What would we get if we put all the ai we had before in the same computer?
We'll get an AI that can have phone conversations in 100 languages. It will beat everyone in Edge of Danger! And beat anyone in chess, Go, poker, and many video games. This artificial intelligence will be able to recognize any object or face and detect cancer even better than most doctors. It will also be done and creative, inventing things, discovering the laws of physics, and identifying new drugs. Artificial intelligence can compose music like Bach and paint like Van Gogh. Its artistic style is also very original.
Ai-AI will become a modern Renaissance machine — with a skill set unmatched by humans.
"Let's put it all together" is smart.
Marvin Minsky
At some point in history, it is possible for one person to master all human knowledge. But with the specialization of the scientific field and the growth of human knowledge, this has not been possible for centuries.
However, AI can store all human knowledge in its brain. Every encyclopedia, every book, every scientific article, even the content of every web page on the Internet. With the ability to cross-reference, analyze, discover patterns, and summarize written text, as GPT-2 does, it's only a matter of time before we build an entity with an understanding of human collective knowledge.
The intelligence of the algorithm
In 2000, Marcus Hutter, who currently works at DeepMind, discovered a general-purpose ARTIFICIAL algorithm called AIXI.
General AI can be described by an equation.
Subsequently, the algorithm achieves perfect optimal intelligence. It works by calculating every possibility arising from each actionable course of action and then selecting the action closest to the target.
If the goal is to win a chess match, AIXI calculates every possible future chess match resulting from each possible move and selects the one with the most wins and the fewest losses among its possible future moves.
The algorithm has a drawback: it is not computable — it requires unlimited computing resources to solve.
Hutter's formula does provide an important lesson: Intelligence is not complicated—the difficulty is to achieve intelligence with limited computing power. Achieving intelligence with limited resources requires finding and employing shortcuts—matching patterns and applying heuristics—to replace brutal computing.
It seems that we now know what to do. DeepMind's AlphaZero is a learning algorithm. OpenAI's GPT-2 is an algorithm for understanding. BAIR's CycleGAN is an innovative algorithm.
With intelligent algorithms, human-level AI has only one obstacle: primitive computing power.
Neural networks
AlphaZero, GPT-2, CycleGAN, Microsoft's speech and object recognition, Google's GMNT translation, all of which are based on artificial neural network technologies that are themselves inspired by the brain.
Neural networks have been in use since the 1960s and were a major topic of interest to AI researchers in the 1980s. But when neural networks can't do anything useful, the field of artificial intelligence stagnates.
However, their failure is not due to the wrong principle, but to not having enough computing power.
In the 1980s, trying to build useful AI with computers was as difficult as trying to go to the moon with a bottle rocket. This principle is correct, but power does not exist.
As computing power increases, larger neural networks with more neurons and more layers become computationally viable. These networks are called deep neural networks.
In artificial neural networks, inputs, such as pixels of an image, are processed layer by layer by layer neurons until the final output is obtained.
Deep neural networks just use more layers.
Each layer of the network can identify unique features individually. This is what Google unveiled in 2015: DeepDream, a visualization technique seen at every layer of deep neural networks.
Each layer "sees" something, each layer extracts a different element of the picture. Computers produce hallucinations.
Today's deep neural networks can have dozens of layers and millions of neurons. The recent breakthrough is not the result of discovering some new intelligent paradigm, but the result of building faster processors and larger data sets. The current AI revolution is almost entirely the result of increased computing power.
Thanks to pioneers like Marvin Minsky in the 1980s and Jürgen Schmidhuber in the 1990s, we have had the right approach for decades. However, until recently, we have not been able to run and train networks large enough to display interesting and intelligent behavior.
Now, as the capabilities of our best computers approach the primitive computing power of the human brain, we are seeing more and more human-like achievements in the field of artificial intelligence.
How much time is left before our computers and the artificial intelligence they provide surpass us?
Calculate trends
Improvement is the essence of technology, and its rate of improvement is exponential.
When the amount of growth is proportional to the size of something, there is an exponential increase. For example, interest paid to a bank account is proportional to its balance.
As long as there is exponential growth, there will be a constant times the time.
In 1965, Gordon Moore, who would later found Intel Corporation, noticed that the power of computers was doubling almost every year. This trend dates back to before the 1960s and has continued ever since.
The productive capacity of the world economy is developing in the direction of infinity
As long as there is a feedback mechanism, there will be exponential growth. For example, the more knowledge we have in making computers, the faster and more powerful we can make.
The faster we build computers, the faster we collect and process information to expand our knowledge, which includes new knowledge about making computers. This cycle repeats itself and feeds on itself.
We observed that every indicator of our knowledge grew exponentially — the number of new patents filed, the number of scientific papers published, and the total amount of digital data stored.
The number of patents (representatives of technological progress) has grown exponentially. Image source: Our Data World
Today, it's just a starting point. If the power of computing technology continues to double every year, then in 10 years, our computing power will grow 1,000 times: 1,000 times faster and 1,000 times more memory.
Applied to artificial intelligence, it means that artificial intelligence becomes 1,000 times smarter than before every 10 years.
In the next 25 years, there will be about 30 times the growth. In terms of capacity and price performance, this is 1 billion times higher than today's technology, which is already quite powerful.
Ray Kurzweil
Even if AI catches up with us, it won't stay on our level for long. It will fly past us.
Intelligence explosion
We feel as if we are in the midst of something big —a paradigm shift, the next stage of life, and so on.
The record of these feelings dates back at least to the late 1950s.
The theme of one conversation, the continuous progress of technology and the constant change of human way of life, led us to see that there was some fundamental singularity in human history beyond which human affairs as we know it could not continue.
In 1958, Stanislaw Ulam recounted a conversation he had with John von Neumann
Danny Hillis, who works with Minsky at the MIT Artificial Intelligence Lab, likens it to being in the middle of the s-curve.
The first step in the story I told took a billion years to complete. The next steps, like the nervous system and brain, took hundreds of millions of years. The next steps, such as language and so on, took less than a million years. And the next steps, like electronics, seem to take only a few decades.
This process is self-reliant, I suppose, to be described by the word "autocatalytic," where something increases the rate at which it changes. The more changes, the faster the changes. I think that's what we're seeing in this curve explosion. We see this process in self-feedback.
Danny Hillis
We live in the most exciting times in history. But as far as we know, does it pose a threat to human life?
The equation of the end of the world
What will be the next stage of life and what possibilities will it bring? How much time do we have before it hits the world stage? Various indications from different fields suggest that this could continue for decades.
Heinz von Foster is well versed in computer science, neurophysiology, mathematics and philosophy. The Pentagon funded von Foster to create and lead the Biocomputer Lab, where he was a pioneer in the field of cybernetics.
During his career, he published more than 200 papers, but he is best known for the 1960 doomsday equation.
Von Foster's equation for the end of the world is the result of an analysis of trends in population growth. He and his students collected and analyzed the size of the population over the past two thousand years. They found that it was growing faster than exponentially.
This growth is not exponential, but hyperbolic.
When the exponential trend doubled at a constant rate, von Foster found that the time between doublings was shortening. They plotted the time at which this doubling was expected to reach 0 — and if the population grew along this trend, it would reach infinity. They came up with the following prediction: 2027 AD±5.5 years.
A similar pattern has been found in economics. Economic historian James Bradford DeLong collected data to estimate the world's GDP for the previous 1 million years. Again, when plotted, it shows a decreasing trend in time between two consecutive doublings.
It suggests that at the beginning of the 21st century, the economic doubling time will reach zero.
Two researchers created a model of population, technology, and inventors to estimate when the world's technology developed. Their conclusions are:
Extremely simple mathematical models have been shown to be able to explain all changes in the macrodynamics of the economy and population in the world's nearly two thousand years of history, from 99.2% to 99.91%.
Andrei Korotaev and Artemy Markov
The data trends are so clear and consistent that data from someone in their time in ancient Rome or the Middle Ages can predict that this trend will end sometime in the 21st century.
The limits of intelligence
A singularity in history
An analysis of the history of technology shows that technological change is exponential, contrary to the "intuitive linear" view of common sense. So in the 21st century, we won't experience 100 years of progress – it's more like 20,000 years of progress (at today's rate).
Ray Kurzweil, 2000
Demographic, economic and technological growth trends all point to technological singularities that will emerge in the near future. This is a moment when machine intelligence far exceeds human intelligence.
Once realized, humans will no longer be the drivers of technological development. Not limited by scientists, inventors, and technologists, the only limit to the speed of technological progress will be the computing resources available to AI-based scientists, inventors, and technologists.
Over time, how much computing power can be purchased for $1,000.
As of 2018, our fastest supercomputer, Summit, has more computing power than a human brain.
In decades of continuous technological advancement, our PCs and smartphones will catch up with Summit's computing power. Around this time, the total computing power of our machines will exceed the total computing power of all human brains.
Important historical developments conform to a binary scale, marking a decline in the exponential number of time intervals, with each interval half the size of the previous one [...] Apparently approaching zero in the coming decades.
The remaining series of faster and faster revolutions will converge into an Omega point between 2030 and 2040, when the computing power of a single machine will be close to the original computing power of all human brains combined. Many readers of this article should have been alive at the time.
Jürgen artificial intelligence pioneer Schmidhuber
Super-smart superpowers
Irving John Good is a mathematician who used a computer with Alan Turing to crack German codes during World War II. Good was one of the first to realize the implications of machines that can improve themselves.
Let's define a superintelligent machine as a machine that can far exceed all the intellectual activities of any one person, no matter how intelligent that person is. Because designing a machine is an intellectual activity, a superintelligent machine can design a better machine; At that time, there will undoubtedly be an "intelligence explosion", and human intelligence will be far behind. Therefore, the first superintelligent machine was the last invention that humans needed to make.
Owen John Goode, 1965
Such intelligence would have many attributes that we might call superpowers. Nick Bostrom's book Superintelligence outlines six superpowers that superintelligent AI might possess:
| Super smart | skill | relevancy |
| 1. Intelligence amplification | AI programming, cognitive enhancement research, social epistemological development, etc. | • The system can guide its intelligence |
| 2. Develop a strategy | Strategic planning, forecasting, prioritization and analysis to optimize opportunities to achieve long-term goals | • Achieve distant goals • Overcome wise opposition |
| 3. Social manipulation | Social and psychological modeling, manipulation, persuasion | • Utilize external resources by recruiting human support • Enable "closed" AI to convince its gatekeepers to release it • Persuading States and organizations to take certain actions |
| 4. Hacking | Find and exploit security vulnerabilities in computer systems | • AI can requisition computing resources over the Internet • Boxed AI may exploit security vulnerabilities to evade cybernetic limitations • Theft of financial resources • Hijack infrastructure, military robots, and more. |
| 5. Technical research | Design and modelling of advanced technologies (e.g., biotechnology, nanotechnology) and development paths | • Create powerful military forces • Create a monitoring system • Automated space colonization |
| 6. Economic productivity | A variety of skills make cost-effective intellectual work possible | • Generate wealth that can be used to purchase influence, services, resources (including hardware), and more. |
Given its superpowers, a superintelligence that unites against humans would be a curse. We have little chance of beating it.
However, if we had superintelligence, that would be a good thing. It can cure any disease, design any technology, solve any problem, and even end world hunger and poverty.
Superintelligence allows us to see progress over the next 100 years and 20,000 years.
Everything that civilization can offer is the product of human ingenuity; When this intelligence is amplified by the tools provided by AI, we can't predict what we'll achieve, but eradicating war, disease, and poverty will be at the top of nowhere. The successful creation of artificial intelligence will be the biggest event in human history. Unfortunately, this may also be the last time.
Stephen hawking
For two decades, we have witnessed the rise of intelligent machines. Creative machines, machines that learn, make us mistakenly think they are human. If so much progress can be made in such a short period of time, what will happen in the coming centuries as the power of computing technology continues to grow exponentially?
There are limits that even superintelligents can't overcome.
For example, limits such as the speed of light and the density of black hole matter. The laws of physics imply physical limits to computer processing speed, data density, and energy efficiency.
However, the ultimate physical limit of computation is extraordinary. Bremerman Limit makes the speed of the computer as fast as possible to reach a mass of 10^50 operations per kilogram per second.
The Summit supercomputer reaches 10^16 per kilogram per second, meaning we are about 10^34 and it is impossible to build the best computer. Before we reach that limit, computing technology can be doubled another 112 times.
The best physically possible computational material is called a computium. It's something in science fiction, but we can use known physical boundaries to speculate on the capabilities of computer atoms.
Humans often think of themselves as the pinnacle of intelligence, but in fact, even if all the intelligence of the human brain adds up, it is only a fraction of what is possible:
Distinguish between orders of magnitude diagrams of different intelligences
In the chart above, each increase of 1 represents a 10-fold increase in computing power.
The pocket calculator can do an average of 10 arithmetic operations per second, so on the chart it is "1". A modern smartphone that performs 1 trillion calculations per second, "12".
Chimpanzees, humans and Summit supercomputers are around 18. The total computing power of all the computers in the world is about 10^21 per second, and the computing speed of 7 billion human brains is 10^28 per second.
At this scale, all humans together are somewhere between a pocket calculator and a Matoshka brain, a hypothetical computer powered by a star that operates 10^{48} per second at the limits of Laundauer's computational efficiency.
The 1 kilogram of computronium, though much smaller but operating at physical limits, has 100 times the capabilities of this stellar-powered computer, reaching 10^{50} operations per second.
Jupiter weighs about 10^27 kg. Convert the entire mass of Jupiter into computronium and you can have a computer with a computing power of 10^77 per second.
Ultimately, AI will have powerful computing power, and through computer simulations, it can explore the entire evolutionary history of other worlds and civilizations in a fraction of a second.
Consider that in human history, about 100 billion people have lived. If each person lived an average of 40 years, the total amount of time that humans experienced was 4 trillion years.
Four trillion years is about 10 to the 20th power of seconds. Because every second of human brain activity involves 10^18 operations, then all the experiences that all humans have can have 10^38 operations.
A 1 kg computronium of artificial intelligence can experience the entire human race in a trillionth of a second.
A Jupiter brain's computronium can dream of 10^39 civilizations in one second.
Is our own existence in such a dream?
The potential capabilities of future computers raise many questions such as "Are we living in computer simulations?", "Is it possible to create new universes?", "Is life insignificant in grand plans?""