What is intelligence? – no really…
Much discussion of AI focuses on defining intelligence. Kevin Kelley makes a persuasive argument that intelligence is not a single thing, but is competence in any measurable dimension. The old idea of an IQ test providing a single number of intelligence for a person is laughably naïve. Likewise, comparison of non-human intelligence with human intelligence is also fruitless.
Are dolphins more intelligent than humans? In their ability to navigate underwater, certainly. Is a calculator app more intelligent than a human at doing arithmetic? Of Course. Is a human more intelligent than a horse because humans can tell horses where to go? Perhaps. But if the human is sufficiently drunk, the horse is smart enough to get him home safely, unlike the drunk human attempting to navigate a brainless automobile. It all depends on the context in which some action can be measured as a successful expression of knowledge.
Without exception, intelligence is expressed by the ability to do something – specifically, an expression of knowledge. This follows from the general definition I give for intelligence: As a requirement, life forms input information, process it using memory and logic (create knowledge) and may express their knowledge as an energetic action in the environment. We call this intelligence. By no accident, this is also the definition of a Turing machine, although Turing and Von Neumann never cast their eye on the thermodynamic aspect of Turing machines. In addition, it is the basis for the recent description of the connectome, which links the totality of an organism’s functions into an irreducible whole. You can’t just chop out the brain and have a useful model to emulate. The Theory of Constructal Information (CTI) describes this as the “completeness requirement” for the definition of life forms as irreducible units.
Amplification of energy is the purpose of intelligence
Next, every existential organism is bound by thermodynamics. Expressing knowledge (intelligence) in the environment, by definition requires energy. So, the ability to control energy as an input or an output is a key feature of intelligence. How much energy does your intelligence control?
Let me now assert that all intelligent life forms are required to amplify energy from their input to output. This is simple to prove, but off topic for today. For example, our survival instinct is our drive to harvest energy to stay alive. We expend energy to obtain more energy than we need for bare survival in order to grow, evolve and/or reproduce. We use tools and automobiles and computers to further amplify the energy available to us as individuals and groups. Moreover, the same energy amplification instinct drives every one of the dependent cells in our bodies, along with the bacteria, yeasts and viruses. This principle is true for all life forms, regardless of their mechanism of operation, DNA, human social or technical.
I will argue that control of energy is of equal importance to raw processing power in assessing the future of our new machine life forms. And life forms they are – coevolving with us humans for millennia, now moving from dependent servants to independent actors.
Information processing computers were historically optimized to minimize both the input energy and the output energy. As incomplete dependent life forms, they amplify our ability to process information which is of great value, returning to us humans more than the energy (time, money) needed to buy and operate the computers. In other words, their ability to reproduce themselves is dependent on their ability to provide value to us in excess of their cost of ownership. If they create more value than they consume, we buy more of them, driving their reproductive cycle. Even biology itself shows us that the ability of individual organisms to reproduce themselves is not a requirement for life. Take the ants for example.
Most legacy computers input information as typing, trackpads, voice or data input from storage or communication. It requires little energy. I checked on Amazon, and could find no keyboards that require gasoline engines. Likewise, most output is just more information – processed to be sure, but low energy. How much energy does it take to create a PDF and email it? The article you are reading is the embodiment of my expressed knowledge, dubious as it may be. The process of writing and submitting it, even accounting for my breakfast, which powered my keyboard finger-wagging, and your pizza or avocado toast or whatever you consume to power your reading, doesn’t amount to much energy compared with say, crushing a car into a three foot square cube. In the physical world, big energy levers are being pulled.
When intelligent machines get control of substantial energy
Today, advanced computers are being harnessed to cars, factories and many physical, energetic systems. This defines the AI/robotic revolution. Any intelligent machine, to be useful to its owner or itself, must amplify energy. Therefore, I submit for discussion that the energy impact of machines capable of independent action is much more important than their presumptive intelligence.
The boundary conditions of our control over these systems are now coming into view:
– What we as a human society want the machines to do
– What the machine does for itself based on its own value system, with or without our personal consent.
At first, we will use our self driving cars to provide us more personal energy – eating, learning or having sex while in transit. But, what happens when the cars network themselves into an independent IoT, whose mandate is to optimize transportation efficiency by whatever means possible. We may find ourselves at the beck and call of the cars themselves. “I’m sorry, you can’t go to the pizza parlor now, it’s too far away, so I’m taking you to a much closer tofu bar.”
Suppose an AI charged with energy management by a world government decided that since they need cheap energy in Africa, the rational course is to strip mine all the coal available in the United States and ship it over there? What if it raised a robotic army of drone machines and went about its task without regard to the wishes of the humans? It wouldn’t require any great super intelligence, especially if one or more humans give it direction. What happens when people protest and try to take out the machines? Oops, middle of winter, the power in Washington DC goes off for a week or two and the protester’s cars no longer function.
Follow the energy.
So wrapping up this rant, I think we need more discussion about how machine intelligence controls energy, rather than worrying about the hypothetical development of some super intelligent brain in a jar, as much fun as that is.