understand why biologists worship biodiversity even more than economists worship technology. 9:48 The wildly varied stock of DNA, evolved and accumulated over billions of years, is the source of evolutionary potential, just as science libraries and labs and universities where scientists are trained are the source of technological potential. 10:06 Allowing species to go extinct is a systems crime, just as randomly eliminating all copies of particular science journals or particular kinds of scientists would be." 10:17 So to say that we don't all agree on this framing yet would be an understatement, right? We still... We've normalized industrial meat to the tune of one soccer field of rainforest lost a minute, every minute. 10:30 We've normalized air pollution that kills eight million people a year, every year, because those things are convenient. Those are the trade-offs. 10:40 Now, the climate clock has ticks remaining, but it's ticking faster now, and that's an important thing to understand. 10:47 But barring an asteroid or supervolcano explosion or both, climates usually change over millennia or longer. Again, we've sped ours up and in the wrong direction, 10:58 but we still have some time on the scale, some room for error to do as much as we can. And we've begun to course correct. 11:05 And the metaphor I stumbled on for this one was like Captain Jack Aubrey in the Southern Ocean, right? 11:11 Being chased by some massive Dutch ship of the line, freezing cold water rising in the hold, her masts splitting down the middle. 11:19 Who knows how many midshipmen are already flung over the side or just exploded by cannonballs or grapeshot. 11:25 But somehow we'll push on with the belief that we'll make it out of this, that clear skies and calm waters are just around the corner. 11:34 So with cl- with climate, we've barely enough time to turn the proverbial ship around, knowing, of course, that millions have already suffered and many more will suffer during the transition. 11:47 There are plentiful known unknowns when it comes to the climate crisis. We know about the heat and the drought and the flooding and the storms and, of course, what we can build with unlimited renewable energy. 12:00 Known unknowns start to go into what the contributions might be from eight million people a year who would have otherwise died from air pollution.We don't know, but we're sure as hell gonna find out. 12:13 And on the impact side, these are incredibly complex systems we fucked with, and there are real tipping points we know about with inevitable outcomes, but we don't understand them yet or how they play together or any of that. 12:25 So we've triangulated the information we do have, and many of us, you folks included, are operating at maximum warp to build a radically better future and to atone for the past, to build and multi-solve with shit like solar panels over dwindling reservoirs, to shut off the gas, literally, to map the ocean floor, to protect it and the waters and the creatures above it. 12:48 Great news. We can actually use AI, we already are, to move faster on those. But what will be the costs to access the power we want? 12:59 I got into sci-fi writing because I wanted to help imagine what's just beyond our reach, not too close, not too far, and to question how we get there and how we deal with it. 13:11 Years later, I find myself here trying to help tens of thousands of readers more effectively put a dent in the universe. Used ethically, AI can help us put one hell of a dent in the universe. 13:25 The future positive known unknowns of AI are abundant. New medicines at very little cost, but for which diseases? New ways of learning languages, but what might actually be the most effective way to do so? 13:41 Is that gonna be overturned, our idea of how to do that? New ways for less educated workers to compete and contribute alongside more educated ones. But where? Is it gonna require Google Docs or Microsoft Office or 13:56 something else? More productivity among all of us and more free time for some. More time to do what? To find meaning? 14:06 Is it, as Victor Frankl wrote, "a new opportunity to transcend subjective pleasures by doing something that points and is directed to something or someone other than oneself, 14:18 by giving himself to a cause to serve and another person to love?" Are we gonna have time for that? [chuckles] As it stands, the current pace of AI doesn't give us much time, time to react, much less to plan. 14:33 Like the climate crisis on fast-forward, AI is only going to compound on itself until the clock is ticking so fast 14:40 that time doesn't really mean what it used to to us, and that's that disconcerting feeling that climate gives us, but much faster. 14:48 Recognizing we cannot slow AI's progress now, it is essential we ask of ourselves and our money and our tools and our time, what's it all for? 14:59 In one of my favorite books, in one of my favorite series, Wizard of Earthsea, uh, the protagonist of the series, Ged, is one of a few special wizards, a teen who feels his potential and powers are criminally unappreciated. 15:14 So one day at school, he lashes out at a rival, showing off in front of his peers and his mentors, never stopping, though, to question might come of it. W-w-we all did that in high school, right? 15:26 Predictably, it goes poorly. Very poorly. 15:30 So Ged actually spends the rest of his series tempering his mighty powers and atoning for what he wrought literally into the world 'cause he increasingly understands the wide-ranging implications of that one decision and because he has the time to make recompense. 15:47 We're not going to temper anything with AI. I mean, with notable muddied exceptions for nuclear weapons and cloning and germline editing, we don't really temper progress. 15:58 Even with those, we're mostly dealing with known unknowns. Again, context for how fast we keep speeding up. In The Three-Body Problem, Ken Liu translated Cixin Liu's brief summary of human progress into English. 16:13 "Humans took more than a hundred thousand Earth years to progress from hunter-gatherer age to the agricultural age. To get from the agricultural age to the industrial age took a few thousand Earth years. 16:25 But to go from the industrial age to the atomic age took only two hundred Earth years. Thereafter, in only a few Earth decades, they entered the information age. This civilization possesses the terrifying ability 16:38 to accelerate their progress." [upbeat music] Hey, it's Quinn. One of my favorite things about this job is to not actually do this job. Stick with me. Sometimes you get to sign off. You know what I mean? 16:57 When I'm forcing myself to not think about climate change or malaria, I'm usually cooking, eating, downloading apps, listening to music, or reading dragon smut. 17:07 And every month, I share my favorite updates on these in a members-only newsletter called Not Important. 17:13 If you wanna receive it, head to importantnotimportant.com/upgrade or just click the link in your show notes to start your free thirty-day membership trial. 17:22 [upbeat music] So, right. I don't say this lightly. 17:35 Today's AI co-pilots might be obvious, but we've really started to manifest a future of unknown unknowns. So describing the scope of real AI as anything but everything, everyone, everywhere, all at once 17:54 would be a disservice, and we're just not prepared for that transition. It's one thing to adapt to a sea that is rising slowly but surely over decades and centuries. 18:08 It's another, as we know, to adapt to a novel coronavirus for which we have no natural immunity, or AI tools that have literally just this weekUnlocked vast educational productivity improvements, but which could quickly overturn our understanding of education and productivity, of employment, of inequality, of biological research, and a million other building blocks of society that we can't possibly foresee. 18:37 AI, or honestly, really just fancy machine learning, has been a part of your life for a decade now, from social media to online advertising to Siri to mortgages and policing. But compared to, again, just this week, 18:54 those tools were primitive at best, with results that have been decidedly mixed. Known knowns are most obvious instincts and biases at work, more connected and made faster. That's all they were. 19:09 And I can't believe I'm saying this, but as relatively limited as these new tools are compared to what we've always imagined artificial general intelligence or AGI would be, 19:19 we're not relatively that far off from a version of it. There's a very long way to go, but time-space doesn't mean the same thing to AI as it does to us. Think of the AIs that learned to play chess and go. 19:34 They taught themselves, and then they became better by playing millions of games against itself. We can't do that, right? So it doesn't mean the same thing to AI as it does us, time. 19:45 And so these large language models that can inhabit different personalities on demand, all while somewhat accurately now posing as a law student, radiologist, or a musical historian, a microeconomist, an action movie screenwriter, 20:00 that's a multimodal paradigm shift we're not quite ready for. And that's just the start. 20:05 As of today, we've entered a world and empowered a technology that we simply do not understand, much less are able to control or rein in. Look around at the news. 20:14 We barely know how to handle a late pandemic, early climate crisis economy with known measurable inputs and outputs. Many things about AI are out of our control. 20:27 But knowing what we can control and operating with purpose can upvote fantastical opportunities and alleviate some of the inevitable and unimaginable losses. 20:41 That's why we have to ask all the hard questions right now. 20:45 I firmly believe we can celebrate a new era like this while simultaneously questioning not only the ethics of who makes the underlying technology, what, and for something like face scanning, who it's made from, who profits from it, 21:01 and who will suffer from it. And that's already this week getting more difficult to answer. So this is otherwise intended to be a more timeless piece, if that's possible in the AI era. 21:14 But it's important to understand for a moment how some of the primary players, for example, OpenAI, how they have evolved from a well-funded open research nonprofit to in part a closed for-profit quickly, aggressively. 21:30 Google, one of the other major players, has always been for-profit. So OpenAI's pivot and feedback loop aren't difficult to understand or even a new idea, right? 21:39 They said becoming a for-profit entity enables them to compete for talent, sure, so they can use access to increased funding for more research with that talent, yep. 21:47 The subsequent intellectual property from all of that becomes a further profit mechanism, yep, enabling them to hire even more talent, so on and so on. Yeah, of course, that's what for-profits do. 21:59 But they've also refused now to share any more research and said sharing in the past was a mistake because doing so alongside their effort to bring about AGI would give bad actors too many pieces to put together on their own, right? 22:13 That is, it's not because it would torpedo their partner and sugar daddy Microsoft's new business model the way Microsoft torpedoed their ethics and safety team just this week. Those two things happened at the same time. 22:27 These moves require more pointed questions from us. Is backtracking their way to win the arms race versus more traditional for-profit companies like Google and Meta and others? Is it for safety? 22:39 Is it to preemptively eliminate opportunities for regulation and enforcement with the lawsuits that already started? Without more context, I think we can safely assume all of these are true. 22:51 But AI doesn't operate in isolation. The exact opposite. It's everywhere. So without answers to those questions, asking broader questions becomes more difficult too. 23:03 Climate change is the other thing that touches everything. What are the raw mineral and climate impacts of NVIDIA's chips that are required for these? 23:10 What are the power requirements for a day of use even now at the beginning? How much should our precious water cost to cool the data centers we'll somehow become even more reliant on? Who should regulate all this stuff? 23:26 States? Countries? UN? All of them? No one? The self-regulating market? How are they going to self-regulate for ethics and safety without an ethics and safety team? 23:40 Google set the pace by laying off their ethical AI team years ago. So we can only assume, because this is what they tell us, that this is the way forward. And to paraphrase the Mandalorian himself, this is not the way. 23:54 And certainly not when we're dealing with a future of unknown unknowns. Maybe governments will step up. 24:01 Before legislation and regulation comes understanding, not simply how something works, but what its potential may be and who it could affect to protect the vulnerable and still leave room for innovation and to maximize 24:16 the universal good it can do.There's examples of that. Regulation is not perfect, but that's done well what it does. A compromised octogenarian Congress isn't the answer. That's the obvious dig, though. 24:30 And if it isn't clear, I don't think anyone has the answer. And willful ignorance definitely isn't the fucking answer, which is why it's so vital we ask better questions all the time. Big questions, hard questions. 24:43 Now, one analogous climate era example I've talked about would be, and it's uncomfortable, how can tens of millions of people continue to live in the American West knowing it's well into desertification, even with all these rains? 24:58 A more future positive AI question would be, 25:01 if the cost for pharmaceutical companies to research new medicines drops ninety percent, how can we cap consumer costs for those new medicines or devices or whatever to provide for universal access to them, especially if AI is gonna make suddenly so many jobs expendable? 25:20 These are the questions we have to ask. Here's what we do know. Here's where to start. These are our known knowns. First step, 25:28 training these foundational models requires very specific chips and enormous amounts of power, both of which are super tenuous geopolitical questions right now. We have to think about this actively. 25:42 Derivative versions of the models from the API or public research require far fewer chips and far less power. 25:49 They can be trained more specifically and literally run right on your phone already 'cause the broader work is already done. But for those who seek to profit from them, the work is never gonna be done. 26:01 These tools will struggle at times to live up to all the hype, including so much of what I've posited here. It's gonna take time, less time and more time. 26:09 But they will eventually technically make jobs and entire industries like graphic design, screenwriting, editing, nonfiction writing, accounting, architecture, software development, data science, market research, legal, customer service, a-and so many others expendable. 26:27 Don't delude yourself about that. These co-pilots of today will become the pilots of tomorrow 'cause it's more cost efficient. There's no going back. 26:38 These tools, like the workers they're replacing, you and I, are very imperfect, often inaccurate, and biased. We'd like to believe they know more than they do, when in reality they are incapable, so far, 26:52 of making decisions on their own. But they will inevitably grow and change. As they grow, they'll surprise us, so we expect more from them than they're capable of. That's what we do. That's how we grow things here. 27:06 But this is also what we do. We are a species that finds enormous meaning in work, in creativity, in expression. We are most happy when we are connected live and in person, and when we have a purpose to work towards, 27:21 even if a rare few of us get to actually choose our work and that purpose. And on the other hand, even if many of us could stand to work a little bit less. So progress is never gonna intentionally slow. 27:34 But we can manage the transition by knowing ourselves as best we can. Think about it again. 27:40 We can electrify cars, but what about the tens of thousands of people who don't just service combustion engines, and what are they gonna do about their jobs? 27:47 But what about the fact that a lot of them just love working on machines? How do we support that? We can automate checkout, customer service, food prep, or bartenders, or whatever. 27:56 What is the cost of less human interaction? People just wanna open a restaurant. How do we accommodate all of those versions? 28:06 I wanna say clearly, if there are tools that let us spend more meaningful time with young people and the elderly, to rebuild our relationships with nature, to make it easier to converse with one another in whatever language, 28:19 to personalize learning, to increase crop yields, to distribute our clean energy more efficiently, to increase access to financial services and essential infrastructure services, 28:31 to provide for a more robust and equitable safety net, to predict natural disasters and speed recovery, to make wellness more universal with an increased emphasis on preventative health, we should build those. 28:44 We should use those. And those tools are here or coming, and that is wonderful. 28:49 But we have to try to understand the known trade-offs as best we can and steel ourselves for the rest, considering our most basic needs and who we are and how we operate and make decisions. 29:00 So now is the time to ask big questions about, yeah, social safety nets and reinvigorating hands-on work industries and improving labor standards and economic diversification and trade schools, retraining, and more to support one another, to make, as the popular saying goes, for a soft landing. 29:20 Look, after all this, time might not actually be real. It's a long story. We'll get into it later. But until we make some serious advances in theoretical physics, the past has already happened. 29:31 Tomorrow's always right around the corner, except now faster. There's no going back. 29:37 So we have no choice but to go into tomorrow with our eyes wide open to make sure we don't automate what gives us life, to take away the livelihoods of those that rely on the act of creation to find meaning for themselves, 29:53 and whose creations often provide it for the rest of us. And now, the news. In climate change news, Joe Biden approved the Willow project in Alaska, and it sucks. 30:06 But on the other hand, the US is on track for a major clean energy milestone. I can't explain enough if we just improve transmission legislation how faster this... much faster this would go. 30:17 Anyways, the world's first honeybee vaccine is like magic. You can use a new Washington Post tool in your show notes. 30:24 See how early spring came to your neighborhood.In the future, it still might cost one trillion dollars per PPM of CO2 removed, and we should pay for it. The SEC's climate disclosure rules are coming. 30:37 Eighty-five percent of business executives say they're not ready. Why not? And why did insurers slash Hurricane Ian payouts? More to come on that. 30:46 In COVID news, COVID made maternal health outcomes much worse, and especially among Black people. In great news, there doesn't seem to be an association between Paxlovid and a COVID rebound. Again, that's great. 31:01 In food and water news, the EPA proposed their strongest ever standards for keeping coal plant pollution out of your water right after they went after forever chemicals. What a great week for water and the EPA. 31:13 Apparently, blueberries are just packed with pesticides, which sucks. A Supreme Court case could reshape indigenous water rights in the parched Southwest. 31:25 In health and bio news, what's secondhand stress, and why do I feel it all the time? Next gen bed nets are coming and could save so many lives. 31:33 And Medicaid expansion might finally be coming to hundreds of thousands of people in North Carolina. And in computer news, there's no computer news. I'm kidding. I just talked about it for twenty minutes. 31:44 But in addition to all the AI news and mid mass tech layoffs, workers on H-1B visas have sixty days to find a job or leave the country, and that can't be the right way to do it. That's it for this week. 31:57 Hit subscribe to get next week's issue straight to your feed wherever you listen to podcasts or wherever you're watching this, I guess, just on YouTube. Anyways, to go deeper, visit importantnotimportant.com. 32:07 Thanks for being a part of our community, and thanks for giving a shit. Have a great weekend. 32:14 [outro jingle] [upbeat music] That's it for this week's conversation. 32:26 For more conversations, scroll back in the feed or visit podcast.importantnotimportant.com to search by name, topic, whatever. Thanks for sharing. Thanks for leaving a review, and thanks for giving a shit. 32:42 [outro jingle]