gilbert@cs.glasgow.ac.uk (Gilbert Cockton) (04/05/89)
In article <819@htsa.uucp> fransvo@htsa.UUCP (Frans van Otten) writes: >Jerry Jackson writes: >Please see the difference between "many simple tasks, all the same" and >"many different and difficult tasks". But yes: AI was invented (at least) >20 years ago. The cheque clearing system you write about does understand >how to process a check. No it doesn't, it relies on humans to get the cheques to the right place, and to input cheques where the magnetic characters can't be read (hint on how to slow down cheque clearing:-)). The decision on 'bouncing' a cheque, part of cheque-clearing, is rarely made by machines. Cheque clearing is a human-machine system with a clearly defined set of subordinate tasks assigned to the machine. Humans hold the system together, and therefore only they understand cheque clearing. The automated tasks possess no concept of cheque clearing in all its glory. >So when we understand how the human mind works, we can build a machine >which has properties like "consciousness", "understanding" etc. No, you assume these are properties of mind. Until you give me a sensible account of the role of 'mind' in human agency, I cannot accept or reject anything you state on the issue. How about the act of making a cup of tea. Where does mind come in in the Chinese Tea Room? As for it's artefactness, it is quesionable whether any artefact fully Copies any entity in the physical world. Indeed, it may be impossible to fully synthesise anything, since there is no objective test for knowing that a natural entity is fully understood. There are many objective criteria for knowing that something is not fully understood, as the natural entity and the simulating artefact perform differently. Natural entities and simulating/surrogate artefacts can only be equivalent in so far as they perform the same way under a finite and enumerated set of conditions (tasks for mind machines). Under these circumstances, 'complete understanding' is impossible. The fact is, societies only seek knowledge on useful things. The post-war knowledge for its own sake institutionalised is a minor deviation which is on the way out. Total knowledge is uninteresting. Sensible folk restrict themselves to useful knowledge with an obvious application (I count filling out existing applied theories as useful, so 'basic' research is not ruled out by this dogma). The only things that matter are tasks, and even these are slippery, as I/O equivalence cannot be tightly defined for most interesting tasks. The Turing test is thus an uninteresting subjective game. People are unlikely to agree that a system is 'intelligent'. It depends on who you ask, and what they ask the system to do. Given all these epistemelogical problems - and more (see all 17 volumes of obscure Euro-drivel) - I stick to my argument that computer simulation cannot advance our understanding of 'mind', rather it always lags behind it (even pulling it back by showing gaps in current understandings). The gap between understanding and computability is even larger, due to the lack of sources used by strong AI research. Current computer models come nowhere near our cultural understanding of human agency, and given the preference for hacking over reading, the gap is unlikely to be closed. -- Gilbert Cockton, Department of Computing Science, The University, Glasgow gilbert@uk.ac.glasgow.cs <europe>!ukc!glasgow!gilbert
rjc@aipna.ed.ac.uk (Richard Caley) (04/07/89)
In article <2728@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: >I stick to my argument that computer >simulation cannot advance our understanding of 'mind', rather it always >lags behind it (even pulling it back by showing gaps in current >understandings). So now showing up gaps in our current understanding is a bad thing? We should perhaps skip allong happily constructing partial and possibly self contradictory theories? -- rjc@uk.ac.ed.aipna "Politics! You can wrap it up in fancy ribbons, but you can't hide the smell" - Jack Barron
gilbert@cs.glasgow.ac.uk (Gilbert Cockton) (04/10/89)
>So now showing up gaps in our current understanding is a bad thing? We >should perhaps skip allong happily constructing partial and possibly >self contradictory theories? Of course not. Folk who need to spend months writing a computer program to do it though can't be using their grey matter to its limit. The question is, has computer simulation of 'mind' exposed ignorance unknown to mainstream cognitive psychologists. What's computer simulation got to do with 'self-contradiction' as well? Is computer logic really the arbitor of consistency? If not, Strong AI is an expensive way of finding holes in knowledge. -- Gilbert Cockton, Department of Computing Science, The University, Glasgow gilbert@uk.ac.glasgow.cs <europe>!ukc!glasgow!gilbert