Hi Purma,
Yes you are on the right track with where I'm going.
Thank you for describing all the detail. It's helpful. I wish I'd written it.
Although, rather than hand picking the low hanging fruit, I was thinking more of a universal, fully-automated fruit picking machine, with automated preparation, canning and distribution and recycling. Did I say also fully automatically designed by computer? ;-)
It is true that a general method of predicting outcomes resulting from the introduction of designs in the world was not feasible.
In the simple picture, the variety available through the predicting systems was less than the variety created by the world with or without the change in variety resulting from the introduction of the new design.
Put in a slightly more useful bit more complicated manner, the dynamics of distribution of variety shaped by the predicting system and its management were ineffective in mapping or influencing the variety distribution dynamics of the world with or without the change in variety resulting from the introduction of the new design, due to lags and lack of capacity
Things have changed on several fronts to make the project more possible (computers, networks, algorithms, publicly accessible large datasets...). A crucial leverage point is to be sparse with data.
More seriously, one of the things to remember about such an endeavour to predict outcomes of designs is what is NOT needed in theory terms. For example, there is no need at all to know how designers design things. Nor is there much need to know about colour theories or what is in them. In the latter case, I suggest the key thing to know about colour theories (which is what I asked about) is the benefits of using different forms of measurement metrics for colour - and NOT in terms of how designers use colour, rather in terms of how those metrics can be best integrated into a predictive process about other outcomes. Similarly for colour we possibly don't need to know anything, theory or prior data, about how people respond to colour or academics gathering such information and writing papers about it. The relevant response data can probably be gathered incidentally within trials of a project or via proxies or social media or sales data, and incorporated via self learning algorithms - without actually explicitly analysing the data.
An interesting thing about some of these processes is how they disintermediate and make redundant some of the current roles of research academics particularly in the older traditions based in deriving theory from large amounts of hand-collected evidence (Ken will likely want to respond to this). They point to several changes in what it means to be a university and for a significant shift in what research universities are likely to be funded.
Thanks again.
Warm regards,
Terry
-----Original Message-----
From: [log in to unmask] [mailto:[log in to unmask]] On Behalf Of Purma Jukka
Sent: Monday, 22 February 2016 6:04 AM
To: PhD-Design - This list is for discussion of PhD studies and related research in Design <[log in to unmask]>
Subject: Re: Assume fixed number of colours in design?
Hello,
I am sympathetic to predictive theory effort, though I think it is too general to succeed. If the goal is a general predictive theory of consequences of design, then I see to be on par with these challenges:
- predictive theory about reception of a theatre play
- predictive theory about critical reception and sales of a book, and if there will be film versions
- predictive theory of box office success for forthcoming movie
- predictive theory of stockmarket response to a new listed company Based on amount of flops and failures in all of these fields, and the huge money involved in trying to predict and prevent these failures, I’d say that predictive models employed are not very reliable for individual cases. There certainly have been thousands of people trying to create and improve these models. These are chosen for comparison because in all these the new thing will enter the whirlpool of existing things at that certain point of time. Predicting the response requires predicting the state of that whirlpool, the environment where the product will go live.
E.g. the splash that the design of mid-2000’s iBook would create would be very different in 2000, 2005, 2010 and 2015
However, predictive theories that are narrow and dumb *can* work surprisingly well. Predictive theories are funny in that way, which has seemingly bothered many participants, that they don’t have to be based on true premises or plausible theory to succeed really well. Three examples:
1. There is the old joke about physicist starting with ”we assume a cow to be a sphere with equally distributed mass”. We can use our expertise and credentials in biology to say that the theory is wrong, cows are not such things. Such theory would however work for many problems where the goal is to predict how many cows can be put into some space or vehicle.
2. I want to create a model about traffic in Helsinki to analyze congestions when more cars are put into system. Assume that I have complete data about all traffic moving in some time range. Based on that I can calculate for every crossroad that in specific time of day, there is a specific probability for car turning right, going straight or turning left. Putting cars following these probabilities into simulation would in overall reproduce Helsinki traffic, and adding more cars following the same probabilities would create congestions in right places. However, looking at behaviour of one car would reveal really pointless driving. The model also wouldn’t be able to predict what would happen if there are new interesting target for drivers, e.g. a shopping centre, added into the system. The point is that the model of driver who randomly choses which way to turn using some probability function is as wrong as a spherical cow, but would be right for this task.
3. I want to make a system that understands what I like in music and suggests me music that I would like, e.g. it predicts my enjoyment in music. In a way it does what my friends more invested in music have always done using their deep understanding of bands and how they sound. After some effort, the final system to understand my taste in music doesn’t analyze the songs or care about the content of the music *at all*. It imitates taste in music by having all of the titles in record collections of large amount of people and finding correlations, where high percentage of those who have A also have B. If I have A, but don’t have B, then it recommends me B. All without knowing anything about content of A and B. Result is a system that understands my music taste, but doesn’t understand music at all. Same algorithm would work with books or any collections that people gather in time: modern art, films, theatre plays — it doesn't matter how deep or complex the works collected are, if the process of building the collection has a certain form. We all have used such systems by now. (It also wouldn’t be able to predict how new band/play/book/film would be received, though it could provide some hints how a new album/book/film from a known author/band would perform.)
It is bizarre that a useful prediction of musical taste is so easy, when the phenomenon of taste is so beyond explanation and intertwined to so many things. There may be much cognitive psychology linked to phenomenon of artistic taste, but it has long, long way to go for any explanatory adequacy.
I think that Terry is looking for some similar low-hanging fruits, to find predictable ways in how we react to designs, ways that can be extrapolated, short-cuts to take. In my opinion he should be, other paths are just too difficult.
Jukka
(p.s. Previous examples, esp.example 2 can also be used against using predictive power of a model/theory as a proof for correctness of its theoretical claims. This is often seen when making claims about human cognition based on promising results of neural network models. )
> Klaus Krippendorff <[log in to unmask]> kirjoitti 21.2.2016 kello 18.33:
>
> Terry
> I thing Gunnar's suggestion "make it real , and make it now" should nudge you in the same direction.
>
> You ask for an example of a predictive theory based on past observations, I already gave you one: predictions of election results based on interviews of potential voters. It presumes that the sample is representative of a population of voters and that their preferences do not change.
>
> However, designed interventions, if effective, should invalidate these predictions. Having (from a particular perspective) desirable consequences is the aim of designs. In the reality of elections designers compete in the name of particular candidates. So, it is not as simple as Herbert Simon once suggested that a predictive theory must include the effects of that theory on what it aims to explain.
>
> In a nutshell: Terry, I suggest you become real and not lost in abstractions.
>
> Klaus
>
> Sent from my iPhone
>
>> On Feb 21, 2016, at 11:05 AM, Terence Love <[log in to unmask]> wrote:
>>
>> Dear Klaus,
>> My understanding is that predictive theories can be at any level of
>> abstraction we choose. Its only in the singular case where you wish
>> to try to make an exact theory replica of reality in all its details
>> that you need to be exactly representative. Do you know ANY theory that does this?
>> Klaus, please accept my apologies for the brevity. I spent too long
>> on the posts to Ken and Gunnar and need to sleep now.
>> Warm regards and thanks,
>> Terry
>>
-----------------------------------------------------------------
PhD-Design mailing list <[log in to unmask]> Discussion of PhD studies and related research in Design Subscribe or Unsubscribe at https://www.jiscmail.ac.uk/phd-design
-----------------------------------------------------------------
-----------------------------------------------------------------
PhD-Design mailing list <[log in to unmask]>
Discussion of PhD studies and related research in Design
Subscribe or Unsubscribe at https://www.jiscmail.ac.uk/phd-design
-----------------------------------------------------------------
|