Predictions made by Mati Roy in early 2020
Author: Mati Roy | Created: 2020-01-11 | Updated: 2020-03-01 (Adjusted: 2020-11-07) | Published: 2020-11-07
Quality: Those are notes I took for my present and future selves; they weren’t taken with the purpose of informing others.
Importance: 3⁄10. I don’t think all those predictions are important / transformational. It’s likely not important for you to read this.
Epistemic status: My credences are already quantified for each prediction. But my predictions are not independent of each other. For example, shorter AI timelines could mean shorter timelines on all those predictions, and vice versa. I might also be miscalibrated on long-term predictions — I don’t have first hand experience with that having only been born recently.
Context: I wrote some predictions in January-Ferbruary 2020 (some with links to PredictionBook and Metaculus with the exact date available on those platforms). I hadn’t published it at the time because there were a bunch of other topics I wanted to make predictions on. They are mostly predictions made on a 10-100 years time horizon.
Value to me: Making this research was much more time consuming than I thought. It was an interesting exercise, and will likely be interesting to check back on those predictions in the future. Here’s how I felt going down this rabbit hole.
Medium: Ideally, I would like a prediction platform where I could post my comments and predictions on public questions, like Metaculus, but with the additional option of having a personal public page automatically aggregating all my predictions and comments, similar to what I did here.
50 billion animals are raised and killed in factory farms every year. Most experience extreme levels of suffering over the course of their lives due to intense confinement and the removal of body parts. The meat industry is also one of the largest contributors to climate change, with 14.5% of global greenhouse gas emissions. (source)
A reduced population of farmed animals would also likely reduce the risk of pandemics.
It would also likely allow us to create meat for cheaper.
It might also provide a more robust food supply.
It would also allow us to create healthier meat, with less contaminants.
It would also allow us to create a wider range of food, such as synthesizing new forms of meat.
It might also make it easier to sustain human life on other celestial bodies.
85%: Meat and/or dairy products from cellular agriculture will be commercially available by January 1st, 2030. (PredictionBook)
75%: A company with its main activity being cultured meat and/or dairy products will be valued at >= 1bn 2020USD by January 1st, 2030. (PredictionBook)
52%: The U.S. plant-based and cultured-based retail market will be worth more than 9bn 2020USD throughout 2029. (PredictionBook)
Predictions using Elicit: https://elicit.ought.org/builder/nUEGbEPgb (h/t Jungwon)
2020-01-22: Memphis Meats receives 161 million USD in its Series B fundraising, more than the total investment in cultivated meat companies so far which was of 155 million USD. (source)
“In the next few decades I believe that cell-based meat will become a major part of ourglobal meat supply.” -Richard Branson, investor (source)
There are now at least 43 cultivated meat companies on five continents producing 15 types of meat, plus a handful of B2B companies. BlueNalu, Future Meat Technologies, Finless Foods, Wild Type, and Aleph Farms all raised $10 to 20 million Series As in 2019. Memphis Meats, however, is the first company to raise a Series B round. (source)
The U.S. meat and poultry industry accounts for $1.02 trillion in total economic output or 5.6 percent of gross domestic product (GDP). (source)
The plant-based meat category alone is worth more than $800 million, with sales up 10 percent in the past year. Plant-based meat now accounts for 2 percent of retail packaged meat sales. (source)
Investment trends: https://www.gfi.org/industry
Having a fast, general-purpose quantum simulator could help a lot in designing new drugs, materials, solar cells, high-temperature superconductors, chemical reactions for making fertilizer, etc.
Breaking existing public-key cryptography.
Optimization and machine learning
I put ~15% probability that noisy intermediate-scale quantum will solve at least one useful problem and/or be economically viable by January 1st, 2030. I’m not sure how to properly operationalize this either. Ex. of applications: certified randomness.
Quantum supremacy (ie. being able to do something which we can’t do with classical computers) is a spectrum, but we’re currently transitioning to it. (Metaculus)
I’m not sure how to operationalize this; maybe the ratio of energy it takes for a classical computer to solve the same problem as a quantum computer? Here’s another idea: “Will it cost less than $1000 to calculate the private key of a 2048-bit RSA public key in 2030?” (Metaculus).
Quantum-computing experts are now finding themselves obliged to repeat a constant refrain: it’s not just about the number of qubits. One of the main measures of the power of a quantum circuit is its so-called depth: in effect, how many logical operations (‘gates’) can be implemented in a system of qubits before their coherence decays, at which point errors proliferate and further computation becomes impossible. How the qubits are connected also matters. So the true measure of the power of a quantum circuit is a combination of factors, which IBM researchers have called the “quantum volume”
But I couldn’t find any graphs of the maximum quantum volume achieved over time.
It would also be nice to operationalize how much noise a quantum computer has.
John Preskill says:
we might succeed in building large-scale quantum computers after a few decades of very hard work. (source)
[noisy intermediate-scale quantum] might also have other useful applications. But we’re not sure about that. (source)
given the spectacular progress by Google and others over the last few years, my guess is that we’re at most a decade away from some small, special-purpose quantum computers (with ~50-200 qubits) that could be useful for quantum simulation
a shift from pure academic research to venture capitalists and industrial efforts just within the last 4-5 years [from 2018-06-06] (source)
Note: It could be economically viable, yet not economically benefitial (ex.: if people only buy quantum defenses to secure themselves against quantum attacks).
China’s National Laboratory for Quantum Information Sciences is a US$10 billion project due to open in 2020. (source). Europe’s Quantum Technology Flagship is a €1 billion euros project that opened in 2016. There are ~7000 researchers publishing ~8,500 articles per year [as of 2014] (source).
Longest quantum entanglement distance: February 2020 record: 50 km; previous record was 1.3 km (source).
Investment trends: https://www.cbinsights.com/research/report/quantum-computing/
Popularity trends: https://www.cbinsights.com/research/quantum-computing-startup-ecosystem/
Other predictions: Metaculus
More info: https://www.scottaaronson.com/blog/
Economic prosperity of having cheap and abundant energy. Energy is a $8.5 trillion energy industry.
It takes roughly 55,000 barrels of oil to heat 10,000 homes for one year. With fusion energy, it would take one liter of deuterium and tritium, extracted from water, to power those 10,000 homes.
That limited residue has a half-life of a few hundred years, not tens of thousands of years, and hence is more manageable than nuclear fission waste.
And whereas those 55,000 barrels of oil would release 23,500 tons of carbon dioxide, fusion produces no emissions and will have a lifecycle carbon intensity lower than solar or wind (as measured in CO2 from all construction, manufacturing and operations per kWh produced). (source)
pro-fusion advocates also say that fusion reactors would be incapable of generating the dangerous runaway chain reactions that lead to a meltdown (source)
But fusion reactors have other serious problems that also afflict today’s fission reactors, including neutron radiation damage and radioactive waste, potential tritium release, the burden on coolant resources, outsize operating costs, and increased risks of nuclear weapons proliferation. (source)
Also see: Pure fusion weapon — Wikipedia
Now , privately funded fusion energy start-ups like TAE Technologies, Commonwealth Fusion Systems and General Fusion [...] are leaving the laboratory and starting to build prototype demonstration plants. (source)
The JET (Joint European Torus) is the world’s largest operational magnetically confined palsma physics experiment. It started its operation in 1983. (source)
ITER is a large international project to build a magnetic confinement plamaphysics experiment 10 timers bigger than the current biggest. The project was officially initiated in 1988. The building costs were over US$14 billion by June 2015, and the construction of the facility is expected to be completed in 2025. Initial plasma experiments are scheduled to begin in 2025, with full deuterium-tritium fusion experiments starting in 2035. Since the 1950, more than 100 fusion reactors have been built. (source)
DEMO, ITER’s successor, is planned to be constructed in 2024-2033, and have a first phase of operation in 2033-2038. After which, it would be expanded and updated, with the second phase of operation starting in 2040. In 2012, the roadmap aimed to have an electricity generation demonstration in 2048. (source)
PROTO, DEMO’s successor, is planned for after 2050. PROTO would act as a prototype power station demonstrating electricity generation on a commercial basis. (source)
The fusion energy gain factor, usually expressed with the symbol Q, is the ratio of fusion power produced in a nuclear fusion reactor to the power required to maintain the plasma in steady state. The condition of Q = 1, when the power being released by the fusion reactions is equal to the required heating power, is referred to as breakeven[.] [...] With typical fuels, self-heating in fusion reactors is not expected to match the external sources until at least Q = 5. If Q increases past this point, increasing self-heating eventually removes the need for external heating. At this point the reaction becomes self-sustaining, a condition called ignition. Ignition corresponds to infinite Q, and is generally regarded as highly desirable for practical reactor designs.
Operating above engineering breakeven, a machine would produce more electricity than it uses and could sell that excess. One that sells enough electricity to cover its operating costs is sometimes known as economic breakeven. Additionally, fusion fuels, especially tritium, are very expensive, so many experiments run on various test gasses like hydrogen or deuterium. A reactor running on these fuels that reaches the conditions for breakeven if tritium was introduced is said to be operating at extrapolated breakeven.
As of 2017, the record for Q is held by the JET tokamak in the UK, at Q = (16 MW)/(24 MW) ≈ 0.67, first attained in 1997. ITER was originally designed to reach ignition, but is currently designed to reach Q = 10, producing 500 MW of fusion power from 50 MW of injected thermal power. The highest record for extrapolated breakeven was posted by the JT-60 device, with Q<sub>ext</sub> = 1.25.
Optimists believe that we are five years away from the first proof-of-concept fusion energy demonstration plant and 15 years from a commercial roll-out at scale.
I mentioned three private start-ups in the beginning of this piece because they all are on viable paths to commercialize fusion in the next five to 15 years. If any of them win, we all win. But they all are underfunded.
TAE Technologies (formerly Tri Alpha Energy) has raised $800 million over the past 20 years and earned notoriety for being backed by Microsoft co-founder Paul Allen. TAE has been experimenting on a $100 million reactor and talks about commercializing the technology in five years, for which they are expected to start raising additional capital soon.
Commonwealth Fusion Systems is backed by the Italian oil company Eni and the celebrity billionaires of Breakthrough Energy Ventures. They expect to add fusion power to the grid in 15 years but have raised only $115 million so far. Again, too little.
General Fusion, based in Burnaby, British Columbia, is currently engineering a $350 million demo plant, set to go live in five years. If that demo plant operates well, General Fusion could have its first commercial plant built in 10 years from now. To date, General Fusion has raised funds from the Canadian federal government, Khazanah Nasional (Malaysia’s sovereign wealth fund), Cenovus Energy, Braemar Energy, Jeff Bezos and Chrysalix, among others.
These are three of about 22 private companies in the fusion game, so of course, there’s intense competition for the few investors willing to back these high-risk and capital-intensive but very high-return innovations.
Google’s US$10-million project aimed to test the cold-fusion claims rigorously in a field that lacked credible scientific data, says Trevithick. (source)
The main source of tritium is fission nuclear reactors. (source)
Graph of US investment: https://i.huffpost.com/gen/2470392/original.jpg
Projection for leading fusion projects: https://2oqz471sa19h3vbwa53m33yj-wpengine.netdna-ssl.com/wp-content/uploads/2017/07/fusion-timeline.jpg
Kurzgesagt video: https://www.youtube.com/watch?v=mZsaaturR6E
As of September 2016, there are 711 known asteroids with a value exceeding US$100 trillion.
Although part of the wealth is at current market value, which would drop a lot with a higher offer.
Unlike Earth, which is protected by its magnetic field, the Moon has been bombarded with large quantities of Helium-3 by the solar wind. (source)
This could prove very useful if we develop viable nuclear fusion.
On April 24, 2012 a plan was announced by billionaire entrepreneurs to mine asteroids for their resources.
Nearly 38,000 people died in car crashes in 2016, according to the National Highway Traffic Safety Administration, and most serious crashes are due to human error. (source)
Autonomous vehicles could add $800 billion [per year] to the US economy, a new report says.
In 2015, some 15.5 million workers in the U.S. worked in jobs related to driving, according to an August 2017 report from the Department of Commerce’s Economics and Statistics Administration.
Only 3.8 million of those workers operate motor vehicles such as a truck or taxi
The 2017 median pay for heavy and tractor-trailer truck drivers was $20.42 per hour, which adds up to $42,480 per year, according to the Bureau of Labor Statistics.
It is estimated that autonomous cars could eliminate 300,000 driving jobs a year, according to a May 2017 report from Goldman Sachs. But that won’t happen right away; the report estimated that from 2025 to 2030, autonomous cars will be 20 percent of car sales.
Nearly 38,000 people died in car crashes in 2016, according to the National Highway Traffic Safety Administration, and most serious crashes are due to human error.
self-driving cars could contribute to major gains in fuel efficiency, lower transportation costs to the consumer and increase access to rural areas.
Car owners will be able to put their cars to work as a taxi when they aren’t using it. In fact, most people won’t need to own a car because of that.
Given that fleets of self-driving cars are better coordinated, traffic will also be reduced.
More info: [video, 4:29] The Simple Solution to Traffic
Given that on-demand cars will be as cheap as owning a car, and that these cars will be safe, smaller cars (for 1 or 2 passengers) will make sense.
This added efficiency will also result in less cars being built and less greenhouse gases per distance.
Although given it will be cheaper, more distance might be driven.
50% Level 3 autonomous cars will be commercially available by 2030. (PredictionBook)
50% Level 4 autonomous cars will be commercially available by 2036. (PredictionBook)
50% Level 5 autonomous cars will be commercially available by 2042. (PredictionBook)
50% Level 5 self-driving air vehicles will be available by 2030. (PredictionBook)
Clarification: Needs to be available for commercial use on public roads in at least one city (not just on the highway).
When will Level 3 self-driving cars be available commercially? (Metaculus)
When will Level 4 self-driving cars be commecially available commercially? (Metaculus)
When will Level 5 self-driving cars be available commercially? (Metaculus)
In what year will half of new cars sold in the US be fully autonomous? (Metaculus)
Key Volkswagen Exec Admits Full Self-Driving Cars ‘May Never Happen’ (source)
In 2020, you’ll be a “permanent backseat driver,” the Guardian predicted in 2015. “10 million self-driving cars will be on the road by 2020,” blared a Business Insider headline from 2016. Those declarations were accompanied by announcements from General Motors, Google’s Waymo, Toyota, and Honda that they’d be making self-driving cars by 2020. Elon Musk forecast that Tesla would do it by 2018 — and then, when that failed, by 2020. (source)
Distance driven (ie. training data acquired)
Number of disengagement per kilometer
Number of accidents / deaths per kilometer (notably compared to humans)
Human driving produces one fatal accident in every 100 million miles driven. (source)
A quick primer on the now-industry-standard SAE International rules on how to discuss self-driving abilities: Level 0 is no automation whatsoever. Level 1 is partial assistance with certain aspects of driving, like lane keep assist or adaptive cruise control. Level 2 is a step up to systems that can take control of the vehicle in certain situations, like Tesla’s Autopilot or Cadillac’s Super Cruise, while still requiring the driver to pay attention.
Get past that and we enter the realm of speculation: Level 3 promises full computer control without supervision under defined conditions during a journey, Level 4 is start-to-finish autonomous tech limited only by virtual safeguards like a geofence, and Level 5 is the total hands-off, go literally anywhere at the push of a button experience where the vehicle might not even have physical controls.
February 2020 notes
Volocopter and RTA Dubai cooperate on eVTOL Autonomous Air Taxi (AAT) (source)
You can buy a car that will automatically brake for you when it anticipates a collision, or one that helps keep you in your lane, or even a Tesla Model S (which — disclosure — my partner and I own) whose Autopilot mostly handles highway driving. (source)
In 2018, Waymo drove 1.2 million miles in California, with 0.09 disengagements every 1,000 miles.
Coming in second is General Motors’ Cruise, with about half a million miles and 0.19 disengagements per 1,000 miles. (Cruise argues that since it tests its cars on San Francisco’s difficult streets, these numbers are even more impressive than they look.)
The company doesn’t release specific figures, but its filings for its IPO last year said that it had driven “millions” of miles.
March 18, 2018, was the first time a self-driving car ran down a pedestrian.
A report from the National Transportation Safety Board implicated Tesla’s Autopilot system in another lethal 2018 accident; while the driver had his hands off the wheel, the car steered into a concrete divider and crashed, killing him.
A full investigation hasn’t yet been conducted on three more recent deadly Tesla crashes.
The problem, according to NTSB chairman Robert Sumwalt, is that drivers assume Autopilot lets them take their attention off the road, when they shouldn’t.
the laws surrounding self-driving cars vary enormously by state, with 29 states having passed legislation.
2014: $167 million
2017: #3 billion
First 3 quarters of 2018: $4.2 billion
Total so far [early 2020]: at least $16 billion (source)
Tesla has been building an autopilot feature since 2014, and both Apple and [Alphabet] are developing their own self-driving car models.
Traditional automotive companies are also investing. Ford recently announced that the automaker plans to spend $4 billion on autonomous vehicles by 2023. General Motors will pour $100 million into self-driving cars, and Toyota launched a $2.8 billion self-driving car company in Tokyo. (source)
it is exasperatingly difficult to get a good estimate of how long until self-driving cars happen for real for the typical American, both because no one knows for sure and because companies have incentives to publicize optimistic estimates. The companies boast about their progress but don’t publish their mishaps.
Fully artificial kidney
End-stage renal disease, or chronic kidney failure, affects nearly 2 million people worldwide, including more than 570,000 in the United States, and costs the nation almost $40 billion each year for treatment. The most effective treatment is kidney transplantation, but those organs are in short supply; last year, only 16,812 kidneys were available for transplant, leaving 92,000 patients on the waiting list, according to the Organ Procurement and Transplant Network. (source)
50% Fully artificial kidneys will be publicly available by 2042. (PredictionBook)
The FDA announced today (April 9) that it had chosen three renal device projects to pilot a new regulatory approval program called Innovation Pathway 2.0, intended to bring breakthrough medical device technologies to patients faster and more efficiently.
The artificial kidney project, which is targeted for clinical trials in 2017, was selected for its transformative potential in treating end stage renal disease and for its potential to benefit from early interactions with the FDA in the approval process.
investors and granting agencies to fund the expected $20 million it will cost
8.5 million people in 2030 (source)
50% CI: 8.3-8.7
Male birth control pill
50% A male birth control pill will be commercially available in the US or EU by 2045. (PredictionBook)
Still, 10 years to market seems optimistic to him. It’s expensive to make an idea into reality, and lengthy, well-designed studies are required to make sure such a contraceptive is safe, effective and reversible.
The promise of a safe and effective oral male contraceptive “has been always on the verge of becoming a reality, but I feel like I’ve been hearing that for a long time now,” said Najari, who was not involved in the new research. “I think it’s going to take at least 10 years, if not longer.”
More information: https://en.wikipedia.org/wiki/Male_contraceptive#Research
Man mission to Mars
50% A hole will be drilled down to Earth’s mantle by 2090. (PredictionBook)
The Integrated Ocean Drilling Program has been in operation since 2003 [...] The effort is expected to take many, if not dozens, of years, and may require $1 billion.
Notes without predictions
At least one other CRISPR baby will be born by January 2030.
In its 2016 power production report, the Geothermal Energy Association (GEA) found that only 6 to 7 percent of global geothermal power potential has been tapped. However, production is on the rise and the GEA predicts global geothermal energy production will more than double by 2030. (source)
The main battery technology used to be lead-acid, then came alkaline, then nickel-cadmium, and now lithium-ion in many different chemical configurations. In the future, we expect solid-state batteries, and perhaps boxes with nanoscale flywheels, flow batteries, or supercapacitators.
Bloomberg expects 1,000% (a tenfold) growth between now and 2025, reaching 11% market share for BEVs in the total private vehicle (PV) market by then. It then anticipates another 170% growth from 2025 to about 30% in 2030. That is followed by only 80% growth from 2030 to reach a 55% market share in 2040. Why Bloomberg expects the growth curve to flatten when the products get better and cheaper instead of following the normal S-curve is not discussed.
Battery development is on a technology curve. Characteristics of this curve are a ~14% decline in price ($/kWh) and a 6% increase in density (kWh/kg). Combined with power electronics, battery and motor management computers, motors, and gears, a combined decrease of 7% per year is used.
Environmental concerns call for increasingly stricter exhaust standards, making internal combustion engines and exhaust-treating systems more expensive, resulting in an expected 3% yearly increase.
Number of drug-resistant infections
The number of drug-resistant infections will increase (70 percent)
Antibiotics are great when they work, but because we’ve overused them, more and more infections are becoming resistant to them. 2020 will continue that trend.
I’m basing this prediction on two high-level reports released last year: one from the Centers for Disease Control and Prevention (CDC) and the other from the United Nations. The reports presented some pretty terrifying findings: Resistance to second- and third-line antibiotics (often the last lines of defense) is projected to almost double between 2005 and 2030. If we don’t make a radical change now, drug-resistant diseases could kill 10 million people a year by 2050 — up from 700,000 a year now.
Note that I’m not predicting the number of deaths from drug-resistant infections will increase relative to 2019. It’s totally possible to make progress on that front. In fact, such deaths have decreased by 18 percent since 2013, in part because more hospital professionals started to heed experts’ warnings. But even if we manage to stabilize or decrease the death toll, the number of drug-resistant infections could still rise — and the trend suggests that’s likely to happen, because we aren’t addressing our overuse of antibiotics with anything like the necessary speed. — SS
Facial recognition will be banned in at least three more cities (70 percent)
In 2019, we saw a growing backlash against facial recognition technology. San Francisco, Oakland, and Berkeley banned it in California, as did three communities in Massachusetts: Somerville, Brookline, and Northampton. In 2020, I predict we’ll see at least three more cities institute a ban on the controversial tech.
To be clear, I’m talking about a ban that applies to city departments like police; I think outright bans that would also cover businesses, individuals, and federal agencies are way less likely.
I’m partly going off local news about particular cities — Portland is currently deliberating a ban, and the western Massachusetts city of Springfield might be next. Last year saw mounting pushback against facial recognition from AI researchers, groups like the ACLU, low-income tenants in Brooklyn, and many more. Their protests seem to be growing bolder, not quieter.
I should note that according to Pew Research Center survey data, most Americans are now in favor of police using facial recognition. I don’t think a nationwide ban is in the cards for 2020 (sorry, Bernie). But a lot can still happen on the city level, and I think it will. — SS
Global carbon emissions will increase (80 percent) (source)
Average world temperatures will increase relative to 2019 (60 percent) (source)
California has a wildfire among the 10 most destructive in state history (60 percent)
Future of bets
A recent Supreme Court decision has cleared the way to bring an estimated $300 billion in black-market sports betting into the light. (source)
Notable early applications: controller for a VR headset; computer navigation for handicapped people
Metrics: words per minute typed; vocabulary it can detects + accuracy
Other topics I was interested in predicting
Technology / politics
Fuel cells cars
3D printed building (note: Dubai has built the world’s largest 3D printed building)
Small nuclear reactor (note: Pentagon awards contracts to design mobile nuclear reactor — skimmed)
Daily imaging of Earth (see: Planet Labs, Inc.)
Remote work: Telehealth
Smart cities; private cities (note: Smart City Will Privately- Owned Towns Drive the Development of Smart Cities? — not read)
Vaccine without needles (not read)
Advance autocomplete (ex.: drawing)
FDA average years to approve drug
Most populous cities
Genetically modified animals (operationalisation: If DNA alterations continue to require FDA approval by default, how many intentionally genomic DNA altered animals will be determined as safe to eat, by the end of July 2025? ; Metaculus)
Emergency called automatically
Physical money doesn’t exist in the US anymore
More money in proptech
Porn with created from images of people of your choices
Porn generated by AI based on keywords of your choice
Reinforcement learning to generate porn tailored to your preference
North Korea and South Korea unification (PredictionBook)