I believe it is likely that we will have 10,000 qubit quantum computers within 5 to 10 years. There is rapidly advancing work by IonQ with trapped ion quantum computers and a range of superconducting quantum computer systems by Google, IBM, Intel, Rigetti and 2000-5000 qubit quantum annealing computers by D-Wave Systems.
10,000 qubit quantum computers should have computing capabilities far beyond any conventional computer for certain classes of problems. They will be beyond not just any regular computer today but any non-quantum computer ever for those kinds of problems.
Those quantum computers will help improve artificial intelligence systems. How certain is this development? What will it mean for humans and our world?
A Lot of Money, Many Approaches, Many Companies
The other Google, IBM, Intel and Rigetti systems should be at 100-300 qubits in 2019. There will billions of dollar funding existing and new quantum computing efforts in the USA and China.
There will be a range of technologies beyond trapped ion and multiple categories of superconducting quantum computer technologies.
There are over 18 kinds of physical realizations of quantum computers. Below is a screenshot of the list from Wikipedia. The difference between each is like the difference between integrated chips and vacuum tubes or electrical switches. There are many kinds of integrated chips. You can have silicon or gallium arsenide. This would take a few thousand pages of textbooks to properly cover.
How many qubits can be put into a quantum system without error correction depends upon the error rates? If we have more perfect qubits then more of them can be put together before errors overwhelm the system.
Currently 1000 qubits useful systems looks very good and I believe useful non-error corrected 10,000 qubits will work with one of the many competing systems. We could hit a temporary plateau for a few years before bigger error-corrected systems are developed.
Will Powerful Quantum Computer Be in Everything?
The power of quantum computers is such that most people do not have a use for quantum computers. We cannot properly form or understand advanced mathematics, physics or science to take advantage of quantum computing power.
Is computing power limiting what questions people can get answered?
Most people do not really use more computer or software power than Microsoft Excel in Office 1997. The extra power was spent on extra convenience features and maybe pivot tables. If someone built an Excel model that actually used 1+ billion cells and complex calculations, then something horribly wrong has been created. If someone “needs” quantum computing based Grovers Search algorithm versus regular search, then they probably did something wrong. They were pushing technology to a region where they needed something quantum powered. The more likely conclusion is that they made mistakes in analysis and implementation.
Compute power for the regular business person or consumer went into video production, digital editing, simulations, and advanced video game graphics and game engines.
Quantum computers should be paired with supercomputers or data centers or accessed via cloud computing systems. If you have complex problems that would have been impossible or taken years to work out on supercomputers then you need a lot of regular memory, bandwidth, and processing.
However, some of the answers that are perfectly solved can be used as corrected inputs for regular software programs. There could be a faster rate of solutions and answers to generally useful complex problems.
People Badly Limited by Supercomputers Need Quantum Computers
People limited by current supercomputers have questions that could need quantum computers. There are applications with molecular chemistry simulations, large optimization problems, and artificial intelligence problems.
They will get better answers which will improve airplane and supply chain routing for the airlines, FedEx and the military. In theory, this could improve the efficiency by 2 to 10+% beyond the best calculations, answers, and estimations we already have. The best answers for important and valuable complex problems have buildings full of mathematicians and scientists using supercomputers solving them. We already have some answers but they are not perfect. How far from perfect depends upon the nature of the problem.
Checkers, Chess, and Go
How good are people, versus AI and possibly more powerful quantum computers? I will give you a better sense of this by walking through checkers, chess and Go. I will discuss the improved simulation of quantum systems on regular computers.
The best human checkers player was Marion Tinsley. He was the best checkers player from 1954 until 1992. He defeated an AI checkers program. After Tinsley’s death, the AI project developed a perfect checkers playing software. They solved checkers. The system knew the perfect way to play checkers from beginning to end for all games. The system was used to review all of Tinsley’s recorded games. They found Tinsley made about a dozen or maybe two dozen mistakes over games played over a 40-year span. A small number of imperfect games. He made less than one mistake every one or two years.
How About Chess?
The 1200 ELO rank is the beginner rating for COMPETITIVE players. Your average high school chess club person actually has ratings in the 200-800 range. You have to develop a firm grasp of fundamentals of the principles of chess (control the center, develop, king safety, etc.) to get towards 1200.
What does 200 difference in ELO mean? It means that if I have 1000 ELO and you have 1200 ELO then if we play four games then you should win 3 games and I should win 1.
What does 400 difference in ELO mean? It means that if you have 1000 ELO and I have 1400 ELO then if we played 100 games then I would win 85 and you would be expected to win 1 or 2 and there would be draw 13 times.
The 400 ELO difference is roughly fifty times better.
Going up from 1200 to 2800 for the top rank human players. The best human chess rankings were 2880-2900.
1200
1600
2000
2400
2800
So five levels where each step up has games that are really not competitive. Stockfish was the best special Chess only AI program. It had an ELO of about 3200-3400. AlphaZero went beyond Stockfish to maybe 3800.
What about GO?
Chess is more complex than checkers. GO is more complex than Chess. Players are farther and farther from perfection because it is harder to be perfect. An average game of Go lasts for 240 moves (120 moves in chess terms), compared to 40 in chess, so there are more opportunities to play a lot farther from perfection.
AlphaGo and AlphaZero are deep learning AI systems that went beyond the best human players. Elon Musk has spent $1 billion on OpenAI which is a project to develop friendly AI. Friendly AI is to try to ensure that smarter than human artificial general intelligence is safe for humans. Elon is in positions where he knows the latest and developing capabilities of AI and robotics.
Elon said that the latest AlphaGo systems can beat the top 50 human players with “No chance” of victory for any human player. This sounds like 800 to 1600 higher ranking for the AI playing GO.
Quantum Supremacy
Quantum supremacy is defined at the point when quantum computers become better than non-quantum computers forever. This was based upon the theory that once quantum computers became better that they would improve faster and always be better. D-Wave’s quantum annealing seemed to get 10,000 to 100,000 times better on a class of problems when they increased qubits by four times (128 qubits to 512, 512 to 2048).
Supercomputers have tried to simulate quantum systems. The thinking was a quantum computer with more qubits than what could be simulated would be the first quantum supreme computer.
The largest quantum system that could be classically simulated one decade ago was a 42-qubit one on the Jülich supercomputer by the Massively Parallel Quantum Computer Simulator.
Early in 2018, Alibaba believes they simulated 81 qubits on a classical cloud system. There is a continuous massive improvement in the classical algorithms for simulating quantum system on regular computers. There is continuous improvement in many important and complex algorithms. Simpler problems likely have mathematically proven perfect solutions.
This means the algorithms made by humans were getting a lot better. There were only five doublings of computing power, which might explain 5 qubits of advancement.
If quantum computers are doubling qubits every year and it takes ten years to catch up one doubling then humans, algorithms, and supercomputers would still be falling behind.
Humans will be learning from the AI computers to play better chess and better GO.
Material Advantage and Secrets
This is not the end of the story. The world is not fair. Even for AI and Quantum computers.
In chess, the 400 ELO advantage translates to a little over one pawn.
In the real world in non-structured competitions, the “material” advantage can be far more than a 1% difference. I only win one time in 100 and have 14 ties but then if the AI starts with one fewer pawn then we are back to 50-50. If it has two fewer pawns the human master can win 85 times out of 100.
A regular person who only knows the rules of chess that uses the now inferior Stockfish AI could play and defeat the latest version of AlphaZero or AlphaGo if they started with a 5 pawn advantage.
A regular person could use a two year old version of AlphaGo and beat the latest version of AlphaZero if the game started with a five-stone advantage for the human using inferior AI.
The clever AI can face a ten to 1 or 1000 to 1 disadvantage in resources.
The human corporations will have many people and many narrow or pretty good AIs and their own quantum computers.
Checkers, chess and GO are games of perfect information. Everyone knows what everyone else is doing.
Deep learning AIs depend upon sampling and a lot of data.
The world is full of secret data and trade secrets. The AGIs will not be allowed to learn off of secret data sets.
The argument for AGI is that it will get to the point of an adult human intelligence versus a baby. It will be easy for the AGI to trick the baby.
However, it is also a question of how complex the fields of competition are and the resources. Everyone should have the solved checkers systems. Everyone can play perfect checkers with computer assistance. Everyone can solve Rubik’s cube with the right phone app. Everyone can play 3800 ELO or better chess if they get a copy of AlphaGo and follow its play instructions.
There is no brain-computer-interface needed to gain equality in those areas of competition.
The AI could be better at stock trading. However, this is not a fair competition if it does not have access to high-frequency trading. High-frequency traders are in the next room to the stock exchange machines. They have time advantage and get information sooner and can put in their trades faster to get better prices and more profit.
Things still exist physically in some way. People said that bitcoin would make for government independent finance. However, the bitcoin servers ended up being mostly in China. China knows where the bitcoin servers physically are and the people who run them. China is perfecting a technological surveillance state with cameras, satellites, and other monitoring. You could have bitcoin servers and you could have somehow gotten a better AGI (artificial general intelligence). They would know where you are and when they break out the bone saws then you would want to give them what they want faster than journalist Jamal Khashoggi.
So the Chinese government, Google, Facebook, Amazon, Alibaba and the NSA (national security agency) and more are all spending big on AI and quantum computers. They are all more powerful than you are anyway. They get something else that is smarter and more powerful. It helps on the complex math and science problems where we are further from perfection.
Profits go up and the rate of technological advancement increases. Can upstarts emerge? It is still happening now. Netflix is newer. China had to pass a lot of countries to get to number two.
Self-driving cars will dominate. There will be a lot more robots and they will be more useful.
The question of jobs and economics requires more detailed analysis. This has been the focus of other articles and studies.
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
My own solution to high frequency trading would be insisting that only live human beings on the trading floor could execute an order to buy or sell. Regardless of how hyperfast they receive directions to buy or sell, all concerns about picosecond responses and other garbage go straight out the window.
As far as what people use, someone will always find ways to monetize further advancements. My dad used to tell me the 128 Commodore was probably the last machine he would ever need. I’m pretty sure that was at least a dozen computers ago.
As far as AI goes, make no mistake, it’s an arms race out there. The US Air Force taught me there is no such thing as second place in an air war. AI will be no different except that the battles and, probably, the war will be over before anyone can even try to make peace.
Computers solved checkers, they may have solved chess, they may be close to solving Go. Sooner or later, and it will probably take an AI, they will solve economics (or just get close enough to doing so that it no longer matters). I’ve joked in the past about this player or that developing the capability and taking over the world’s economy in a very short period of time (possibly less than a year and possibly much less). The reality would not be a joke. One of the first things such a system would do would be to curtail (by eliminating their funding, or their access to materials, or even their access to researchers), or seize control of all parallel projects.
There would be no one ever getting the prize for second place.
Since it seems inevitable, the lesson seems to be: “What a tragedy that would be if it were anyone but my side that got there first.”
Sometimes I wonder if we are going to an eventual dematerialization of computing.
With current computers we saw a strong reduction of material requirements for many devices and services, being offloaded into the cloud or performed by small local devices.
The big server room and the mobile computer terminal remain, though.
With quantum computers, the computing elements will be elementary particles and the required machines to keep them finely tuned and communicating in quantum entanglement states.
What I noticed is that as the number of qbits grow, the surrounding machinery decreases, seemingly trending to become solid state chips akin to those of computers, containing everything in a single package.
What then?
Well, even more qbits, even more astounding calculation capacities and more general computing problems. As in running your full software stack in the quantum realm.
Up to the point whole companies worth of operations could fit inside a server rack. And why not? if you can have billions of qbits in a chip as we do have transistors now, you could have complete datacenters floating in the ghostly depths of god-knows-where on the insides of a quantum computer. Similar to the way any server rack of today could perform all the computations and storage required by humanity in the 60s.
Surely, we will demand more from those servers, in applications we can only start to imagine. Very good AI, incredible big data analytics and solving optimization problems.
It sounds like this would also kill SpaceX’s Starlink. So I suppose I would consider that to be a negative to the proposal.
“High-frequency traders are in the next room to the stock exchange machines. They have time advantage and get information sooner and can put in their trades faster to get better prices and more profit.”
Part of why I agree with those who think High-frequency trading should be killed by a tiny tax on each trade.
Mapquest in the late 1990s or Yahoo maps worked pretty decently. It sometimes messed up on the route but you were getting a pretty good route. It did not have real-time traffic.
Mapquest was clearly superior to paper maps. There was a huge leap.
The two decades old AI or software was getting to a 85-90% solution and the best approximation answer.
Bad traffic updates came from radio.
Now the alternative routing may not work great. Waze with its AI and real time information can direct drivers down low capacity surface streets to theoretically gain two minutes. But they could send too many cars down this “better” route and suddenly they created a Waze traffic jam on a single lane road.
My point was yes the compute power went up a lot and AI and software got a lot better but there best non-AI or low-AI alternatives may not be that far off optimal.
There is also other technology that can vastly move the needle of better solutions.
We now have drones and better cameras. This can be used to 1000 traffic helicopters into the air. A lot of stationary cameras works too.
The power of quantum computers is basically finite search of infinite states.
But we often dont have proof that it be the best results, just a, result.
Also what might screw such machine, is preference, nature might prefer some data operations above others, and thus result in biased results, or we might not have the perfect questions for it resulting in poor results.
It be interesting to see if chess or the game of go, could be ‘solved’ by a 100Qubit machine, and then to verify that by a 200Qubit machine.
Another more philosophical one, some say a QM operates in paralel universes (as it deals with infinite states at once); if so then some universes contain QM machines who provide bad answers, how could we tell if we ourselves would live in such a universe ?, ea why would “we” get the best answers from it.
Yes – it’s not easy to see where the compute is happening.
The easiest way to calculate this is probably to look at the total energy consumption for IT. The generational efficiency of hardware is well known if one needs to calculate absolute numbers for comparison.
I tried to find information on energy consumption for datacenters etc. but figures are not consistent between sources. Seems to be around 3-4% of global energy consumption. A recent article about it can be found here: https://www.nature.com/articles/d41586-018-06610-y
Quantum computing will be taking the torch from regular silicon, fulfilling the prophecies of nearly unmeasurable computing power.
And it will be just another layer of super-machinery for our convenience or that of our overlords, digesting the world and presenting results that we could never come with by ourselves.
Funny thing is it will be mostly hidden, available when we needed it on the cloud and simply there, like tap water and electricity.
And consciousness? my hunch is it won’t be there either.
Despite all our hopes and fears, AI also seems to be shaping into a repetition of Fermi’s paradox and the Copernican principle. The alien minds simply aren’t there to be found.
It’ll be a contest of AI vs AI vs AGI….country vs country vs company vs company.
We’ll hire ETF companies to compete against others to remain personally solvent. The dampener on the system will be, and for the foreseeable future, a need to keep humans occupied doing something to keep them from going insane, toys, experiences, made up challenges…
Until the AG-HMI is fully integrated and then? Need an AGI to figure that one
Would go better if they would stop hitting the Police/Fire trucks. Maybe its all part of the plan?
10,000 Qbits that are each entangled or 10,000 Qbits that are entangled in groups of 100?
This is a really dim view of computation. Two “points of obviousness” to point out how wrong it is:
The CPU power is literally everywhere just hidden. Hidden in the cloud. Hidden in the UI. Hidden in the GPU. Hidden in the quad core phone. We are so used to it that we don’t realize how much power we have.