Who Will Represent Humanity in Interstellar Space?
In a recent Paradigm podcast interview, Matt Geleta asked me who should represent Earth in interstellar space? He reckoned that humans cannot survive the long trip and so humanity will likely use an avatar with artificial intelligence (AI) as its flagship towards a possible encounter with aliens in interstellar space. In that case, which humans should the AI system replicate as a representation of humanity?
I replied that humanity’s ambassador should contain the most complex neural network system that can be launched to space, potentially better than any human who ever lived on Earth. There were 117 billion samples of human intelligence that inhabited Earth over the past ten million years.
Assuming a normal (Gaussian) probability distribution of intelligence quotient (IQ), the most intelligent biological brain which ever existed is ~7 standard deviations away from the mean. If we construct an AI system that is more than 7 times better than the average human brain, it would perform better than any human who ever lived.
Based on Moore’s Law, if the number of parameters in our most advanced AI systems will grow exponentially with a doubling time of 1–2 years, then within 2 decades artificial neural networks will be more complex than biological neural networks.
This will allow humanity to launch an AI system more impressive than any human brain that ever existed on Earth by the middle of this century. I will be proud of such a system more than a replica of a human, for the same reason that I am proud of my daughters’ accomplishments even if I cannot replicate them.
However, the space launch of an advanced AI system may take much longer than the time required to produce it on Earth. The root of the hurdle is in the required power supply. Currently, the training of Large Language Models (LLM) consumes gigawatts of electric power.
Increasing the number of parameters in them or neuromorphic AI systems by a factor of ten thousand so as to exceed human performance, would require tens of terrawatts given their current architecture. This represents ten times the global electric power consumption on Earth. Providing this much power to a spacecraft is challenging. For comparison, the human brain runs on merely 20 watts.
It is remarkable how difficult it is for our technologies to reproduce what nature does so effectively thanks to billions of years of evolution. Indeed, it is difficult to maintain human hubris in light of our limited scientific accomplishments. Consider nuclear fusion as another example.
Experimentalists at the National Ignition Facility (NIF) of the Lawrence Livermore National Laboratory (LLNL), achieved breakeven in the nuclear ignition of 0.2 milligrams of deuterium-tritium fuel over ten nanoseconds whereas the sun achieves hydrogen fusion for billions of years in a fuel reservoir that is 37 orders of magnitude larger in mass.
These two shortcomings of our technologies are related. Once we develop a concept for a compact fusion reactor, it could power our most advanced AI system in space.
Given that most stars in the Milky-Way galaxy formed billions of years before the Sun, it is possible that aliens resolved both technological challenges long ago.
In case they already sent their ambassadors out of their home exoplanets to interstellar space, these alien avatars might be smarter than any of the senders. In that case, it would be fun to watch how our ambassador interacts with theirs.
After encountering each other, the two ambassadors might choose to partner and ghost their senders, like teenagers falling in love and avoiding their parents.
If that happens, we will keep wondering about their whereabouts, and have no other choice but to connect directly with the aliens instead of delegating this task to our AI avatars. After all, there is nothing more rewarding than a dialogue between a biological brain on Earth and an extraterrestrial brain on an exoplanet.
