Quantum Computing

Papers etc. on quantum computing

1.) Quantum walks of correlated photons - O'Brien, J.L. et al - Quantum computing moves a step neaere

2.) Dancing the quantum dream - Paul Parsons - General description of quantum computing

3.) Rules for a complex quantum world - Michael Nielsen - Discussion of quantum information systems

4.) The fall of the machines ( or 'all you need to know about robots') - Michio Kaku - Problems with robotics and artificial intelligence

5.) The Singularity: A Philosophical Analysis - David Chalmers - Disappointingly weak discussion of AI

1.)

Quantum walks of correlated photons

O'Brien, J.L. et al

Science, 17 September 2010: Vol. 329 no. 5998, pp. 1500 – 03 DOI: 10: 1126/science 1193515

This paper published in 'Science' looks to bring quantum computing much closer to reality. Two entangled photons moving through circuits in a silicon chip have managed to perform a quantum walk. A quantum walk is analogous to the classical random walk and involves a superposition of possible positions, the outcome of which is the result of a calculation. Previous quantum walks have involved only one photon. The move to two photons from one photon is considered important. The photons are entangled and each increase in the number of photons entangled produces an exponential increase in the complexity of the problem that can be solved. Thus if a one photon walk can deal with ten possible outcomes, then a two-photon walk can deal with 100, and a three-photon walk with a thousand and so on. Having mastered the step from one to two photons the move up to multi-photon quantum walks is regarded as a relatively straightforward step. From a practical point of view a photon system is easier to handle than systems using trapped atoms or ions, because these latter require extreme conditions to remain in superposition.

The possibility of quantum computing has been around since the 1980s, but the technical difficulty of preventing decoherence of the particles or 'qubits' involved has sometimes made quantum computing look a bit like nuclear fusion, as a potential major leap forward, mired by insuperable technical problems. This latest photon study is claimed to bring practical quantum computing much closer. Before this 25 years looked like an optimistic estimate, given no apparent break through in the technology, now ten or even five years looks like a possibility.

The potential impact of quantum computing on our society and economy may well be underestimated, if only because of a culture of denial relative to the limitations of classical computers. A quantum computer is in effect the mother of all search engines, able to search through a much larger number of possibilities than will ever be possible for any classical computer. Diseases of cells such as cancer, Parkinsons, Alzheimers and MS might yield to quantum computing that could simulate processes in cells that are beyond the scope of classical computing. Further to this, problems that have stymied artificial intelligence and robots for more than half a century may relate to the inability to search through all the possibilities needed for perception within complex and ambiguous environments. If so the development of practical quantum computing may well lead on to the building of autonomous robots capable of carrying out day-to-day activities that have so far eluded robots. This could produce the most profound revolution in our social and economic structure since the advent of agricultural thousands of years ago, with robotics being applied to many areas of activity that are now exclusively human. This traumatic change was described in a popular science fiction book a few years ago. '...... it was the quantum brain that put an end to that. With its invention, robots became more practical than humans for most forms of labour. His school history books had recorded how the former work-based social and economic order had been overthrown in the course of a single turbulent generation. Now work was the privilege of the rich and those of the intellectual elite who sought to join their ranks. The unceasing hum of robotic factories, depots and data centres provided for everything else. Idle comfort and virtual reality entertainments had become the lot of the masses.'

Finally, what are the implications of this development for consciousness studies. In a general way the idea of quantum brains and quantum consciousness may become more acceptable. It has often been remarked that concepts of the brain are related to contemporary technology, with the Greeks likening it to a wax seal, the early twentieth century to a telephone exchange and the late twentieth century to a classical computer. The existence of quantum computers might make it much more normal to liken the brain to a quantum computer, something that is currently easy to dismiss as a silly fantasy. At a more practical level, quantum computing should allow the sort of examination of cells that would determine whether or not they involved quantum coherent functions. There's also a warning here for all those who score cheap points by ridiculing quantum consciousness, they could later be pulled out of the crowd as a 'how wrong they were quote' on the lines of 'no future for the horseless carriage'.

2.)

Dancing the quantum dream

Paul Parsons

New Scientist - 24th January 2004

The idea of quantum computing was originated in the 1980’s by the physicist, Richard Feynman, best known for the theory of quantum electrodynamics (QED). He realised that no conventional or classical computer could simulate the complexities of the universe. Some equations for relationships in the macroscopic or classical world and for some mathematical problems have to be solved by equations that also describe aspects of quantum physics. These equations can be of the type that mathematicians call NP-hard, NP standing for non-polynomial. In these, the number of computational steps needed grows exponentially with the number of variables involved, and it is possible to have calculations that would take longer than the life of the universe to solve.

Feynman’s suggestion was for a physical system, such as electron’s bouncing off one another, that would reflect the same mathematics as the problem that one wanted to solve. Furthermore, it turns out that once one can solve one NP-hard problem, one can solve a lot of others.

David Deutsch at Oxford University carried out some of the initial research on Feynman’s suggestion, and described how to encode a quantum computation that could solve an NP-hard problem, and in 1994 Peter Shore at Bell labs showed how quantum logic gates could find the factors of large numbers that underlie secure codes, such as those used by banks.

It has also been suggested that quantum computing underlies perception and decision taking by brains, and explains the inability of artificial intelligence systems based on classical computing to imitate the performance of organisms.

Quantum computing uses qbits, analogous to the bits in classical computing. In the experimental devices used to date, this has usually involved trapped ions or atoms. The machines require that the qbits remain isolated from the environment for long enough for useful calculations to be performed. Once the qbits interact with the environment, they decohere, and become useless for computing purposes. So far, despite substantial funding, practical quantum computing has not been achieved, and some critics have likened the technology’s problems to the failure to find a workable form of nuclear fusion.

3.)

Rules for a complex quantum world

Michael Nielsen

University of Queensland

Scientific American – November 2002

Nielsen’s article discusses the development of quantum information systems. He begins by pointing out that information has a physical basis, which applies to both classical and quantum types of information. With classical information the basic unit is the ‘bit’, which is either ‘0’ or ‘1’. With quantum information the basic unit is a ‘qubit’. The special qualities of qbits as opposed to bits is that they exist as a superposition, and also that they can be entangled with other qbits, so that they can be effected by a change in any of the other qbits in their group. Quantum computing is in principle much faster than classical computing, so that, for instance, a problem that would tie up the resources of a classical computer for 150,000 years might require the resources of a quantum computer for only a second.

Nielsen stresses that entanglement is not an all or nothing property. Some systems of particles have more entanglement than others, and entanglement can be measured as an information resource. The amount of entanglement in a system is defined as the number of copies of some fixed unit of entanglement.

The problem for the development of quantum computing is that the larger a quantum system becomes, the more likely it is to interact with the environment, and lose the quantum properties of superposition and entanglement. To achieve quantum behaviour in a complex system it is necessary to maintain isolation from the rest of the environment. It is relatively easy to isolate a single atom in a trap, but more complex arrangements become progressively more difficult. However, there are laboratory examples of larger but isolated quantum systems, as with superconductivity.

An individual qubit will give the same ‘0’ or ‘1’ answer as a classical bit, but it represents a probability weighting between the two possible outcomes, for instance a 30% probability of ‘1’ and a 70% probability of ‘0’. However, quantum computing becomes more interesting than this, when it involves multiple particles that are entangled in a group. Particles that are members of such a group do not have their own quantum states, only the group has a quantum state. A measurement on one particle instantaneously provides information about other particles.

However, the probabilistic nature of quantum mechanics makes it impossible to transmit. Thus when a series of quantum particles, that are entangled with another distant series are measured or read, the series that becomes available is completely random, and although the distant particles will show an identical or correlated series this will also be completely random, so the first observer will have no idea what an observer of the distant particles is going to see before they make their measurement. Therefore there is no faster than light transmission of conventional information and no violation of special relativity.

The concept of quantum error correction has become important in recent years. Classical information already has a system of error correction, so where one of three triplets of bits flips out of position, it can be corrected, and the message that the triplets represents is thus preserved. Previously it was thought that it was impossible to have error correction in quantum signals, and sceptics saw this as something that would prevent the development of most quantum computing and signalling. However Shor and Steane at Oxford University demonstrated how to achieve quantum error correction by passing qubits through the equivalent of logic gates without actually measuring or reading them, something which would cause decoherence. Penrose and Hameroff in discussing their Orch-OR model of quantum consciousness have suggested that the lattice structure of microtubules is suited to quantum error correction, and they argue that such error correction would help to shield their proposed quantum processing in microtubules.

4.)

The fall of machines (or 'all you ever need to know about robots)

Michio Kaku - Financial Times, 12th April 2008

In this article, Kaku discusses the shortcomings and disappointments of Artificial Intelligence (AI). He remarks that in contrast to more successful technologies, the physical laws that underpin it are not well understood. Despite over half a century of research and investment, AI has not mastered aspects of basic human/animal perception such as pattern recognition, nor the conclusions that we draw about the world under the heading of common sense.

Perception & common sense
The early days of AI in the mid 20^th century seemed promising, as computers achieved rapid strides in areas such as algebra and chess playing, with the most advanced machines beating grand masters at chess. So computers were better than humans at some of the things that humans regarded as difficult. The problems with AI arose when designers tried to get computers to do the things that humans found easy, or humans and animals, even quite primitive animals, did automatically.

It was found that robots performed very poorly at navigating complicated environments, much worse than insects such as fruit flies that appeared to have only a fraction of the computing power of some of the more sophisticated robots. Common sense involves judgements such as ‘water is wet’, ‘mothers are older than daughters’, ‘animals do not like pain’. The understanding of simple childrens’ stories requires a grasp of such common sense facts. They are obvious to humans, but have so far eluded the robot mind. Computers and robots depend on algorithms, systems for calculating, and no algorithms have been found to express the truths of common sense.

Top-down & bottom-up
Two possible approaches to robotics evolved over the years, the top-down approach and the bottom-up approach. The top down method involves trying to programme in, more or less one-by-one, all the things that the robot needs to know. These top-down robots have remained clumsy. Some commentators on AI have taken the view that the problems of perception and common sense are what mathematicians call non-polynomial (NP) hard. An algorithm for human perception or common sense could be written, but it would take an impracticably long time, possible longer than the life of the universe, for the calculation to reach a conclusion. It is thought possible that quantum computers, utilising the entanglement of numerous quantum particles as computing bits or qubits, could solve such NP hard problems, but the technology necessary for such computers is also proving difficult to master.

The bottom-up method involves the robot learning how to function by trial and error, literally by bumping into things. Bottom-up robots tend to use a form of computing called neural networks, which tries to mimic the human brain by performing a large number of calculations in parallel, and then bringing these together into a conclusion. The bottom-up approach had some success in mimicking the movements of non-flying insects, and this is essentially the technology used by the Mars landers. However, these robots have been unable to mimic the more complex movements of mammals. In respect of the bottom-up method, some commentators have claimed that the approach does not achieve unique solutions, and is therefore operating on a sort of guess work that may not be adequate for more complex activities.

The role of emotions
Kaku goes on to discuss the role of emotions, an aspect of the brain that was virtually ignored by neuroscience through most of the 20^th century. However, some researchers are now taking the view that emotion is the thing that distinguishes humans from robots. Emotion has come to be seen to be an important factor in evolution. Liking things or experiences is broadly a way of distinguishing the things that are good for us, and particularly so when dealing with the decisions that animals or hunter gatherers have to take.

Emotions are instantiated in regions of the brain known as the limbic system. This communicates directly with the prefrontal part of the brain, which is responsible for reasoning and decision taking. Where brain damage interrupts communication between the limbic system and the prefrontal, the patient retains the ability to reason, but may be unable to make even the most trivial decision. Thus it would appear that some emotional colouring is needed to clinch the human decision making process.

This leads on to a further problem. The electro-chemical processing performed by neurons in the limbic system is essentially the same as other processing in the parts of the brain that carry out mathematical calculations or those whose functioning is unconscious, and are thus directly analogous to silicon computer processing. But in contrast to these areas, the limbic system produces the subjective experience of our conscious emotions, of which there has never been any sign in computers and robots. This confronts us with the so-called ‘hard problem’ of consciousness of emotion or of other forms of experience. There is no agreement about what it is in the brain that gives us subjective experience of emotion, and therefore it is impossible to specify a computer that would produce these conscious emotions, that appear essential to decision taking.

Despite these problems the confidence of many in the AI industry remains undented, as it has always done in the unsuccessful past, and no doubt the industry will continue to convince governments to pour funds into this particular black hole.

5.)

The Singularity, a Philosophical Analysis

David Chalmers

Journal of Consciousness Studies, 17, No. 9-10, 2010, pp. 7-65

INTRODUCTION: David Chalmers wowed the first Tucson conference on consciousness, when he confronted people with the difficulties of incorporating consciousness into the scientific paradigm, coining the phrase 'the hard problem'. This makes this latest paper the more disappointing, with its tendency to subside into some of the sloppier ways of consciousness studies, and in the process evading the hard problem of qualia. Chalmers admits to the difficulties experienced in developing artificial intelligence/robotics (AI), and says that the problems encountered should lead to a reassessment of our beliefs on the subject. Despite this, he expresses confidence in reasonably near term success, but surprisingly without suggesting any means to bridge the gap between past failure and proposed future success. Later it is suggested that this successful AI will have the same type of desires and preferences as humans. This really poses the question of how emotional qualia which appear to at least in part drive preferences, arise and thus brings us back to that famous Tucson conference and the hard problem. But instead of giving us anything interesting, Chalmers shuffles the problem away by grouping emotion with the more obviously computerisable action of reasoning, as being personal properties, and leaves us there.

In this paper David Chalmers discusses the singularity, here meaning the point at which computers or robots become more intelligent than humans. Extrapolation of computer hardware processing capacity is not seen as a useful trend in this respect. The solution is seen as being in software. Chalmers argues that the main problem or bottleneck for artificial intelligence (AI) is to find the right algorithms, and that no researcher appears to have come close to this as yet.

Chalmers skates round the Penrose question of possible non-algorithmic processes in the brain, by appealing to other papers by himself that are claimed to show that Penrose's argument fails. Given the generous length and discursive style of this paper, a few pages on the non-algorythmic question might have been justified. However, Chalmers does say that if Penrose's version was correct, it should be possible to build a mechanism that replicated the Penrose-type quantum process. This was also admitted by Penrose.

What Chalmers does not say is that building such a quantum system is likely to be a more difficult and lengthy process than any classical computing route, which is all that has been available to AI to date. Practical quantum computers have not so far been developed, although there have been some promising reports of possible systems recently. The Penrose-style quantum computer would however be more sophisticated than currently planned quantum computers, because it requires computation to continue until the point of a hypothetical objective reduction of the quanta involved. However, Chalmers simplifies the situation at this point by saying that all he requires is a machine with intelligent behaviour, and that it need not necessarily have an internal subjective life.

Chalmers acknowledges that the development of AI has proved surprisingly difficult. The history of the subject has resounded with confident claims of success just round the corner. In the 1990s popular books were published predicting machine takeover of the planet in the early part of the first decade of this century, but the beginning of the second decade finds us as far as ever from having autonomous robots. Chalmers remarks that in terms of the technology and thinking of their time the confident claims were not obviously unreasonable, and this fact should lead us to a reassessment of our beliefs.

This makes it very surprising that two sentences later he says that he still distinctly favours the view that AI will eventually be possible. For a sophisticated philosopher, it is surprising that he does not provide some causal link between the past failure of AI and his confidence in future success, at least in terms of some proposed technical possibilities. He doesn't really attempt this very necessary explanation, instead jumping ahead to what would happen once AI was developed. Later in the paper, it is claimed that we have the capacity to emulate brains and should therefore be at the point of AI take off, but no reason is given for this optimism in the face of the still yawning gap between robots and even quite primitive organisms in terms of autonomous behaviour.

Later on Chalmers arbitrarily chooses to attribute desires and preferences to AI systems without any real build up as to why this should be so. The significant thing about desires and preferences is that they are at least partly driven by emotions that are qualia or subjective experiences. A sentence or two after this, Chalmers quickly elides emotional experience with reasoning, as person type activities. But this is unacceptable. No one would deny some form of reasoning in computers, as already exists at least in the type of computers that defeat chess grand masters. What is skipped over here is that the qualia of emotions link to preferences, in a way that they do not directly link to reasoning, and the two functions should not be shuffled together to look like one thing. What is disappointing is that Chalmers who wowed the first Tucson consciousness conference by pointing out the short comings of conventional thinking about consciousness has now apparently subsided into the sloppy ways of modern consciousness studies, where the core hard problem of qualia is simply skated over.

Later in the paper, Chalmers attempts a functionalist justification for assuming subjective consciousness in machines that are not essentially different from existing computers. The argument is that as we have no idea how consciousness arises in biological systems, therefore there is no reason why it should not arise in non-biological systems. This arguably falls foul of Occam's razor. We know that consciousness arises in biological brains, and under Occams razor we should stick with that explanation unless it becomes unworkable. With non-biologically arising consciousness we are complicating the explanation of consciousness with a new theory.

In so far as our understanding of biological systems describes something like a classical machine or mechanism, that description has proved to be no help with consciousness. The more reasonable assumption is that biological systems have something that our existing machines do not, and that this feature produces qualia. This is not meant to imply a dualist solution, but simply that the very complex construction of organisms may provide us with something not so far given us by neuroscience text books. This does not mean that conscious machines could not eventually be built, but it may require the copying of something in organisms that we have not yet understood, or at least not understood in relation to consciousness studies or AI.

var _gaq = _gaq || []; _gaq.push(['_setAccount', 'UA-17430256-2']); _gaq.push(['_trackPageview']); (function() { var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true; ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js'; var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s); })();