At present computers can translate most documents from one language into another in such a way that a person who reads the document in the language into which it has been translated can at least get the gist of what is being said. However the use of such translations is limited, and the criterion for a successful translation is that it be useful. Someone relying on such a translation of a legal document, for example, would be very likely to end up in trouble as the computer would produce a translation which missed or altered many of the finer points produced by subtle wording that make or break legal transactions. In a completely different situation we would doubtless end up extremely frustrated by a computer translation of a work of fiction or of a poem. Though we might form a reasonable idea of what was going on in the book or poem we would lose almost all of the subtleties of the original authors use of language. Descriptive passages, jokes, accents and dialects, the general atmosphere of the book and attitudes of characters not explicitly expressed would all vanish. It is questionable whether the original narrative could actually survive such a process in many cases but even if it could the best part of the book would be gone. Reading a detailed description of the plot is no substitute for reading a book. It is reasonable to expect that computer translations of other kinds of documents, philosophy, instruction manuals, literary criticism, political manifestos etc, will be inadequate for similar reasons. However, it seems there is a generally held belief that it is only a matter of time before computers can do translations as well as if not better than humans. Why not? Over time computers have surpassed human beings in many areas. Many people believe that computers will eventually be better than human beings at everything, so why should translation be any different?

I am not setting out here to save translators jobs or to assert the eventual superiority of technology. I hope only to show that once we have defined our terms we will be able to see that the debate about whether computers will be able to translate as well as human beings actually hinges on a more fundamental debate in philosophy of mind about the role of our actual experience as human beings in explaining our actions.

Before we start it would be good to say a little about translation. At the most basic level translation is the process whereby we seek to take what is expressed in one language and express it in a different language. There are two general approaches that we can take to this process. First, we can aim for ‘formal’ equivalence, seeking to convey the exact message of the original text in terms of form and content. Linguistic features of the original such as vocabulary and syntax are relayed as faithfully as possible. Conversely we might aim for dynamic or ‘functional’ equivalence, aiming to produce the same semantic effect on the target language readers as was produced by the original text on source language readers. By ‘semantic’ I mean actual experience of meaning and so if two audiences experience the ‘same semantic effect’ I mean roughly the same experience of meaning. It is worth noting that whether or not formal equivalence has been achieved depends on whether the correct rules determining corresponding sentence structure and vocabulary have been followed and so we judge success here by which process has been carried out and not by the end result. On the other hand we judge the success of a dynamic translation by its product. If it produces the same semantic effect in its target language readers as in its source language readers it is a good dynamic translation and if it does not it is not, irrespective of the processes actually used to create the translation.

It is a little more difficult to say what I mean when I talk about computers, and how I distinguish computers from human beings. Computers are often defined as input output systems. Taken as it stands, this definition does not rule out anything from qualifying as a computer. A stone is a computer. I input throwing, it outputs shattering or a ‘thunk’ noise. Something similar will be true for any inorganic substance. Plants animals and human beings are computers capable of returning various outputs in response to a range of inputs. Even abstract conceptual things like numbers can be thought of as input output systems. A concept so general that it refers to everything that does or could possibly exist is of no use at all and we certainly do not mean the word ‘computer’ to refer to everything that is or could be in general conversation, so we shall need to restrict our definition somewhat. I do not wish to propose a complete set of necessary and sufficient conditions for being a computer here. It will suite our purposes if we can come up with a couple of the necessary conditions for being a computer in such a way that we stand a chance of distinguishing some things from computers in virtue of the fact that they do not possess these features. Hopefully it will not be controversial to say that if a thing, x, is a computer then x is a thing whose outputs in response to inputs is determined by a set of rules that can operate independently of any semantic interpretation.

On what grounds to I insert the clause that the rules must be able to operate indipendantly of any semantic interpretation? First off whatever else is a necessary condition of being a computer, semantic capability is not. All the things we currently call computers do not have semantic abilities so far as we know, and so having semantic capability is at least not a necessary condition of being a computer. I go further than this however in saying that computers must be able to carry out their operations uninfluenced by any semantic ability they may have. My reasons for making this statement are conceptual: we define one set of things that are not computers as things that necessarily have semantic ability as part of their make up. If something does not have semantic ability it is not a part of that set of things. This set of things can be thought of as the group of things that are ‘sentient beings’ as opposed to automatons (for now we can think of all automatons as computers). If a thing is to be called a computer the process, namely the set of rules that determines outputs in response to inputs, must be able to remain the same were the semantic capability to be taken away, otherwise we will call it a sentient being instead. This definition allows for the possibility of a conscious computer. As I take human beings to be part of the set of sentient beings I can now safely assume that human beings are not computers.

The meaning of the statement that if x is a computer then x is a thing whose outputs are determined by a set of rules depends also on what we take ‘rules’ to mean. There is much debate on this matter so I do not wish to say anything to definite or open to objection if I can help it. It is probably possible for anything to act as a rule or set of rules. I think it will be enough for our purposes if we think of a rule in a functional way: something is a rule or set of rules at a given time if it functions as such at that time. We have already said what it is to function as a set of rules. A rule or set of rules is something that exhaustively determines the outputs of a given thing in response to inputs. Nothing so far has been said that implies that computers are deterministic systems. A set of rules that explains and predicts the outputs of a thing in response to inputs can do so exhaustively, so far as I am concerned, even if its explanations and predictions are irreducibly probabilistic. Thus in addition to allowing that computers may have semantic capability I am also admitting that they may well not be deterministic systems.

A system defined in the way that I have defined computers will be able to manage formal equivalence translation just as well if not far better than a human being. ‘A thing whose outputs in response to inputs can be exhaustively explained and predicted by a set of rules’, something that carries out a process where a result is obtained from a starting point via a set of rules, is bound to have no problem achieving success in an activity where the criterion for success are that the right rules be followed. It will only be a matter of ensuring that the thing is governed by just the right set of rules, and then by definition it will be carrying out perfect equivalent translation. We can see this merely by thinking about the very limited definitions of the concepts we have come up with so far.

However it is not immediately clear just from looking at the concepts involved that such a computer will be able to achieve successful functional equivalence translation where success is judged by the semantic affect of the process, not the process itself. A computer can operate according to any set of rules which do not depend on semantic interpretation for their make up. I do not want to suggest that sets of rules operating independently of semantic interpretation cannot produce any possible semantic result, However, it is not analytically true that they can either. There is no reason to believe that there are no results that cannot be achieved by semantic (or ‘sentient beings’) systems alone (as there is no reason to assume that there are not results that cannot be brought about by a system that depends on semantics for the nature of its operation). It would not be surprising if some of these results that non semantic systems (computers) might not be able to bring about were semantic.

And so you see that I really have not made a contribution to either side of the debate about whether computers will ever be able to translate as well as human beings. Instead I hope to have clarified the debate a little and shown that there is no analytic reason to suppose that computers will be able to bring about all the results that ‘sentient beings’ such as human beings can. Of course there is no analytic reason to suppose that they cannot either. This is evidence that the debate is empirical rather than philosophical, but it is not proof. Someone may be able to develop the concept of ‘computer’, ‘sentient beings’ or even ‘human being’ more fully until some conceptual reason emerges to think that there are things that people can do and computers cannot, or that there is nothing a human can do that a computer can’t. I have a hunch however that it will prove extremely difficult to come up with a more fleshed out concept of ‘computer’ that does not prove controversial (indeed I’m not sure that my extremely limited definition will not ruffle some feathers). Likewise it is a certainty that any reasonably developed definition of sentient or human beings will probably prove unacceptable to some. In the meantime it is no small gain that we see that the question of whether computers will ever be able to translate or do many things human beings can do will depend on the result of the debate about whether semantic systems can do things that non semantic ones cannot. There is a strong tendency at the moment to talk about human beings, not necessarily as if they were computers, but as if everything they do and say can be explained in the same way that we explain the workings of a computer. Clearly if what I have been saying is correct then this is only an assumption. There is a related tendency to think that computers can in theory if not in practice eventually achieve anything that is logically and physically possible. This too is an assumption and not a position that arises out of an understanding of what computers are. It is also interesting to consider the implications that this might have for the problems of solipsism. It is possible that if there are no other beings in the world but myself non conscious things might still be able to act in a way indistinguishable from other people. However, it is also possible that they might not. If empirical or philosophical evidence were ever mustered to show that there are certain things semantic things can do that computers cannot, and if we know that there are entities out there doing these things I shall at least know that there are other sentient beings in the world beside myself, if not other people. Because of its bearings on these and other questions it seems well worth continuing to investigate empirically and conceptually whether there are things computers cannot do that humans can.

Is this look-up table method really translating, and if not, what do we want from a system before we are willing to say it is really translating?

In the remainder of this article I will try to show the following:
•    Translation is an activity that we identify by its function. We say that something is translating when it produces translations, regardless of the process used to obtain them.
•    When we ask ‘is this computer really translating?’ what we actually should be asking is ‘is it really the computer that is doing the translating?’
•    The answer to this last question is not black and white.

Before I start I want to make a useful distinction:
•    I will use the term ‘function’ when we are only concerned with the input and output of a procedure, that is, we are not concerned with the method or any (possibly semantic) by-products.
•    When I want to talk about a procedure as a whole, that is, when I care about more than just the input and output, I will use the term ‘process’.

For example, performing the same calculation in your head, with a calculator and on a computer could all be said to be the same function yet different processes. Note that this divide is purely pragmatic, intended to be useful rather than in any way strictly defined or objective.

My entire argument of course depends on what exactly we mean by ‘real translation’. My opinion is that, although we could think of endless sets of requirements for a translation to be ‘real’, there are only actually two types of requirement. That is, all possible requirements would fall into one or the other of the following categories:

1.    We can claim that real translation must involve semantics; the translator must translate using some idea of meaning, rather than simply crunching symbols. (A qualitative requirement on the translation method).
2.    We can set some quantitative requirements for real translation, that is, we could choose some testable selection of requirements. For example: Translation isn’t real unless the program has only been trained on X documents; Translation is only real if less than Y linguistic rules were hardwired into the translating device, etc. (A quantitative requirement on the translation method).

I intend to show that neither type of requirement enables us to define ‘real translation’, and we should instead approach the problem in a different way.

Translation is effectively a function. Certainly human translators experience translation as a process, but when we ask ‘Can computers translate as well as humans?’ we are not asking ‘Can a syntactic system feel like a human does when translating?’, but rather, ‘Can a syntactic system produce outputs from inputs at the standard a human can?’. More precisely, we are asking: ‘Can a computer mimic the human translation function?’ not ‘Can a computer mimic the human translation process?’.

As the input and output of the translation function are both totally syntactic (ie, they are just some blocks of text) , any semantic involvement must necessarily be in the method (that is, in something other than the input or output). It is a result then that whether a machine was translating using some semantic method would only be of importance if we wanted to create machines to emulate the translation process. We can conclude then: It may well be the case that human level translation must have some semantic element in order to function, but it can not  be a requirement. All translation, is in fact ‘real’, and we should instead be concerned only with what exactly is performing the translation. That is, we shouldn’t be asking ‘can machines do real translation?’, but, as all translation is real, we should just be asking, ‘Is it really the machine doing the translation?’

When considering the look-up table method for performing human level translation totally syntactically, it would be easy to wonder exactly when the inputs are really taking place. Is the text to be translated being input when we hand it to the final program and ask it for a translation, or is the input in fact happening when the required translation is initially created (perhaps by a human) for insertion into the look-up table? If we decide that the input is occurring only when we give the finished look-up table program a document to translate, then we are done; between the input and the output only the machine is involved, and as argued above the translation is certainly real. The machine is translating.

What if we decide the input is really at the start, at the point where the look-up table is originally constructed? Personally I think this is reasonable. Then it seems that it’s not the computer that is doing the translation, but its creator, as in order to make the look-up table they must have produced all the original texts and all the translations. I want to argue though, that this is only an issue with this particular method, rather than with machine translation as a whole.

Consider a similar system to the look-up table, but this time, rather than having the entire look-up table in it’s memory, it instead has a very large look-up table, and some basic ability to extrapolate. Imagine for example that it doesn’t have the sentence ‘Today I feel really strong’ and its corresponding translation in its knowledge base, but it does have ‘Today you feel really strong’, ‘Today I feel really weak’, Today I feel strong’, ‘Today you feel really weak’ and ‘I feel really strong’. It seems quite reasonable to think that the computer would be able to use this information along with some basic reasoning to translate ‘Today I feel really strong’. Now ‘Today I feel really strong’ has for the first time been input when we hand the final program the sentence and ask it to translate it. It was never previously translated and put into its knowledge base. So would we say that the computer is translating this sentence? Or would we argue that the input, in some convoluted form, had in fact been made earlier? We are again at a subjective wall; we seem to yet again need to agree on some method for classifying. This time, when the computer is doing the translation, and when it is not.

All translation is real, we ask only what is doing the translation. This seemingly can’t be dealt with objectively, yet it would appear difficult for even the most zealous of those in opposition to the possibility of computer translation to come up with some criterion of their own that made it theoretically impossible for a machine to ever be doing the translation itself. Sometimes it might be cheating, but I don’t think anyone can argue that it always is.

]]> http://notthatcomplex.wordpress.com/2009/09/11/when-is-it-cheating-submitted-by-t-j/feed/ 1 notthatcomplex http://notthatcomplex.wordpress.com/2009/09/08/could-god-turn-out-to-be-an-alien/ http://notthatcomplex.wordpress.com/2009/09/08/could-god-turn-out-to-be-an-alien/#comments Tue, 08 Sep 2009 13:03:50 +0000 notthatcomplex http://notthatcomplex.wordpress.com/?p=43 ]]> My guess is that most religious people would say no. They would probably say that it’s ridiculous to suppose such a thing and that the mere asking of such a question belies a complete lack of understanding of theism. I’m interested in just what that misunderstanding is. I suspect that most non religious people would also think it absurd to ask if God might be an alien, but it seems to me that it is harder for us non believers to put our finger on exactly why this is. With my best Richard Dawkins hat on I’m going to run through the possible conceptual differences between God and an incredibly powerful and knowledgeable alien then talk briefly about why they don’t seem convincing. Then I’m going to speculate, as far as an atheist can, just what it is that must be included in the concept of God if he is to be distinguished from an incredibly powerful natural force. The real point of this note however is to invite people who do believe in God to reply and try to explain how they understand the difference between God and an incredibly powerful natural being. I want an explanation of the religious persons concept of God to juxtapose with the atheist concept.

You might think that it’s ridiculous to compare God to an alien, but wait, you’ve not heard about the alien yet, and this is one special alien. Call him Fred. There could surely be beings who know a lot more than us right? Well Fred does. In fact Fred knows everything. We call people like Fred ‘omniscient’, similar to God who is also omniscient this respect. I don’t see how anyone could object to the possibility of an omniscient being who was not God. Perhaps there are a finite number of facts about the world to be known (though I doubt this). There are lots of natural beings who know some of these facts. A natural being could know more and more until, like Fred and God, he knew them all. Or perhaps there are an infinite number of facts. A being who had this infinity of facts would certainly be qualitatively different from any being we have ever known, but still, I see no conceptual reason that such a being should not be discovered. At least, if it’s logically possible that God should be omniscient it must be logically possible that Fred is omniscient.

Another thing about Fred. Perhaps by dint of his natural powers or perhaps because of his incredibly advanced technology he is what we call omnipotent. So powerful there is nothing he can’t do, ‘infinitely powerful’ if you like, same as God. It may be unlikely that such a being as Fred exists but if God can be omnipotent it must be logically possible.

It might be objected that there is more to God than this. God controls the destiny of men, sends people to Heaven and Hell, is the father of Jesus, sits on a throne surrounded by angels, etc. it might seem superficial to make these things the distinction between Fred and God, but perhaps people who believe every last word of the bible to be the literal truth might be able to make sure they don’t make any mistakes when they refer to God by saying something like ‘by God I mean the being who has done all the things that the bible relates to us that God did (and so I certainly won’t accidentally mean Fred the alien)’. This is a stronger point if we beleive that ‘meanings are not in the head’ or at leas that the meaning of a term is largely determined or affected by its reference. But wait a second, if Fred is infinitely powerful he can control the destinies of men, create a heaven and a hell then send them to the first if they behave and the latter if they do not, be the father of the man called Jesus (or perhaps if omniscient beings cannot interfere with the definitions of concepts and so Jesus can only have one father, that spot being taken by the real God, just a man identical to Jesus who is also called Jesus). He could even sit on a throne surrounded by angels and have a white beard if he chose. This is what it is to be omnipotent.

More seriously, it might be objected that Fred is a natural being and not God because Fred is ‘within this world’, within space and time, while God is traditionally thought to be in some way outside it. However I see no reason why Freds omnipotence shouldn’t allow him to go outside of space and time. If there were something like this that he couldn’t do he wouldn’t be omnipotent because being omnipotent means being able to do anything.

So couldn’t it just turn out that Fred is God? If we met Fred, or some earthly manifestation of him, and we learned that he had all these attributes, wouldn’t it be sensible to think that we had made contact with God? If I know that a person has prominent characteristics xyz and then I meet someone who is xyz I normally assume that they are that person. This at any rate is the way that scientologists seem to have reasoned. But anyway, I think that most religious people would still say that it is absurd, perhaps even blasphemous, to say that Fred is God. As an atheist however, I think that I have run out of ways of making a conceptual distinction between an omniscient and omnipotent alien and God.

Lacking a real understanding of religion it is unlikely that I will be able to say much about what I suspect a real religious understanding of God might be like. Hopefully people in a better position to know will give a better account in replies to this article. For now all I can say is that it seems to me that it is only possible to construct a concept of God that is not essentially a concept of an omnipotent, omniscient being if you take a fundamentally different world view to that taken by your average atheist. This world view will probably include a wealth of new concepts which are themselves very hard to understand from outside the system (perhaps there are some equivalent atheist concepts that are hard for theists to understand).

This is no longer the case. Nowadays we can be acquainted with almost anyone through the media, and the media tends to acquaint us with those people who are the best performers of various activities. That is to say the best performers in the world; the media has no interest in relaying stories to us about the most intelligent person or best runner in the neighborhood. With the pool of acquaintances widened and selected in this way it is no longer statistically likely that there will be any activity at which we excel all of our acquaintances. In fact it is incredibly unlikely – there is now only one top spot available in each activity where before there were very many. This, coupled with the fact that it matters to human beings to be the best at things, has resulted in an increasing division of labour in areas where an increase in the division of labour cannot be explained by the pursuit of more efficient profit creation. There is an incredible level of specialization in hobbies and sports, for example. People don’t just collect stamps, they focus on a specific style of stamp from a specific era. Cyclists don’t just enter races, they focus on 5 mile time trials or endurance rides to the exclusion of everything else (in the Tour De France some riders just train for riding up hill, try to win the mountain stages then drop out as they have no chance of doing well on the later flat stages). Intellectuals are no longer interested in everything, being confined first to their subjects, then to increasingly narrow areas of expertise within these subjects. Of course I am not saying that increased awareness of competition is the only reason for specialization in these areas, but it is a significant factor.

As a result of this state of affairs we have a hitherto unknown contradiction for the individual in today’s society. In an age that increasingly emphasizes the value of being the best by bombarding us with images of people who have achieved the highest possible level of attainment is has become massively more difficult to actually be the best.

Some people might suggest that this is only going to be a problem for ambitious people. The majority of people never aspired to be the best at any of the things they do and so they are not bothered by the fact that it is now nearly impossible to achieve this. It is true, sensible people will probably see that it is not worth trying to be the best for just the reasons outlined above, but it was not always a rash idea to aspire to make being the most able at some activity part of ones identity because this used to be a much more achievable ambition. Indeed I think that there probably used to be so many ‘top spots’ to go round that people did not need to aspire to them and work towards them with the devotion that we now associate with those who take it all the way to the top. A little talent and interest was probably sufficient. You didn’t have to be an extraordinary person and it really was true that ‘everyone is good at something’, that people would most probably find the thing they felt they meant to do in life. Moreover, finding out what you were good at was probably a very valuable thing. In fact I suspect that it was valuable in a way that few things are these days and that we do not know what we are missing. It would have been a source of identity and pride, and a source of comfort too in the face of failures in other areas. This is just speculation though.

So I am saying that it is better to be a big fish in a small pond, that the pond has got much larger, and that this is bad. This is probably not a very original or a very profound observation and I am sure that someone will have made it before in much more detail and with much more evidence. I am only making it here because I think that it may set the stage for one of those rare occurrences where a philosophical idea will actually be useful and help us to solve a problem if we can manage to make it generally accepted in our culture.

The harm that is done by the state of affairs described above is only possible while we subscribe to the idea that more of anything is always good, or at least utility producing. This is a background assumption of classical economics, and it is a good assumptionwhen it comes to predicting peoples behavior because most poeple act as if it were true. More material possessions, more money, more power, and more ability. Not ‘a lot of’ money or ‘a great deal of’ power. There is no objective amount of anything that it will be good to have, all that we know is that it is better to have more than other people have, or more than we used to have. This idea that more is always better compels us always to look at how well other people are doing when we want to see how well we are doing. There is no where else to look, no other measure. However, if we were first to do away with this idea (perhaps not in all areas), then to replace it with something like an idea of ‘good enough’, this might go some way towards removing the problem. This does not mean that there needs to be a simple binary good enough/not good enough approach to things. All that we really need is some kind of objective scale against which to measure ourselves. To put it another way, we need to be able to talk meaningfully about being good, bad or all right at something without needing to take into account how good other people are at it. If people can really accept this way of talking they need not be bothered by information about individuals who outperform them by a considerable measure. Just because someone else is brilliant at something it will not mean that they are no longer good at it.

I am not setting out here to save translators jobs or to assert the eventual superiority of technology. I hope only to show that once we have defined our terms we will be able to see that the debate about whether computers will be able to translate as well as human beings actually hinges on a more fundamental debate in philosophy of mind about the role of our actual experience as human beings in explaining our actions.

Before we start it would be good to say a little about translation. At the most basic level translation is the process whereby we seek to take what is expressed in one language and express it in a different language. There are two general approaches that we can take to this process. First, we can aim for ‘formal’ equivalence, seeking to convey the exact message of the original text in terms of form and content. Linguistic features of the original such as vocabulary and syntax are relayed as faithfully as possible. Conversely we might aim for dynamic or ‘functional’ equivalence, aiming to produce the same semantic effect on the target language readers as was produced by the original text on source language readers. By ‘semantic’ I mean actual experience of meaning and so if two audiences experience the ‘same semantic effect’ I mean roughly the same experience of meaning. It is worth noting that whether or not formal equivalence has been achieved depends on whether the correct rules determining corresponding sentence structure and vocabulary have been followed and so we judge success here by which process has been carried out and not by the end result. On the other hand we judge the success of a dynamic translation by its product. If it produces the same semantic effect in its target language readers as in its source language readers it is a good dynamic translation and if it does not it is not, irrespective of the processes actually used to create the translation.

It is a little more difficult to say what I mean when I talk about computers, and how I distinguish computers from human beings. Computers are often defined as input output systems. Taken as it stands, this definition does not rule out anything from qualifying as a computer. A stone is a computer. I input throwing, it outputs shattering or a ‘thunk’ noise. Something similar will be true for any inorganic substance. Plants animals and human beings are computers capable of returning various outputs in response to a range of inputs. Even abstract conceptual things like numbers can be thought of as input output systems. A concept so general that it refers to everything that does or could possibly exist is of no use at all and we certainly do not mean the word ‘computer’ to refer to everything that is or could be in general conversation, so we shall need to restrict our definition somewhat. I do not wish to propose a complete set of necessary and sufficient conditions for being a computer here. It will suite our purposes if we can come up with a couple of the necessary conditions for being a computer in such a way that we stand a chance of distinguishing some things from computers in virtue of the fact that they do not possess these features. Hopefully it will not be controversial to say that if a thing, x, is a computer then x is a thing whose outputs in response to inputs is determined by a set of rules that can operate independently of any semantic interpretation.

On what grounds to I insert the clause that the rules must be able to operate indipendantly of any semantic interpretation? First off whatever else is a necessary condition of being a computer, semantic capability is not. All the things we currently call computers do not have semantic abilities so far as we know, and so having semantic capability is at least not a necessary condition of being a computer. I go further than this however in saying that computers must be able to carry out their operations uninfluenced by any semantic ability they may have. My reasons for making this statement are conceptual: we define one set of things that are not computers as things that necessarily have semantic ability as part of their make up. If something does not necessarily have semantic ability it is not a part of that set of things. This set of things can be thought of as the group of things that are ‘sentient beings’ as opposed to automatons (for now we can think of all automatons as computers). If a thing is to be called a computer the process, namely the set of rules that determines outputs in response to inputs, must be able to remain the same were the semantic capability to be taken away, otherwise we will call it a sentient being instead. This definition allows for the possibility of a conscious computer so long as consciousness is not essential to the way the computer works. As I take human beings to be part of the set of sentient beings I can now safely assume that human beings are not computers.

The meaning of the statement that if x is a computer then x is a thing whose outputs are determined by a set of rules depends also on what we take ‘rules’ to mean. There is much debate on this matter so I do not wish to say anything to definite or open to objection if I can help it. It is probably possible for anything to act as a rule or set of rules. I think it will be enough for our purposes if we think of a rule in a functional way: something is a rule or set of rules at a given time if it functions as such at that time. We have already said what it is to function as a set of rules. A rule or set of rules is something that exhaustively determines the outputs of a given thing in response to inputs. Nothing so far has been said that implies that computers are deterministic systems. A set of rules that explains and predicts the outputs of a thing in response to inputs can do so exhaustively, so far as I am concerned, even if its explanations and predictions are irreducibly probabilistic. Thus in addition to allowing that computers may have semantic capability I am also admitting that they may well not be deterministic systems.

A system defined in the way that I have defined computers will be able to manage formal equivalence translation just as well if not far better than a human being. ‘A thing whose outputs in response to inputs can be exhaustively explained and predicted by a set of rules’, something that carries out a process where a result is obtained from a starting point via a set of rules, is bound to have no problem achieving success in an activity where the criterion for success are that the right rules be followed. It will only be a matter of ensuring that the thing is governed by just the right set of rules, and then by definition it will be carrying out perfect equivalent translation. We can see this merely by thinking about the very limited definitions of the concepts we have come up with so far.

However it is not immediately clear just from looking at the concepts involved that such a computer will be able to achieve successful functional equivalence translation where success is judged by the semantic affect of the process, not the process itself. A computer can operate according to any set of rules which do not depend on semantic interpretation for their make up. I do not want to suggest that sets of rules operating independently of semantic interpretation cannot produce any possible semantic result, However, it is not analytically true that they can either. There is no reason to believe that there are no results that cannot be achieved by semantic (or ‘sentient beings’) systems alone (as there is no reason to assume that there are not results that cannot be brought about by a system that depends on semantics for the nature of its operation). It would not be surprising if some of these results that non semantic systems (computers) might not be able to bring about were semantic.

And so you see that I really have not made a contribution to either side of the debate about whether computers will ever be able to translate as well as human beings. Instead I hope to have clarified the debate a little and shown that there is no analytic reason to suppose that computers will be able to bring about all the results that ‘sentient beings’ such as human beings can. Of course there is no analytic reason to suppose that they cannot either. This is evidence that the debate is empirical rather than philosophical, but it is not proof. Someone may be able to develop the concept of ‘computer’, ‘sentient beings’ or even ‘human being’ more fully until some conceptual reason emerges to think that there are things that people can do and computers cannot, or that there is nothing a human can do that a computer can’t. I have a hunch however that it will prove extremely difficult to come up with a more fleshed out concept of ‘computer’ that does not prove controversial (indeed I’m not sure that my extremely limited definition will not ruffle some feathers). Likewise it is a certainty that any reasonably developed definition of sentient or human beings will probably prove unacceptable to some. In the meantime it is no small gain that we see that the question of whether computers will ever be able to translate or do many things human beings can do will depend on the result of the debate about whether semantic systems can do things that non semantic ones cannot. There is a strong tendency at the moment to talk about human beings, not necessarily as if they were computers, but as if everything they do and say can be explained in the same way that we explain the workings of a computer. Clearly if what I have been saying is correct then this is only an assumption. There is a related tendency to think that computers can in theory if not in practice eventually achieve anything that is logically and physically possible. This too is an assumption and not a position that arises out of an understanding of what computers are. It is also interesting to consider the implications that this might have for the problems of solipsism. It is possible that if there are no other beings in the world but myself non conscious things might still be able to act in a way indistinguishable from other people. However, it is also possible that they might not. If empirical or philosophical evidence were ever mustered to show that there are certain things semantic things can do that computers cannot, and if we know that there are entities out there doing these things I shall at least know that there are other sentient beings in the world beside myself, if not other people. Because of its bearings on these and other questions it seems well worth continuing to investigate empirically and conceptually whether there are things computers cannot do that humans can.

According to the ethical naturalism of Philippa Foot and Michael Thompson, the word “good” applies to humans in the much the same sense as it applies to animals and plants. All animal species need to behave in certain ways in order to survive and flourish. For instance, bees need to sting creatures who threaten their hives. A bee that stings is a good bee; it is behaving as bees should. Humans need the virtues in much the same way. For instance, humans need to keep promises, or society will fall apart; an individual who breaks promises is defective in the same way as a bee that doesn’t sting. (Foot adds that human action is unique in being voluntary action; goodness and badness in humans are therefore goodness and badness of the will. But this modifies the scheme of natural normativity; it does not fundamentally alter it.)

One discomfort we might feel with this line of thought can be expressed in the following way. “If we know what is good for trees or tigers, we also know what is good for this tree or this tiger. (It would be eccentric to say, ‘Yes, I know strong roots are good for oaks in general, but not for this oak, because this oak is special.’) It is surely otherwise for human beings. My good cannot simply be ‘read off’ what is good for man in general. It is something that I have to work out for myself, taking into account the specifics of my personality and situation.”

This peculiarity of human beings can be captured by saying that we have something called “personality” – something that other living creatures lack. The fact of personality seems to imply the falsity of ethical naturalism. If there is no direct inference from “x is good for man” to “x is good for me” then how can reflection on human nature in general have any relevance for how I should live? However, I think this is too swift a conclusion. I want to argue for a form of ethical naturalism that accommodates the fact of personality.

First, let me give a more precise definition of what I mean by personality. I have a personality insofar as what is good for me is distinct from what is good for my kind. Only the former, not the latter, gives me a reason for action. The ethical demand must address itself directly to me; it does not just apply to me as a member of homo sapiens. “Good for me” is to be interpreted in the broadest sense. It refers to all that is fulfilling or perfective of me. It is not equivalent to “what I desire.” Something can be good for me even if I don’t desire it, although in coming to see that it is good for me I will also come to desire it.

It is a contingent fact about human beings that we have personality in this sense. We can imagine a species X which, although rational, does not have personality. An individual X would draw no distinction between “good for me” and “good for the X.” To say, “φ-ing is good for the X” would be to give any individual X a reason for φ-ing.

The film “The Invasion” depicts a species of this kind. Certain humans are infected with an extra-terrestrial virus, which leaves their physical appearance and intellectual faculties in tact but completely wipes out their personality. They are in effect zombies. Each individual’s interests are at one with the interests of the virus; his overriding goal is too spread it as widely as possible. In one scene, the still uninfected heroine, played by Nicole Kidman, is driving to safety, from where she will be able to create a vaccine. Realising the necessity of stopping her, the zombies throw themselves against her car, killing themselves in the process. They are like bees stinging themselves to death.

Notice that these zombies would use the words “good” and “bad” just as we use it of trees and animals – to mark natural strengths and defects. A zombie who didn’t sacrifice himself when required would be defective in exactly the same way as a bee that didn’t sting an intruder to the hive. He would need to be cured – or, failing that, put down.

Ethical naturalism of Foot’s sort applies very well to creatures of this kind – which is to say that it doesn’t apply to human beings. But can ethical naturalism be modified to overcome this defect? I believe it can. The solution is to be found in Aristotle. Aristotle (unlike Foot) takes account of what I’ve been calling the fact of personality. Every human, claims Aristotle, acts for the sake of his own good as he perceives it. The ethical demand must connect with my projects; it does not take me in automatically as a member of the human species. But – and this is crucial – “my good as I perceive it” and “my good” are not the same thing. I can come to a more adequate understanding of my own good, and one way of doing this is by reflecting on what is essential to me as a human being. Thus the concept of the human good does play a role in ethics, but it speaks to me through my own good, as it were, not over the top of it.

Let me give a concrete illustration of the difference between Foot’s view and my own. Foot claims – surely correctly – that keeping promises is necessary for human beings; society would collapse without it. This, on her view, is sufficient to give any individual a reason to keep his promises, even in situations where it runs counter to his perceived interests. This seems to me wrong. Take an individual – a free-spirited writer, say – who lives in no one place for more than a few years. He gets a feel for the place, uses it as material for his fiction, and then moves on, ending all his friendships. Of necessity, this individual breaks lots of promises. If reproached, he might reply that his own personal good – the good of a writer of his type – depends on not keeping promises. What does he care that keeping promises is necessary for human beings in general?

This reply seems to me decisive against Foot’s style of naturalism. But perhaps we can persuade this individual that he has misunderstood his own good. As a human being, friendship – in Aristotle’s very broad sense – must be important to him. And continually breaking promises in the way he does makes any true friendship impossible. (Note that this is true even if the person he has made the promise to never finds out; the friendship has still been breached.) If we can bring him to see that he has misidentified his own good, then we have given him a reason to change his behaviour – not necessarily an effective one, of course, for he may be weak-willed. And what if we can’t make him see that he has misidentified his own good? Well, in that case we have nothing more to say to him. But a person who really doesn’t care about friendship at all is not fully human. He is either godlike or bestial, in any event beyond the reach of morality.

Take another example. Having children is clearly necessary for our species in the same way that keeping promises is; the human race would disappear without it. Does this make it good for me? Foot would have to say “yes.” But a naturalist of Aristotle’s kind can take a more flexible view. What personal good hangs on my having children? The answer will be something along the lines of “self-perpetuation,” “immortality.” But immortality might be achieved in other ways, through the creation of great works of art or thought, great deeds, or union with God, in which case children might be a distraction and not good for me at all. Thus a naturalist of the Aristotelian type can acknowledge a plurality of goods here, corresponding the plurality of personalities and talents, whereas a Foot-style naturalist must insist on uniformity.