This morning we play the game of thinking about how the brain might go about its business if it were a computer, prompted in large part by reference 1 and a variation on the game played at reference 2.
We suppose the brain has been told for look for something, say a mammal whose common name in English starts with a certain letter, say ‘C’. ‘C’ for coypu or for Chinese pangolin.
We suppose also that the brain strategy is in two parts. Part 1 – the left hand pale box in the diagram – delivers a candidate object from the brain’s store while part 2 – the larger, right hand pale box – checks that candidate against the requirement. Plus a supervisor who does things like going back to part 1 if we need to look at another candidate, quite possibly but not necessarily because a candidate has just failed.
We suppose that the brain stores all its objects – objects in the sense of real world places, times and things - in the one place. It can just poke around, more or less at random in that place. We might think of the store, following brain structures generally, as a two dimensional sheet which can be sampled by generating a couple of random numbers, interpreted as coordinates. Or one can just scan the store in a systematic way, starting top left and ending bottom right, as one might scan the page of a book. The results of such a scan might be more or less random or they might be organised in some way.
Alternatively, depending on the question being asked of it, there might be an index which is helpful: in computers and books, indexes can be very useful - while google must have lots of them – and some of them at least are dynamic, adapting to new objects and to new uses. So the index to a book might tell me on what page or pages I can find the word ‘elephant’. A cunning index might do some of ‘elephants’, ‘elephantine’, ‘elephantiasis’ and ‘éléphant’ while it was at it. And given its behaviour, the brain looks to have some capability of this sort. It knows what object it was last looking at. From one object it can get to other objects, with the latter associated to the former for one reason or another. It may have indexed all the elephants it knows, if that is something of special interest. It may have memorised and indexed various sequences, for example all the tube stations between Upminster and Earls Court (carefully avoiding here any need for branching, which would take us out of the world of sequences).
Maintenance of all these indexes might be partly a matter for background activity, perhaps something which gets attended to during sleep, partly a matter for conscious effort.
So first off, looking at part 1, the brain compiles the query, decides how to handle it. Is it going to go for indexed access or random access? There might also be some context which tells us that speed more important than accuracy, or what the stakes are and such context will qualify the compilation process. Having gone through all this and decided how to execute the query, the brain then actually does the execution and delivers a candidate.
We can now move onto part 2 and the first thing to do is to check for candidates which are either pictures or words. Pictures down the left hand path, words down the right hand path, anything else discarded.
In this, the brain continues to check for priming (see reference 3). Is this candidate already in mind? If it is, this and subsequent checks will be qualified, biased in its favour. Perhaps making mistakes on its behalf.
Picture candidates are then subjected to a battery of picture tests, a battery of tests which is context sensitive. We suppose that testing is a bit hit and miss, sometimes giving the right answer, sometimes not. The order of testing might also be a bit hit and miss. Plenty of noise in the system as well as the bias which may have been introduced by priming. So in this case, we might test whether the picture is that of an animal. Then does the animal have two eyes? Does it have four legs? Does it have fur? After each test the brain looks at the score and decides what to do next. Outright acceptance and promotion to consciousness? Outright rejection? Switch to word tests (change mode in the illustration)? Do another picture test? And this will go on until there is an outcome, or perhaps until the process is timed out by the supervisor.
Word tests go through the same sort of thing. Part of this will be classification. Is the candidate in hand an X, where the mammalian status of an X is known? Perhaps mammalian status is a property for which animal words have been pre-coded. In any event, in the case of a positive reply, the brain can apply a well-known syllogism – well known that is in historical rather than evolutionary time – to either accept or reject the candidate.
Most of the time, given sobriety and plenty of parallelism, something is delivered to consciousness in fairly short order, say within a second or so of kicking the process off, during which time the brain may have sampled hundreds, if not thousands, of its objects.
The object offered up is then subject to a final check in consciousness. Which can be orderly and use a proper algorithm, an algorithm which is rarely wrong. The sort of orderly checking which the unconscious mind seems to find difficult and which often makes mistakes – mistakes which are the basis of many tricks, conjuring and otherwise. Evolution did not make much allowance for tricksters, not that common out in the jungle.
The number of right answers will depend on what is being looked for, will depend on the question. There might only be one right answer (what is the big city on the Thames), there might be lots – or at least several (give me a town on the Thames starting with ‘W’). In either case the process is something of a race: which candidate, right or wrong, will be first past the post?
One tricky aspect of the system proposed here is the need to generate or perhaps just to select suitable, context sensitive tests. In the case of common questions, batteries of tests might be to hand, while, in the case of rare questions, rather more preparatory work might be needed. While google’s deep mind might do something else altogether; something which is probably fairly proprietary.
PS: one might suppose that there are lots of people out there doing exactly the same kind of thing as I am doing here, lots of people for all the lots of ways there are of drawing diagrams such as that above. Not to mention all people doing common or garden work on search strategies. But I don’t come across them very often – perhaps, in the end, because it is more fun to do it oneself than to read about what the other chap is up to.
Reference 1: http://psmv3.blogspot.co.uk/2016/05/new-game.html.
Reference 2: http://psmv3.blogspot.co.uk/2016/05/thoughts-about-self-control.html.
Reference 3: https://en.wikipedia.org/wiki/Priming_(psychology).
No comments:
Post a Comment