How can a machine be taught intellect? - Chapter 16

Artificial consciousness

Is there something special that allows people to talk, think, see or hear? Or are we simply machines? More and more is being known about the biology of perception, memory etc. However, suppose we would know everything about these mechanisms: would we then better understand consciousness? Or is there then an artificial form of consciousness that differs from the human form?

When asked if people are a machine, two answers are possible. We can start with biology and try to understand how neural systems work or we can make artificial systems and see to what extent they correspond to a human being. We will talk about artificial consciousness ( "AC" or " machine consciousness ' ,' MC '), which is closely related to artificial intelligence ( 'AI').

'Mind-like' machines

Descartes believed that the human body is a machine, but that this machine can not provide voice and rational thought on its own. Rational thinking could be the result of the workings of consciousness. However, a spirit is needed for this.

In 1642 Pascal built the first calculator, which was improved by Leibniz. Leibniz believed that matter consists of small 'wisps'. He disagreed with Descartes who claimed that the mind can influence the brain. Leibniz thought about the existence of thinking and observing machines. He presented a thought experiment in which he imagined that there would be a conscious machine that a man could fit into. If one would enter this machine, he could find parts that work on each other and he would not find parts that could explain the perception.

Boole believed that the human mind could eventually be explained by means of mathematical formulas. He has never been able to find these formulas. Yet thanks to him, algebra was used during the computer revolution. Turing argued that problems could be solved with algorithms. He spoke about a simple, abstract machine that could move. He believed that this simple machine could specify the steps needed to solve any problem. Such a machine is also called a Universal Turing Machine. An important part of this machine is that it can be used by inserting chips or brain cells. It is important that the machine carries out the same operations, which the machine also works on.

Craik believed that the human mind translates all aspects from the outside world into internal representations in the head. He stated that these internal representations can be manipulated according to fixed rules and that machines could do this too. According to him, consciousness is equal to mental representations.

Computers became more sophisticated and it was soon said that they were intelligent and could think. So it seemed possible that something like artificial intelligence existed. All attempts to design artificial intelligence consisted of programmers who told the machine what to do. These computers used algorit hms and processed information according to fixed rules. This is now called " GOFAI" ("Good Old Fashioned Artificial Intelligence") . A problem with GOFAI is that the information that computers process is seen as representations of things from the world, but that these symbols are not based on the real world.

In short: the computer itself does not know what it is doing; he is only performing. This is also called ' rule-and-symbol AI' . Searle distinguishes between ' strong AI' and ' weak AI' .

According to ' strong AI' , a computer that uses the right program is intelligent and has a mind as we have. Having a mind is therefore no longer a thing of using the right program. According to ' weak AI' computers can only imitate intellectual abilities (such as thinking and making decisions) . According to ' weak AI' they will never be really intelligent or really have a mind.

Connectionism

Connectionism is based on artificial neural networks ( ANNs ) and parallel processing. ANNs are used to mimic human human cells. The big difference with AI is that ANNs are not programmed, but are trained.

Imagine: a machine must recognize on photos whether a man or woman stands up. People can do this automatically, but a machine needs to be taught this. If we use ANN, the machine must look at a photo and output every time. If he makes a mistake, the neural network of the machine is adjusted so that he no longer makes the mistake. How does such a network work exactly?

The network consists of many units. These units are similar in effect to neurons. The units are connected in a network of parallel connections. A simple network can consist of three layers:

1. an input layer,    

2. a hidden layer, and

3. an output layer.

Connectionist networks do not just do what programmers say they have to do. This is something very different from the ' rule-and-symbol AI' . ANNs have many applications, for example reading manuscripts, controlling robots and as a spam filter.

Embodied cognition

The machines that have been described so far do not have a 'body'. They stand on shelves and they interact with the world. Previously it was believed that the essence of thinking is abstract and rules-based. Thinking could also be done without a body according to this vision.

Embodyed cognition stands for the idea that the mind can only be created if something interacts with the environment. The environment provides us with information storage and feedback that enable perception, intelligence and awareness. To construct machines in this way means working from bottom-up to top-down. This approach is sometimes called 'situated robotics' or 'behavior-based'. A consequence of such a work system is that intelligent behaviors can arise from simple systems, which creates the hope that the consciousness may also arise from the construction of simple systems.

Intelligence without representation

If you program robots so that they do not collide with things in their environment, it means that they do not have to create a representation of the environment in order not to damage themselves. The traditional form of AI assumes that intelligence goes hand in hand with mental representations of the outside world, but nowadays robots are being built that make no internal representations of the outside world at all. These robots can move forward without colliding with complex environments. They can also accomplish tasks such as cleaning up waste. These robots have different control layers that are all responsible for simple tasks. These layers are on top of each other and have limited connections. One layer can suppress the other. This is also called ' subsumption architecture' called. It seems that these robots perform intelligent actions. These findings are important for the discussion about consciousness.

We have the idea that conscious experiences go together with internal representations of the world. But how can a mental representation conscious experience be ? Perhaps it is not useful to think in terms of representations of the outside world, but this can lead to other problems. If we do not reason in terms of representations, what is the alternative?

The Turing test

Turing wanted to test a way to investigate whether machines could think. For example, he thought that it was possible to see if machines could play chess. He stated that if a machine can play chess, then he should be able to think. Turing also carried out a second imaginary investigation. He wondered whether a computer can have a conversation with a person. People were very skeptical about this idea and believed that there should be a trick if something like that would happen. Turing described ' the imitation game' . In this (imaginary) research there is a decision maker (C) who has to choose which of two people is a woman. The man (A) and the woman (B) are in another room, so that C can not see and hear them. C can only communicate with them by asking them the right questions and receiving typed responses. A and B both try to respond as women. It is therefore very important that C asks the right questions.

Turing wondered what would happen if A were played by a machine. Can a male machine behave like a woman? So can machines think? Turing predicted in 1950 that by 2000 it would be possible to program computers in such a way that a questioner would only have a 70 per cent chance of discovering whether a person or a computer gives the answers. Turing was not far wrong. In an experiment in 2008, people were given five minutes to communicate with a computer and with a person. Three of the twelve people thought that the computer was a human being.

Resources:Blackmore; Susan. (2010). Consciousness, Second Edition An Introduction. Abingdon, Oxon: Taylor & Francis.