Will computers ever have consciousness?

Question

Will computers ever have consciousness?

What would it take to accomplish it?

How can we determine if a computer has one?

Answer 1

Will computers ever have consciousness?

Depends on who you ask. 3 possible responses:

1. Consciousness and the mind are non physical phenomena, and computers are physical systems so, no, computers can't be conscious since they lack the non-physical component. The idea that consciousness is non-physical is called (mind-body) Dualism. There are different types of Dualism, and Dualism doesn't necessarily say that the non-physical mind must be found in a living organism. It is possible that the mind could exist in all things, but with no brain to use the mind. It is possible that computers are already conscious without enough "brain power" to show signs of consciousness. (See Panpsychism).

2. Consciousness is a physical phenomenon, the brain is a biological computer and the mind is just software implemented on the brain. This position is known as functionalism or the computational theory of the mind. In which case computers can and most likely will become conscious.

3. Consciousness is physical, but has a non-computational or non-algorithmic component. John Searle, of Chinese Room fame, holds such a view: Consciousness is a purely physical phenomenon, but the simulation of consciousness is not the same thing as consciousness, just like the simulation of bird flying on a computer doesn't constitute real flight. In this case, digital computers won't ever become conscious, but different types of futuristic bio-computers might be become conscious, provided we gain a better understanding of the physics and biology of consciousness.

What would it take to accomplish it?

In the case of (2), we simply need to make more advances in computer programming. 30 years ago we didn't know how to program computers to recognize speech or faces and now we can. Similarly software and hardware capabilities will advance enough in 20 or 30 years to the point where we can program consciousness.

In the case of (3), the physical and bio-chemical principles behind consciousness have to be discovered. Only then can we duplicate them.

How can we determine if a computer has one?

This is the most interesting question in this post. The other answers mention the Turing test. The Turing test is for intelligence (can a computer be as intelligent as a human), not consciousness. We do not know how to determine consciousness, and this problem has been described as the hard problem of consciousness. In fact the idea that on one hand consciousness exists, but on the other we cannot measure it or determine it using any empirical means, is used as an argument for dualism and against physicalism. The reasoning is the following: if something is physical then it can be determined using physical experiments. We know that consciousness exists, but we cannot determine its existence by any physical means, therefore it must be a non-physical phenomenon (See Frank Jackson's knowledge argument).

For more details see the following posts:

• Is the simulation of emotional states equivalent to actually experiencing emotions?
• Can Mind-Body Functionalism be true yet General Artificial Intelligence still be impossible?
• Does claiming that strong AI is impossible imply a belief in substance dualism?
• https://philosophy.stackexchange.com/a/30575/13808

---

As PeterJ points out in the comments, one additional position w/r to the mind-body problem is idealism or immaterialism, which means basically that everything is made up of minds and ideas (whose most famous proponent is Berkeley). In this worldview: the mind and consciousness are immaterial, but so is everything else, so pretty much all bets are off. Presumably in some versions of immaterialism, computers will have consciousness, since they like everything else, are made out of the same stuff as minds.

Idealism/Immaterialism is not a popular position these days and philosophers who work on questions of A.I. almost never subscribe to this worldview, so the question of computer consciousness in an immaterialist universe hasn't been addressed.

Answer 2

The final question is the "easiest" to answer. Our current method of determining whether or not a computer is "conscious" is to see if we can tell it apart from a human. This test is called the Turing Test. If, after talking to both computers running the AI and actual people, we cannot tell who is who, we say that the AI has passed the Turing Test. Is this test really able to determine if an AI is conscious? No. But then again, we cannot really tell whether or not anyone we are talking to is really conscious.

Will an AI ever be created which at least appears conscious? I think so. The processing power of computers is still evolving rapidly. Machine learning has certainly made great strides. But it will still take a while before true AI (what is not referred to as artificial general intelligence) to come into being.

As for what is needed, that is a harder question to answer. There are a number of theories on the topic. My personal theory is related to language. I think that we need to have an AI that has an outer voice and an inner voice, much the same way that we talk to ourselves to work through problems. I also think that we need a better test. The current Turing Test is either an all or nothing result. It does not tell us how far off the mark we are or in what direction we need to move.

Answer 3

In order to answer these questions, we first need to determine what human consciousness is. This is not as simple as it sounds, and leads to one of the more interesting debates in cosmology.

Stephen Hawking (and many other cosmologists and astrophysicists) believe that the universe is deterministic (or algorithmic) in nature. If this is the case, then it means that human consciousness by definition has to be algorithmic in nature because the universe cannot contain something that works beyond its own fundamental laws.

If this is the case, then YES; computers can have consciousness because computers are algorithmic in nature as well. All that would be required (under the mathematical concept of Formalism) is that the computer be running an algorithm of sufficient complexity. This is also known as Strong AI theory. In such a case, the Turing Test (as has been described in Daniel's answer) is the way to detect it. The Turing Test is essentially a 'double-blind' test; if you can't tell the difference between A and B, then A must equal B. (More on this later.)

The problem with this model is that if the universe is algorithmic in nature, then that means that if you know the algorithm, and you know the state of the universe at any given point in time, you can 'determine' the state of the universe at any other point in time, either forwards or backwards. Put another way, it means that free will (and consciousness for that matter) is an illusion and we have no power of choice; every decision we make is a result of the universe's original programming and it only feels like a choice to us because the algorithms in our brains are sufficiently complex that they have formed a localised consciousness.

On the other hand, there are mathematicians like Roger Penrose who claim that the universe is NOT algorithmic in nature, and that algorithms are merely a large subset of the mathematical constructs in the universe. Most importantly he states that the human mind itself is non-algorithmic and has to be in order for us to be sentient. In his book 'The Emperor's New Mind', he explains that this non-algorithmic element is something he describes as Insight.

In Penrose's universe, consciousness and sentience are a by-product of a non-algorithmic function which DO permit free will and allow us to make choices (and changes) to the universe that alter it in such a way that its future state CAN'T be determined from a previous state. This is good news for us because it means that free will we think we have is real.

It is however bad news for conscious computers. Because computers are algorithmic in nature, they could never be conscious, although they could appear to be so if one applied the Turing Test. This is where the Turing Test fails; it's a test of output, not process. Just because a computer can simulate human responses, it doesn't make it the same as a human, IE; Conscious.

What makes the Turing Test appear to work is something called anthropomorphisation, or the tendency for humans to see human like patterns in non-human objects. Anthropomorphisation is why we play with lego minifigs and dolls, etc. It's why we give our cars and GPS systems pet names. When something looks or responds to us in a way that feels human, we tend to attribute a human framework to it. In this case, a computer algorithm that can pass a Turing Test is more like a sophisticated doll.

So; we have a choice to make - either computers CAN be conscious and free will and our own consciousness is an illusion, OR computers CAN'T be conscious because free will is real but HAS to exist outside the bounds of algorithms.

A few caveats on all this.

First, I'm using the word consciousness in the conventional semantic sense as that's the way it's been used in this question. In the research I'm doing for my PhD in machine awareness, I'm building an ontology of terms related to awareness and in that consciousness refers to the 'always on' capacity to assimilate new sensory inputs and integrate the new learnings into an internally conceptualised view of the world. What we're calling consciousness in this answer is really sentience (or possibly liveness).

Second, even if a sentient computer program is possible, that wouldn't make it human. This answer is already long enough so I won't bore you with the neuro-physiology but in short, humans have instincts and emotions that sit below the reasoning centres of the brain. Computers would effectively be all cerebral cortex. This means that they can't have real instincts and emotions like humans even though they can simulate them. In short, a computer can't have a survival instinct (Skynet can't happen the way it's described), it couldn't be happy or sad, etc. Because of this, the idea of applying human rights to it would be invalid.

Finally, we need to be clear what 'intelligent' behaviour really is. I have a pick-up-sticks program that plays the game really well. On some configurations of starting count of sticks, number you're allow to pick up, etc. it's impossible to beat. This isn't because it's an AI, it's because the game is 'math-complete'. Once you understand the mathematical formula required to play the game, it's easy to program a computer to play the game 'perfectly'. Once you explain the mathematical formula to a human, they become just as good.

This last point is subtle but fundamental. We tend to see a response to something that we don't fully understand as 'smart', but this computer program doesn't play the game better than us because it's learned something we don't know, it plays better because it's following a number sequence that's programmed into it. Having access to data we don't may make it look smart, but all it's really doing is blindly following a set of instructions, like any algorithm. That means that we infer intelligence from our interactions with it. It doesn't imply that intelligence.

Long answer, but to summarise; computers can only be sentient if our own sense of free will is an illusion. This is not to say that computers can't simulate sentience, but the true test of sentience is not outputs (as per Turing) but process. A sentient algorithm MUST have the ability to conceptualise meaning as an internal structure, and build on that meaning via new information.