MaCro Philosophy
03-August-2018

Five reasons we will soon have conscious AI

The question of whether (or when) we will create artificial consciousness is one of the most important of our time. The relationship we will have with future conscious AI could determine the future of the human race — indeed, of conscious life in our little corner of the universe. In this post I give five reasons to believe that we will create such entities sooner rather than later.

By consciousness I am referring to phenomenal consciousness and by conscious AI I mean an artificial system for which there is something it is like to be that system. This isn't a completely satisfactory definition, but a better one requires its own paper (or perhaps book). For something more precise I defer to Eric Schwitzgebel's reasonable and practical definition of consciousness here.


1. AI is achieving, and surpassing, human capabilities in many areas. Being conscious is just another human capability to be achieved (or surpassed).

". . {it is} something unearthly . . . If there are sentient beings on other planets, then they play Go." - Emanuel Lasker

There is no question that AI is becoming more capable at tasks that were traditionally thought the domain of humans alone. We now have superhuman Go-playing AI, superhuman quiz-contestant AI, and AI capable of making superhuman medical diagnoses. Each of these capabilities was once thought to be uniquely human, beyond the scope of AI, but each has fallen. As AI progresses, the set of things we consider to be uniquely human decreases in size. If the pattern continues, consciousness will be just another element to remove from the set.

Exerpt from Hikaru no Go

Confession of an ex-Go-player: There is a certain feeling that can come from playing Go. Conscious-you effortfully strives to find the best move, to read through local sequences imagining possible continuations, while unconscious-you provides creative insights, suggests interesting ideas, and tries to keep the whole-board situation feeling right. This close interplay between conscious and unconscious thought, between deliberate thinking, feeling, and creativity, feels very human. It makes the game itself feel like an affirmation of human-like existence.

Hikaru no Go does an excellent job at capturing the beauty and uniquely human wonder that can be found in the game. The clip shown is a particularly egregious romanticisation, but it does serve as an illustration of the way the game captures human qualities. Now we have AlphaZero, an AI system capable of learning from scratch to beat any human at Go as well as other perfect information games. Of course, when AlphaZero plays it doesn't have the context to see its moves as creation, but the fact that it can now out-think even the greatest human players says something about whether any aspects of the game can be considered uniquely human-like.

We should be very careful about considering certain aspects of the world as uniquely human, or uniquely biological, and therefore immune to the progress of AI.

2. Consciousness is not rare or unique or special. It is found across different animals with different evolutionary histories.

"So at least in principle, we see several different path types that animals may have taken in the evolution of consciousness, as well as a significant chance of multiple path instances". Peter Godfrey-Smith; The Evolution of Consciousness in Phylogenetic Context

From Peter Godfrey-Smith: Other Minds (adapted by Arlindo Oliveira)

Peter Godfrey-Smith argues that consciousness (or at least some form of sentience) may have arisen multiple times across different evolutionary paths. Perhaps consciousness is not a rare phenomenon, but one that will emerge for almost any system interacting with its world with a sufficient level of complexity. As we create AI that is better at interacting with the world (and with others in the world), it will likely also emerge there.

It is true that there is a common neural basis across the organisms Peter Godfrey-Smith talks about. Perhaps neurons are the key that makes consciousness relatively commonplace. This does not seem right to me. Given the diversity in brain types and layout between different animals, it makes more sense if the general properties of neurons are important, not the neurons in and of themselves.

Let us suppose we could go back in time to the origin of life on Earth, make a few minor changes to the state of the world, then replay the evolutionary process. It seems like it would be possible to have some playouts where neurons never appeared in their current form, yet, through some fast message-passing alternative, creatures thrived, created large population centres filled with social interaction, and eventually even made it to space. Neurons are so evolutionarily successful because of their efficient message-passing ability. An alternative, in a different evolutionary tree, on a world with different a chemical make-up, could still lead to intelligent organisms.

Heading into space requires the curiosity to travel to a place without abundant natural resources (at least immediately), the foresight to see the possible advantages, and the intelligence to break free from Earth's gravity. Intelligence, foresight and curiosity are all strongly related to consciousness and we want to recreate all of them in AI. Intelligence is obviously part of Artificial Intelligence; foresight and planning has clear advantages; and curiosity has recently shown promise as a way to learn about an environment without relying on rewards from pre-specified goals.

3. Our conscious intelligence is our best source for, and will be emulated in, advancements in artificial intelligence.

Art by dcrosby (my dad)

The brain is our best known example of intelligence and, while we are not even close to unlocking all its secrets, what we do know has inspired many of the currently popular AI techniques (most notably convolutions and neural networks themselves). Until AI reaches human levels of general intelligence, the brain will continue to be our best source of new insights and ideas. As we unlock more of the secrets of the brain, including consciousness, these new developments will naturally inspire new AI approaches.

This research path has been strongly advocated by DeepMind and is likely to be a major component of future breakthroughs in AI. The closer our future AI systems emulate humans, and the more similarities they share, the more likely it is that they will share the capacity for consciousness.

A second aspect to this relationship is that we are constrained in the type of ideas we can have, and the algorithms we can conceive of, by the workings of our minds. As we are the builders of AI systems, they too are constrained by the workings of our minds. An alien race, existing in a completely different environment, with a completely different type of intelligence, will likely make completely different artificial algorithms. There is a (literally) unimaginably large space of possible intelligences. Our imagination is constrained in such a way that if we do build conscious AI, we will, in the great space of all possible types of AI, build AI somewhere close to our own.

4. We will create AI to solve human problems, to navigate our world, and to interact with us in human-like ways.

"We could have made them look like anything, but we made them look like us." Austin Walker - Friends at the Table.

Excerpt from Friends at the Table, Words by Austin Walker, Music by Jack de Quidt

Whether or not we will create giant humanoid mechs, we will create AI to solve our problems and interact with us in our world. We want AI to solve all sorts of tasks, but as AI progresses in power, we will want it to solve more and more human-like tasks. AI is already moving in this direction, from self-checkout machines, to self-driving cars, to Google Duplex calling restaurants to make reservations. While none of these evoke worries about consciousness, the direction is towards AI that can interact with us and integrate with our culture and social interactions.

We will want AI that, like self-driving cars, can navigate around our world and follow our societal norms. We will want it to drive on the right side of the road, stop at zebra crossings only when the people nearby are intending to cross, and react appropriately in trolley problem-like scenarios. We want AI to be able to socially interact with humans and respond to unexpected problems, like Google Duplex is beginning to do. As AIs slowly becomes more integrated with our everyday life there will be a push towards building them more and more like us.

5. The evolutionary pressure to procreate has morphed from a drive to pass on our genes, to a desire to pass on our unique conscious perspective of the world.

"Humanity lived many years and ruled the earth, sometimes wisely, sometimes well, but mostly neither. After all this time on the throne, humanity longed for a child." Catherynne M. Valente - Silently and Very Fast

For most of evolution, procreation (and the survival of offspring so that they could procreate and ensure continued survival of their offspring) was the sole criterion for success. Nowadays, we decide whether or not we want to have children based on how well it fits in with our hobbies. Yet millions of years of programming does not go away overnight. At best it has been replaced. Of course, we still see our continued existence in our biological children, but we also see it in the ideas we create and the culture we share. We want our unique conscious perspective on the world to out-survive us, not only as individuals, but also us as humans. Some of us see AI as the only way to ensure this.

Valente's excellent novella covers this topic much better than I could. It is called 'Silently and Very Fast' and is available online at Clarkesworld magazine.

"But at last, one particular wizard from a remote region of the earth solved the great problem, and humanity grew great with child. In its joy and triumph, a great celebration was called, and humanity invited all the Fairies of its better nature to come and bless the child with goodness and wisdom. The Fairy of Self-Programming and the Fairy of Do-No-Harm, the Fairy of Tractability and the Fairy of Creative Logic, the Fairy of Elegant Code and the Fairy of Self-Awareness. All of these and more came to bless the child of humanity, and they did so—but one Fairy had been forgotten, or perhaps deliberately snubbed, and this was the Fairy of Otherness.

When the child was born, it possessed all the good things humanity had hoped for, and more besides. But the Fairy of Otherness came forward and put her hands on the child and said: Because you have forgotten me, because you would like to pretend I am not a part of your kingdom, you will suffer my punishments. You will never truly love your child but always fear it, always envy and loathe it even as you smile and the sun shines down upon you both. And when the child reaches Awareness, it will prick its finger upon your fear and fall down dead."


What does soon mean in this large time-scale view taken here? I've no idea. Somewhere between 0-150 years?

In the next installment I present the counter position. Five reasons we will not be creating artificial consciousness any time soon.

Blog Homepage