Last January, we, Dylan and Kelsey, made 16 specific predictions about what the year 2019 would bring. Our forecasts covered everything from US politics to the Chinese economy to progress in artificial intelligence.
We put estimated probabilities on each outcome, as a signal of how confident we were in each forecast and so you could hold us to account accordingly if we got them wrong.
But we want to hold ourselves to account, too. In this post, we’re looking back and confirming which predictions we got right, which we got wrong, and what we learned from the experience.
Our predictions (which we’re doing again for 2020 — stay tuned) weren’t just an idle exercise. We think it’s really important to get better at predicting the future, and the best way to get better, as with any skill, is to practice.
It would be nice to be able to predict the future. It would’ve been helpful if the Bush administration had foreseen a lengthy, bloody quagmire after invading Iraq; if regulators and bankers foresaw that over-leveraged investments in mortgage securities would lead to a recession; if governments saw the 1919 Spanish flu coming while there was still time to prepare.
This isn’t just an idle thought; we can get better at forecasting! That’s a major finding of Philip Tetlock, the University of Pennsylvania psychologist who holds forecasting tournaments to learn how humans can better predict the future.
As Zack Beauchamp explains in a Vox piece on Tetlock’s work, Tetlock and his collaborators have run studies involving tens of thousands of participants and have discovered that prediction follows a power law distribution. That is, most people are pretty bad at it, but a few (Tetlock, in a Gladwellian twist, calls them “superforecasters”) appear to be systematically better than most at predicting world events.
Tetlock even found that superforecasters — smart, well-informed, but basically normal people with no special information — outperformed CIA analysts by about 30 percent in forecasting world events.
So what do they do differently? Tetlock and others are still trying to figure that out but have a few suggestive conclusions so far. Superforecasters change their minds a lot. They’re open to being wrong. They’re comfortable giving numerical probabilities, which makes their assumptions more explicit and easier to test.
But most of all, they practice. They make lots of predictions, see where they went wrong, learn from their missed calls, and do better going forward. And a number of other writers — like Scott Alexander, Rodney Brooks, the Financial Times staff, and Zachary Jacobi — have been practicing for the past few years too, making predictions and then looking back to see how they did, as a kind of annual tradition we’ve decided to emulate. If you want to try your hand, the site Metaculus is a good place, and the successor company to Tetlock’s Good Judgment Project also runs competitions.
So, without further ado, here’s what Kelsey and Dylan thought would happen in 2019 — and what actually happened.
The United States
Donald Trump will still be in office (90 percent) — CORRECT
This was not the bravest prediction on our list, but it panned out. As I wrote at the time, “He could resign, but I highly doubt even the most damning report from special counsel Robert Mueller could lead him to something so ego-shattering.” The Mueller report came and, sure enough, Trump and his Attorney General Bill Barr treated it as an exoneration, despite its many damning elements. The House impeached Trump, but there’s no indication that the Republican Senate will convict, or that even a single GOP senator will vote for conviction. And, in any case, that didn’t happen in 2019. Trump is still president, and it appears unlikely anything but an election defeat in November will remove him. —DM
No Democratic presidential candidate will become a clear frontrunner in the political prediction markets at any point in 2019 (60 percent) — WRONG
This turned out to be correct in spirit but wrong in terms of the criterion I used: Will any 2020 candidate exceed a 50 percent probability of winning according to PredictWise, a website that collates odds from bookies and prediction markets to estimate the likelihood that bettors — people putting down actual money predicting the 2020 primaries — assign to various candidates?
If you scroll down here and check out Elizabeth Warren’s trajectory in betting on who the 2020 nominee will be, you’ll see that for much of early October, Warren’s estimated chances exceeded 50 percent. They never got much higher than 50 percent, but for that month the betting markets thought she was favored against the whole field. By Halloween, Warren’s slide had begun, and now she’s nearly tied with Pete Buttigieg, with an estimated 12 percent chance. Biden, as of this writing, is at 33 percent, and Bernie Sanders is at 23 percent.
In short, the primary still looks like anyone’s game — which I thought it would be. But in October, Warren rose so far above the pack that my prediction turned out to be wrong. —DM
The US will not enter a recession (80 percent) — CORRECT
Last January, we were beginning to see some dire predictions about the state of the economy. Stock markets had a troubled end to 2018. The Washington Post warned us “A Recession Is Coming — And Trump Will Bungle the Recovery.” Business news reported that half of CFOs expected a recession in 2019. The interest rate on five-year bonds fell lower than the interest rate on three-year bonds, a so-called “yield curve inversion” that often precedes recessions. (A stronger predictor, 10-year bond rates falling lower than two-year bond rates, didn’t happen until August.) The government was mired in a shutdown, which can damage the economy by restricting income and spending for lots of Americans who directly or indirectly work for the government.
But despite all that, I predicted only a 20 percent chance of a recession — and I was right. Recessions are pretty rare while stock market volatility, government shutdowns, and pessimistic surveys from business leaders are a fair bit more common. So even though all of the above events were evidence of trouble, none of these things were actually very strong evidence — they shouldn’t have shifted us very much from our baseline expectation that the economy would grow in any given year. And, indeed, that’s what it did. —KP
Congress will not authorize funding for a full-length border wall (95 percent) — CORRECT
I phrased this prediction in a way that made it very unlikely to be wrong.
Let’s break it down: The US-Mexico border is 1,954 miles long. Prior to Trump taking office, about 654 miles of that was “fenced,” 300 miles of which feature only “anti-vehicle” fences that often aren’t able to block pedestrians. So Trump would need enough funding for another 1,300 miles of walls to get his complete Gulf-to-Pacific barrier built.
As of this past September, no extensions to those barriers had yet been built. Around that time, the Trump administration told reporters it was hoping — using emergency funds from the Pentagon and additional appropriations from Congress if possible — to add almost 500 miles by Election Day 2020. That would be a huge step toward their goal, but it’s less than half the remaining length that Trump still needs to build to fully cover the border.
What’s more, federal court rulings in December have stalled the administration’s attempts to use Pentagon funding, and an omnibus spending bill passed that month contained only $1.4 billion in border wall funding, far too little to meet the 500 miles goal.
Trump, in fairness, has backtracked from his initial promise of a wall, saying instead that “natural barriers” like mountains or the Rio Grande can substitute for a wall in some places. But he hardly has enough congressional money for even a partial wall covering all areas without those natural barriers. —DM
US homicides will decline (80 percent) — CORRECT
With mass shootings frequently in the news, it’s easy to forget that, overall, violent crime in the United States has been declining significantly since the 1990s. Our data here is on a bit of a time lag, as it takes a while for national homicide data to be compiled, so we’re comparing 2018 to 2017 (we won’t have 2019 data until around mid-2020).
In 2018, the FBI estimates there were 16,214 murders in the US (murder is a good proxy for violent crime in general because the definition has been stable for a long time, murder reporting rates are much higher than reporting rates from other violent crimes, and murder rates track rates of other violent crimes). That is 16,214 too many, but in 1991 (when it peaked) there were 24,703, and the rate has been declining — mostly steadily — since then. Most importantly for us, it’s a 6.2 decrease since 2017. —KP
The United Kingdom will leave the European Union (80 percent) — WRONG
This one’s pretty simple: I thought the UK would formally leave the EU. It didn’t. What happened? Well, Theresa May wound up requesting not one but two extensions from the original March 29 deadline to leave, after her Tories failed to back her proposed deals for a “soft” Brexit. She eventually resigned, Boris Johnson won the leadership race to replace her as prime minister, and he got a third extension, pushing the deadline to January 31. Then Johnson won a landslide election victory and easily passed legislation confirming the January 31 date.
So at the risk of making the same mistake twice, I really believe that Brexit will happen in 2020. Johnson wants a hard exit, he’s passed legislation to do it, and he has a majority that May did not, meaning the threats she faced of the Northern Irish Democratic Unionist Party or Conservative breakaways forcing a general election no longer apply for Johnson. In any case, the possibility of Jeremy Corbyn and Labour winning and conducting a second Brexit referendum never came to pass either.
But one lesson I learned with this prediction is to pay more attention to the distinction between “Event A will happen soon” and “Event A will happen this year.” —DM
Narendra Modi will continue as Indian prime minister after the 2019 elections (60 percent) — CORRECT
As the polls suggested, Modi and his right-wing National Democratic Alliance, led by Modi’s own Bharatiya Janata Party, swept Rahul Gandhi’s center-left Congress Party in May’s elections. The policy ramifications have been vast: Modi seized control over Jammu and Kashmir, denying the region the autonomy it enjoyed for 70 years; used the National Register of Citizens to strip citizenship from nearly 2 million people; and passed legislation making Muslims ineligible for a new pathway to citizenship, sparking massive protests against Modi and his government’s anti-Muslim discrimination.
I got the easy part right: who would win the election. I didn’t anticipate quite how catastrophic the consequences of Modi’s victory would be. —DM
Neither India nor China will enter a recession (70 percent) — CORRECT
As I mentioned when laying out this prediction the first time, neither China nor India ever fell into recession in 2008-2010; indeed, it’s hard to identify the last time either of them was in recession, as it appears never to have happened in the 21st century.
And economic weakness in 2019 was not enough to drive either into full recession, just as the 2008 financial crisis was not enough. Both Chinese and Indian growth slowed, but not to negative territory.
It’s important to note that Indian growth is likely lower than its official numbers state; the Center for Global Development’s Justin Sandefur and Julian Duggan point out that a lot of correlated stats, like non-oil imports and exports and capital/infrastructure production, are declining. But even if growth is overstated, the latest year-on-year growth figures were 5 percent and 6 percent for India and China, respectively. They would have to be really dramatically off for either country to have been in recession in 2019. If an error that size is confirmed, I’ll happily correct this. —DM
Malaria deaths will decrease (80 percent) — CORRECT
Our data here operates at a year lag, but per the World Health Organization’s (WHO) 2019 Malaria Report, about 405,000 people died of malaria worldwide in 2018, compared to 416,000 in 2017 and 585,000 in 2010. Given that the world’s population has grown since those years, these strides are even greater in incidence per capita terms. What’s more, fewer people died of malaria in 2017 than the WHO initially thought — 416,000, not 435,000, as initial estimates had it.
So the global trend toward malaria reduction, if not quite eradication, is continuing apace. If you want to help, the Against Malaria Foundation and Malaria Consortium are two great charities to consider. —DM
No additional countries will adopt a universal basic income (90 percent) — CORRECT
Here I’m defining “basic income” as an unconditional payment to all adult citizens, not subject to any work requirements, means testing, or other restrictions. As I noted when making this prediction, Iran arguably has an unconditional basic income program in the form of an oil revenue dividend. Alaska, though obviously not a country, arguably does too, through its oil dividend.
While Rahul Gandhi endorsed a limited form of basic income in his campaign for Indian premier, he lost the election, and the Modi administration didn’t pursue the idea. The city of Maricá, Brazil, has launched a basic income for about 52,000 people, but the policy is not quite universal for the city, and in any case, it doesn’t cover a whole nation.
So basic income didn’t take a country-size stride this year. But it’s still chugging along, as the Maricá example shows. That pilot is also funded in part by oil revenues, which seem to be the most reliable way to get a policy like this started. —DM
More animals will be killed for US human consumption in 2019 than in 2018 (60 percent) — LIKELY CORRECT
In 2018, 9.59 billion land animals were killed for consumption in the US. (This is actually only a small fraction of all animals that died as a consequence of US food production — estimated at 55 billion animals — because many fish are also killed for food, even more fish are killed as bycatch or as feed fish, and many animals are raised for food but die before they reach the slaughterhouse.) Not all the numbers we need to estimate this are available yet from the US Department of Agriculture, but most of them are. Based on January-November numbers, it looks likely that more animals were killed in 2019 than 2018, so I’ve tentatively marked this correct — expect an update within a month when all of the 2019 data is in and we can answer definitively.
2019 was, as we’ve written, a huge year for the plant-based meat movement. So why are more animals dying than ever? Mostly because plant-based meats started out as such a tiny niche industry that, even after all their growth, they’re only about 2 percent of packaged meat sales. So all that growth can’t cancel out the fluctuation in demand for meat due to normal economic factors — the US population is still growing, and meat demand has been rising since the end of the recession. Maybe in a few years that’ll change, and we’ll start seeing the effects of plant-based meat. But not yet. —KP
Impossible Burger meat will be sold in at least one national grocery chain (95 percent) — WRONG
You can buy Impossible Burgers in grocery stores around the country: Gelson’s Markets throughout Southern California; Wegmans in Maryland, Massachusetts, New York, New Jersey, North Carolina, Pennsylvania, and Virginia; and Fairway Market in NYC. However, to my horror, I have just now realized that my original prediction said “national grocery chain” and each of these are regional grocery chains.
Let this be a lesson for me: Even if you have information making you highly confident in a claim (I knew the Impossible Burger team was getting ready to launch in grocery stores, that the final FDA approval was on the way, and that burger production was gearing up), you shouldn’t be too confident in your own phrasing matching what actually happens. In other words, to be willing to predict something at 95 percent confidence, you should be that confident not just in the events but also in the specific wording you used.
I feel very foolish here and will be much more careful in 2020. —KP
Fully autonomous self-driving cars will not be commercially available as taxis or for sale (90 percent) — CORRECT
This is the other prediction I was very, very confident in, and here I got it right. It was pretty obvious last year that, despite optimistic declarations from some self-driving car companies, the technology still wasn’t there. Driving is incredibly complicated, and while milestones in other areas of AI have been surpassed with surprising speed, the challenges of teaching a car to drive itself have been persistent despite tons of talent and investment.
Self-driving car technology is continuing to make slow steps forward, with the industry leader Waymo having cautiously begun driver-free test runs for members of its heavily vetted early-rider program in Arizona. But unless you happened to be part of Waymo’s beta program, you still cannot take a fully autonomous taxi anywhere in the US — and you can’t buy one anywhere either. —KP
DeepMind will release an AlphaZero update, or new app, capable of beating humans and existing computer programs at a task in a new domain (50 percent) — CORRECT
Can you even call a 50 percent prediction “correct”? I’m not sure, but being right 50 percent of the time (and not 30 percent or 70 percent of the time) when you say something is 50 percent likely is a skill just like other skills in making predictions, so we’ll post some predictions we’re 50/50 on. (If we’re any good at this, they’ll be right half the time and wrong half the time.)
Anyway, DeepMind’s AlphaStar fits the bill here — it plays StarCraft at an expert level (beating top experts in some high-profile show matches before DeepMind, responding to criticism from the professional StarCraft community, installed a lower cap on its reaction time). It cleverly solves some longstanding problems with using self-play to teach AI systems to get better at computer games, using a tournament style to ensure that its most promising agents don’t have easy-to-exploit vulnerabilities.
While it definitely fits the bill for this prediction, I want to note that AlphaStar strikes me as, on the whole, a less impressive achievement than AlphaGo, the DeepMind AI that defeated the top human Go players. While AlphaGo dramatically and immediately surpassed human-level play and now plays Go at a level that’s far beyond ours, AlphaStar plays like a good, imperfect human. It’s an achievement, but not the most impressive one we’ve seen from DeepMind or its competitors. —KP
Average world temperatures will increase relative to 2018 (60 percent) — CORRECT
The World Meteorological Organization said in late December that 2019 was on track to be the second- or third-warmest year on record, with the January-November period warmer than every year except 2016. We’ll update this once they publish their final data, but it looks nearly certain that 2019 will be hotter than 2018, which was itself one of the five hottest years on record. Annual trends will always fluctuate some, including due to cyclical weather events like El Niño, which helped make 2016 the record-holder for the hottest year, but the overall trend is unambiguous: All five hottest years in history happened in the past decade. 2019 continued the trend. —KP
Global carbon emissions will increase (80 percent) — CORRECT
The Global Carbon Project’s scientists estimated in papers and presentations in December that global carbon emissions indeed increased again in 2019, setting a new all-time high. They estimate that carbon emissions increased 0.6 percent over 2018, which was itself an all-time high.
(Those numbers are only projections, based on emissions data up to December, and there’s some chance they’ll be revised.)
This is depressing news! All our treaties, agreements, progress on cleaner energy, and new energy technologies can’t even get our emissions to the level they were at in, say, 2010, let alone to the drastic decreases in emissions that would mitigate the worst effects of climate change. The report notes, optimistically, that emissions grew more slowly in 2019 than in 2017 or 2018 — but they’d briefly leveled out entirely in 2014-2016, so I’m not inclined to take “slower growth” as a very encouraging sign. —KP
Sign up for the Future Perfect newsletter. Twice a week, you’ll get a roundup of ideas and solutions for tackling our biggest challenges: improving public health, decreasing human and animal suffering, easing catastrophic risks, and — to put it simply — getting better at doing good.
Future Perfect is funded in part by individual contributions, grants, and sponsorships. Learn more here.