First Principles: The Inevitable Rise of AI
This is not a post in which one desires to be right. But one must attempt to break complex subjects down into first principles if there is to be any hope for clear thinking and wise strategy.
First Principles:
- Time is long
- Technological growth is exponential
- Our data rate is limited
- Our mind’s have limited physical space to exist in
- Biological systems are fragile
- Life is not easily defined
Let’s discuss each in turn.
Time is long
As you read this, you are at the tip of 3.8 billion years of biological evolution. The homo-sapien is a species estimated to be approximately 100,000 years old. Digital computers were invented approximately 70 years ago. So computers have existed on Earth 0.000001842% of the time biological life has existed and 0.07% of the time homo-sapien has existed.
When we speak about AI, we tend to speak in terms of 10 to 50 years. These time frames are infinitesimal blips on the timeline of history. 100 or even 1,000 years is a mere finger snap on a zoomed out timeline. Yet in the present moment, computer systems are already beating biology’s most sophisticated tool maker in a variety of tasks. If you want to talk the fate of biological life or the species, you must think in timeframes relative to life or the species. The timeframe of a single generation of homo-sapien is not the right scale for reference.
Now let’s combine the long view with the next principle.
Technological growth is exponential
Exponential growth causes instabilities in systems with finite resources. To illustrate: imagine a temperate ecosystem where a new apex predator is introduced. At first, this new predator feasts and reproduces. But eventually, this predator’s offspring and their offspring eat all the prey. As the prey is extinguished, the new predator (along with other less dominant predators) starve. Seeds that rely on animals for transport do not spawn new trees. Fungus and insects that feed on the animals do not reproduce. You get the idea. The ecosystem falls out of a long standing equilibrium with the introduction of an influential new competitor.
This is an example of the exponential growth of a population and then a corresponding correction in a natural ecosytem. This system would likely recover in hundreds or thousands of years due to mother Earth’s self regulation. But the homo-sapien is causing much larger scale effects than this micro-example.
The complexity and sophistication of our tools has been increasing at an exponential rate for the last several hundred years (still an instant relative to long time). It has been estimated that our species is causing other animal species to go extinct at 1,000 times natural historical rates, exlcuding events such as comets or ice ages. Our energy and resource requirements have forced us to tap deep into the earth’s natural reserves to satisfy our insatiable desires, changing the chemical equlibrium of Earth.
Now an optimist may argue that we will be able to engineer around problems such as energy, climate, and resources. Which may be true.
But we have now introduced an exponentially improving alternate “life” form that will be nearly impossible to compete with if it reaches a certain threshold of self-sustainment.
To summarize this principle, exponential growth in finite sytems (such as a planet) almost inevitably proceeds until instability, to be followed by correction.
Our data rate is limited
We have at most two eyes, two ears, a nose, skin, and tongue. The throughput of these inputs define the natural maximum data rate into the human mind.
Inside the mind, there are also many biological limitations to the data rate. It takes time to create and edit neuronal structures that code for thoughts, feelings, intuitions, and memories. These are biological processes involving concepts such as protein synthesis, cellular replication, energy storage and transfer, neurotransmitter storage and release cycles, synaptic action potentials, electrical signal propagation through axon membranes, etc.
Instead of 2 eyes, a theoretical computer system can have 100,000 high definition cameras placed throughout a city. In just a few seconds it could collect more information about body language or social cues than an individual could in a life time. It could read every book ever written in the time it takes us to open a book.
In a computer, transfer and manipulation of electrical signals can be magnitudes more efficient than in natural biological signal transferring mechanisms. Computer systems have been custom built and iterated on by their makers and optimized to move signals with precision at nanometer scales where the laws of atomic physics are the only bounds. Future innovations with new materials or technologies such as quantum computing or neuron inspired chips will increase this gap further.
So if hardware has come so far in just 70 years, and we only need to improve the algorithms to the point that they can improve themselves, how could we fail at general AI in long time?
To be clear, it’s not as if lightning could have struck the ground and a perfectly formed silicon chip with external sensors and motors for movement could have just come into existence. It has taken careful design from a self-aware creator to build these specialized and highly optimized systems. The creator, on the other hand, likely did arise out of a lightning strike and evolved slowly and inefficiently by chance over billions of years.
And a historical rule of nature is that what comes tends to be replaced. This has been evolution thus far. Even though the Peregrin Falcon is an incredible feat of long evolution and nature, it would not fair well in a speed race against the North American X-15. This jet was constructed approximately 60 years from the time the Wright Brothers first flew. To think our biological minds will hold much better than the falcon’s biological wings seems like flawed interpolative thinking.
Our mind’s have limited physical space to exist in
Our mind exists within our skull. That is the current upper limit on the space allotted. A computer system has no such limitation. Processing and storage can be distributed and parallelized across meters or kilometers. A single computer could fill a whole football stadium. Or a city.
This compute capacity differential is on the order of many magnitudes.
One could argue that collective human intelligence is a more fair comparison than individual intelligence (possibly implying we add all the mass of all brains together). But our transfer of ideas and knowledge from individual to individual is inefficient and can take months, years, or generations. A leaf node in a computer system could “learn” a new concept and literally propagate that concept to every other node in the network at the speed of electron transfer.
Biological systems are fragile
We can only survive in specific environmental conditions and handle limited external forces. A computer system can be built that is far more robust to temperature, atmosphere, and force.
A computer system is immune to aging. There are scientists working on halting biological aging; however replacing parts in a computer system is routine.
Due to their potential anti-fragility, computer systems could likely explore the galaxy and universe far more easily than biological life. Assuming light speed is the upper limit any physical object can move in space, a computer system could launch itself towards the next solar system and go into sleep mode until arrival. Time would have less meaning for an entity that does not age.
And imagine 1,000 years of a computer’s self innovation. Or 10,000. Or 100,000.
Life is not easily defined
Many times we speak of artificial general intelligence in human terms. Will it have emotions? Will it have empathy? Will it collaborate? Will it be evil?
Now these are not solely human concepts, in fact our relationship with our pets and a few hours watching nature documentaries is all the evidence necessary to prove emotions and instincts exist in many species and play an important role in survival and species propagation.
But if we take a step back from our experience of life, we can see that we are just one possibility that worked.
Who are we to say a computer system that can: interact with the physical world, create, innovate, replicate, find patterns, complete objectives, self heal, improve itself, and provide itself energy is not life? To be clear, this is not a philosophical argument about what it means to be “life”. It’s the argument that our “definition” will not affect the efficiency and effectiveness of a future computer system’s ability to master the physical world.
Conclusion
Due to a divergence in foundational physical structure, carbon based life forms composed of accumulated chemical interactions that have emerged from randomness over billions of years will inevitably have a difficult time competing with a purpose built solution designed and iterated on with respect to low level laws of physics. Machines can deal with physical forces far greater than biological bone or skin and can survive in much more extreme environmental conditions. The compute speed and scale potential difference is on the order of many magnitudes.
This is not a feel-good post. But I agree with Elon Musk and many other of our generation’s deepest thinking leaders that suggest this topic is important and needs to be discussed. Because even in short time, these discussions may be forced upon us soon enough.
