When the modern computer was first created in the 1960s, people soon started imagining a future with intelligent machines. Some of these visions were dystopic, like Skynet in the Terminator movies. Today, almost everyone takes it for granted that artificial intelligence (AI) on par with human cognition is just around the corner, but is it realistic? Surprisingly, there are good reasons to be pessimistic about the prospects of smart machines.
Materialism and Reductionism
The basis of AI optimism is the widespread conviction in materialism and reductionism. Materialism is the belief that everything is made from dead matter, that consciousness is an illusion, and that the human mind is the product of a machine – the brain. Reductionism is the belief that everything can be understood by chopping it up into its parts and only studying their properties as if it were machinery. It has been the backbone of many scientific and technological achievements.
Its success has led people to believe that it also applies to consciousness and intelligence. All we need to do is to write a clever program and run it on a sufficiently powerful computer and voila: It will be as bright as us.
The Limits of Reductionism
Despite the utility of reductionism, it is a philosophical error to assume that everything is reducible. Consider the following analogy: 99.99% of the universe is an empty vacuum. Everywhere in the cosmos, we see only vast swaths of nothingness, punctuated by an odd galaxy. From our own Milky Way, we know that even galaxies are mostly empty. If you used this to conclude that the universe was void of content, you would almost be correct. Almost. But that speck of matter in the cosmos turns out to be immensely significant. We are made of it, and we live in a place where we are surrounded by vast amounts of it. Matter matters to us.
Thus, we cannot reduce the world to emptiness. We need to account for that exceptional state that we call matter. Similarly, almost all matter in the known universe is dead and unconscious, but we also by extraordinary coincidence happen to be alive and conscious. We know this from our direct experience, and no amount of peering into the dead material world can undo that fact. Consciousness is real and may not be reducible to material causes.
In addition to our direct experience of consciousness, evolutionary theory provides compelling evidence that it is not merely an illusion. Anything that evolves through natural selection must be both heritable and objectively measurable. There is no way for consciousness to be selected for and evolve unless it does something.
So what is the biological function of consciousness? We don’t know for sure, but it likely plays a crucial role in intelligence. This can be demonstrated with a task that is simple for us: object recognition. At an early age, we recognize a panoply of objects with ease, which even the most powerful supercomputers are incapable of today.
The task is so trivial to us that scientists thought it would be easy for computers too. It turned out to be nearly impossible. Although computers can beat humans in highly specialized computationally intensive tasks such as chess, they are nowhere near the ability of even a toddler in categorizing and recognizing objects. The key ingredient seems to be consciousness, which enables powerful intelligence at a mere glance.
Computers will continue to get smarter in specialized domains, but they will never gain awareness because it is not an algorithm. Thus, they will have to emulate human intelligence without the power of consciousness. Don’t be surprised if AI continues to fall short of expectations.