From: Alan R. <ala...@gm...> - 2007-05-17 04:51:23
|
Hi Matthew, My understanding of Assay as it currently exists is, effectively, protocol application with output that is data. (though I note that I don't see an english or format definition attached ... So TeaTemperatureMeasurementAssay =def a protocol application where a digital thermometer is used to measure the temperature of the tea in a cup. The thermometer tip is dipped into the tea until the thermometer goes "beep" after which the number on the display is recorded as output. input: cup of tea instrument: digital thermometer model xxx serial yyy output: data (temperature of tea as read off thermometer, units: Fahrenheit) It is implicit (and could be added as an axiom) that every Assay has a process part that includes an observation. This seems adequate - I'm not sure why one needs to state this explicitly. I suppose that one could argue that not marking this as an observation that a human made and transcribed doesn't alert to a possible class of errors that a wireless transmitting thermometer wouldn't make. OTOH, the best place to record this sort of information might be on the instrument e.g. all digital thermometers model xxx require a person to read and transcribe the temperature. On May 16, 2007, at 5:05 AM, Matthew Pocock wrote: > On Tuesday 15 May 2007, Alan Ruttenberg wrote: >> Hi Frank, >> >> The request I had on the call was for you to list some consequence of >> not making this choice, i.e. to show that there is some inference >> that would or would not be made in the absence of this additional >> class. A schematic example showing this would be helpful. >> >> To put in another way, we all have preferred ways of *thinking* about >> this subject, but a good way to decide whether those thoughts need to >> turn into action is to see whether there is any measurable >> consequence of doing so. This is along the same line of thinking that >> puts the (english and logical) definition of a class as more >> important than its name or label. >> >> Regards, >> Alan > > Hi Alan, > > Let's have a go at a real example that requires the more complex > representation. Feel free to rename things as is needed to make this > OBI-friendly. > > We all know that when performing experiments, there are errors, both > systematic and sporadic, and that the process of making an observation > perturbs the system being observed (by a larger or smaller amount). > Let's > take the case of measuring the temperature of a cup of tea with a bulb > thermometer. This will look something like: > > input: > cup of tea (has unknown temperature tmp1) > thermometer (no meaningful temperature reading) > agent (no knowledge of the temperature of the tea) > > output > cup of tea (has unknown temperature tmp2) > thermometer (has reading r1) > agent with quantitative observation of thermometer reading (o1) > > (agent is perhaps optional - I put it in for completeness sake) > > Clearly o1 is related to tmp1. However, it is not correct to assert > that they > are the /same thing/. We can't just assume that the observed > thermometer > reading is an accurate measure of the temperature of the cup of tea > before it > had its temperature taken. We assert that o1, r1, tmp2 and tmp1 are > all > related, but as good experimental scientists our responsibility is to > describe how they are related and to ask what could have gone wrong. > > * tmp1 could be higher than tmp2, as it could have lost heat to the > surroundings > > * tmp1 could be higher than tmp2, as we know that the process of the > thermometer getting a reading sucked some of the heat out of the > cup of tea > into the thermometer - that's how the thing works > > * tmp2 could be systematically miss-represented by r1, because of a > calibration error > > * r1 could be unrelated to tmp2 because it is a broken thermometer > > * o1 could be a poor representation of r1, because of observer error > > If we collapse the representation of acquiring a qualitative > (quantitative) > observation for some material, then how can we make it plane that > these > sources of error exist, or specify that a protocol must minimize > some of > these, or describe that different pieces of equipment or observation > processes can be callibrated to produce similar quantitative > observations? > How can we describe a range of ways of performing observations and > discover > which ones are good at capturing r1 accurately? How can we specify > that the > loss of heat to the surroundings can be minimized by doing the > experiment in > a thermos flask rather than a plastic cup? > > This is a trivial case, but look at any good quality lab procedure > and it will > have opportunities for these kinds of errors to creep in, and > things in place > to minimize some of them. I'm not making a case for forcing labs to > capture > all of this info. I'm making a case for us ensuring that OBI has > places to > put this stuff, and that the reasoner has access to (potentially > anonymous, > potentially introduced during reasoning) instances representing > what's going > on here. It is all a necesary part of documenting experimental > assumptions, > which makes it more possible to compare conclusions from different > experiments and to select protocols appropriate for the task at hand. > > Matthew |