Hundreds of years of searching for extraterrestrial life, we may find that our first contact will not occur at all with organic beings.
Small Sombrero Galaxy, or NGC 7814We have a problem. In the 10 billion-year-old galaxy, there should be plenty of opportunities for at least one species to break out of its dustbin and settle in the stars, filling all the niches. The fact that these creatures did not come to us brings us to
the Fermi paradox - if life is not incredibly rare, then where is everything? Attempts to scan the heavens in search of signs of intelligent life were not crowned with anything, which only strengthens the paradox. Perhaps the blame is the endless abyss of interstellar space and the narrow time frame in which there are able to communicate creatures that are at a sufficiently small distance from each other to be heard. The mind can be like small craft, passing in the night in the vast ocean. Face-to-face encounters can be extremely unusual.
Another explanation for the great silence of the Galaxy may be that any surviving intellect may be so different from us, to develop so much that we cannot even imagine our form or behavior. As a result, detecting and identifying it will be almost impossible. Pretty depressing.
But there is still a possibility that lies between these extremes, and which has the greatest probability. When, finally, our first meeting or the discovery of activity happens, machine intelligence can appear to our gaze.
This idea is not new at all. In the 1940s, mathematician
John von Neumann studied the possibility of the existence of non-biological self-replicating systems that consider well and do not require a mind controlling them. Later in the 1980s, others expanded this concept, considering the real engineering technologies needed for autonomous, self-replicating devices capable of traveling in space; these machines could travel through the universe and find raw materials to build more and more of them, creating an infrastructure for obtaining energy from space or for human settlements among the stars [this idea was expressed by the authors of the Stargate / approx. trans.].
But if such devices had real artificial intelligence, they would clearly have a more complicated mission. What would be its purpose, and what kind of AI would these machines have? Meeting with an alien machine could help us solve this riddle.
One possibility is that this machine will turn out to be superior in cognitive or analytical capabilities. Such an AI would be extremely difficult to understand, either because of a complex motivation, or because of practical barriers associated with the bandwidth of communications. For such a device, a conversation with us would resemble a conversation with a baby, or an attempt to discuss the collected works of Shakespeare with the help of pictograms. An alien system optimized for processing a large data stream may not even be able to slow down sufficiently enough to notice that we are trying to communicate with it, whether we will use technology for this or not.
Alien AI can also be very scary just because of its machine nature: some animated thing, consisting of inanimate parts, very much like the classic story of a
golem made of clay or mud. Even if a biological alien appears to be shocking in appearance, he will surely have some similarities with us. We can convince ourselves that evolution leads to recognizable behavior and intentions, which we can even sympathize with. An artificial being will not need to follow all these rules of evolution, which will bring the alien alienness to a new level.
A meeting with someone else's AI will not only show us our possible future, but will probably also serve as a fun change to our worldview. When Nicolaus Copernicus in the XVI century suggested that the Earth was not the center of the Universe [the first known heliocentric theories were proposed by
Aristarch of Samos in the III century BC, however, then it did not produce such an effect / approx. perev.], he launched the development of a critical scientific idea: that on a cosmic scale, we do not represent anything special or important. But the encounter with an alien AI can turn this understanding on its head: if the only intellect we meet is artificial, then it turns out that we are still special in something.
If we still meet AI on our cosmological excursions, it probably will not be reasonable in the full sense of the word. I think we will meet with some specialized intelligence - extremely capable of scientific tasks, but otherwise extremely limited.
A recent example here on Earth is AlphaGo Master, a go-to-board board game system that learns from a large catalog of games played by people. A year after writing, she began to beat the best human players. Then came the next version of the system called AlphaGo Zero. For 40 days of training, without referring to the historical data on the games, this new AI began to play better than its silicon ancestor and any living person. Another option, AlphaZero, took just 24 hours to train and beat the well-trained version of AlphaGo Zero - as well as other programs that play both chess and
shogi , sometimes called Japanese chess. Read the interviews with expert players, and you will see that these machines do not even play like people - their strategies are alien and seem unknowable.
All these alphas are an example of a specialized machine. Such AI will help us to answer a whole mountain of scientific questions that are at the very edge of accessibility. The researchers behind the creation of Alpha systems are already reasoning about riddles about protein coagulation — how chains of amino acids are twisted, coagulated, and folded into structures with important biological properties. The molecular laws that work in this case are understandable, but it is extremely difficult to say what exactly a particular configuration will turn into. This really seems like a challenge for some future AlphaFold Zero.
Or take the other tasks of modern science. Cosmological simulations, predictions of climate and weather, depend on physical phenomena, such as gravity, fluid dynamics, and thermodynamics. Computer programs digitize these phenomena, counting billions of iterations of interaction and motion. But these calculations are still limited by the speed and accuracy of the processors. AI with in-depth and independent learning may be a better predictor than these intensive numerical simulations, limited by a monotonous, systematic approach. The intuition of a specialized machine may be able to skip through all these cycles, and “see” a probabilistic answer, instead of having difficulty reproducing its pixelated version. In the same way, the intuition of a specialized machine may be applicable to the root process of scientific deduction and discovery. All the possibilities are yet to be explored, but they may turn out to be extraordinary. Specialized machines can be so useful and revolutionary in empowering discovery and research that there is no motivation to look for something more interesting.
If this scheme of development of machines is developed in our country, it can demonstrate the regularity occurring throughout the entire universe, and affecting the very nature of space exploration. Specialized AIs can be the ideal tools to allow biological life to reach across interstellar distances: more reliable than biology, and not smart enough to raise ethical questions about sending scientists into space according to the first whim of scientists. The launch of millions of such researchers into space seems an inevitable step [such an initiative is sponsored by the Yuri Milner -
Breakthrough Initiative ]. They can be sent to other stars and planetary systems, and customized to recognize the most interesting characteristics of those places, from astrophysics to the presence of life. These machines would reflect the motivation of their biological creators: perhaps a little curiosity, perhaps something else.
Another excellent reason for specialized machines to explore space is when thinking about creating a more powerful general-purpose AI. It is possible that an IIN can be built from a multitude of smaller specialized AIs, each of which is engaged in a specific, complex cognitive task, put together. These connections may be localized or scattered, recalling the architecture of biological neural circuits. There may be a specialist in facial recognition, in natural language, in calculations, in tax optimization. You can imagine the AI, whose task will be to learn how to combine specialized AI in something like ION - something like a manger for developing machine intelligence.
The most difficult task in the nursery for AI will be to familiarize the team of specialists with the complexity of the real world. Nature does not consist in achieving fixed goals; it is full of noise, randomness and trillions of interacting parts. For example, from the very moment of embryo formation, it undergoes a constant change. At first, these are just a few cells that communicate with the world at the molecular level. With the development of the embryo, the complexity of the organs for registering light, sound, touch and smell - the portals to experience is expanding.
In short, the best way for creator AI creatures to improve it would be to allow him and his components to study an untidy Universe. A single planet can be complex and rich in information, but space is filled with worlds that represent millions, billions, trillions of natural test tubes, each of which has its own story about natural selection and chances. The distribution of specialized AI among stars gives us a way to take advantage of these endless natural experiments and input data.
The path to creating a better AI can be parallel to biological evolution only by some artificial means. Think about the development of the human brain. The main process does not allow retry attempts. It begins with the fact that stem cells quickly generate the original structure, the
neural tube , multiplying at a rate of 15 million divisions every hour, and are connected to the developing embryo. This amazing assembly of the brain is followed by training — the brain builds the contours and cuts them as the experience creates its unique life path. And although we are incredibly flexible and able to perceive and learn all sorts of skills and knowledge, as individuals we are also, unfortunately, limited to the genetic alignment that has fallen to us. And to be honest, although we can constantly learn and change, we rarely generate new wonderful talents throughout our lives.
The machine may not be so limited, especially if its critical parts are engaged in the study of space, being designed to change and transform into something new. Specialized components of larger intelligence may appear and disappear in an ever-changing pattern. Some cognitive skills may be useful in certain contexts, but may simply interfere in other situations. In other words, machine intelligence can be extremely flexible, changeable, and fast-paced. As a result, a huge diverse AI zoo can lurk in space.
In a sense, such AIs resemble microbial machines that are at the beginnings of life on Earth — forming the smallest units of intelligence that cross the Universe, and combining together, as biologist
Lynn Margulis suggested, to create multicellular creatures, larger organisms, and the mind itself. Moreover, when species have long been extinct in the relentless stream of natural selection or self-destruction, such machines may remain.
AI outside the Earth can be difficult to recognize. In the same way, as some resident of the Mongolian steppe of the XII century would consider the ro-mob both magical and meaningless, we may not be able to recognize or interpret the presence of specialized machines by the age of a billion years. Conversely, these machines may not be looking for conversations with us, or in general, any other intelligence.
But the recognized meeting even with one specialized machine will change everything. She will inform us that the galaxy is full of intelligence, and will be able to cause the assumption that our future may contain the disappearance of the biological presence. And most importantly, this discovery can tell us that at the moment we can be the only natural mind aware of all these facts. This is because biology, capable of producing AI researchers, is likely to evolve or become extinct by far smaller time scales than the life of these interstellar machines, and we already live in a 10 billion-year-old galaxy [the authors series Stargate Universe / approx. trans.].
Detecting alien AI can open the way for us to explore space, highlighting it. It may also give us ideas about the nature of its creators, these intellectual ancestors, who may have had a biological form. How exactly this research process will look is extremely difficult to imagine. Even a single specialized AI may not be a single device, but a swarm of smaller components that will be very difficult to classify. However, imagine that by questioning or disassembling, we can eventually solve the riddle of the origin of someone else's AI. We can find evidence of the existence of another organic species, or discover that in the whole chain of development there are only machines.
Caleb Scharf is director of astrobiology at Columbia University in New York. Author of the book “Extrasolar Planets and Astrobiology” (Extrasolar Planets and Astrobiology, 2009).