Decision Trees (was Report Back from IxDG NYC Face-2-Face)

14 Jan 2005 - 6:54am
9 years ago
2 replies
838 reads
Mike Baxter
2004

<Ben>My favourite management consultant, Gerry Robinson, writes about this
in
his book "I'll show them who's boss" (which supports the 2 TV series he
did for the BBC). He calls it "decision tree", and espouses it as the
the ideal way to proceed dynamically through a complex problem, focussed
on a goal, and avoiding getting bogged down on too much detail at
once.</Ben>
I did quite a bit of work on the idea of decision trees to underpin IxD a
few months back and they suffer certain quite fundamental problems. If one
node in a decision tree cannot be resolved then access to all the branches
under it are blocked (the decision made at one node determines which of the
branches under it are chosen).
Complex problems, which most of us are dealing with most of the time,
normally have rich inter-dependency. In other words, the decision you make
about one feature has knock-on consequences about the options available in
other features. This is always difficult and often impossible to model in a
hierarchical decision tree structure. So often the type of decision-making
best suited to IxD is to focus on one decision node (a component part of the
entire decision), think of the options available, constrain them if
appropriate and then look at the impact of these alternatives on the other
decision nodes. Then move on to another decision node and repeat. What you
end up with is an iterative design process gradually refining and narrowing
the range of options until you reach the best (or least worst!) solution to
the entire problem.
This is best thought of as a decision NET rather than a decision TREE and as
far as I'm concerned it is a much better model of how we intuitively think
as designers. My understanding from research into this stuff is that
decision trees are well suited to complicated problems - a problem where
there is a single correct solution to a problem (e.g. why has my television
stopped working). Most IxD problems are complex:
- they have better or worse solutions, not right and wrong ones;
- they require iteration and have no clear stopping rules;
- they cannot simply be deduced from any given set of information;
- they concern business processes which often produce very different
outcomes given very similar initial conditions;
- they are based, to some extent, upon intuition and judgement on the part
of the decision-maker.

Mike Baxter
http://www.saleslogiq.com

Comments

14 Jan 2005 - 8:52am
Ben Hunt
2004

<Mike>
Most IxD problems are complex:
- they have better or worse solutions, not right and wrong ones;
- they require iteration and have no clear stopping rules;
- they cannot simply be deduced from any given set of information;
- they concern business processes which often produce very different
outcomes given very similar initial conditions;
- they are based, to some extent, upon intuition and judgement on the
part of the decision-maker.
</Mike>

I think that's true, Mike, there are times when we need to apply
intuition and gut-feel to a problem, looking at the whole rich fibre of
what-ifs. Sometimes the way to the solution can appear as out of the
mist in a crystal ball, but I still *feel* that decision tree discipline
can be appropriate to a lot of IxD problems.

The key, to my mind, is to define the parameters of the problem. The
main parameter is the goal: what will it be like when you've succeeded?

If you're clear about that, and clear on range of movement, you can ask
at each node what choice will provide the biggest step towards the goal.
As long as that step stays within the bounds of design parameters,
accept it and carry on. The end result should be a clean design solution
that's made up of a few big, bold steps.

Applying this to each of your points above:

- They have better or worse solutions, not right and wrong ones;
> True. We have mechanisms to evaluate better or worse: using scenarios
and goals, a solution that lands a persona cleanly at her goal with the
fewest steps and least effort is better than one that doesn't do it so
well

- They require iteration and have no clear stopping rules;
> Maybe iteration is a habit? Goals are clear stopping rules. Once
you've proved you've achieved good progress towards the goal-set for a
step in the decision-tree, that step is done and you proceed.

- They cannot simply be deduced from any given set of information;
> Qualitative research information and creative imagination are the
data, and the deduction is a mixture of common-sense, creative thinking
and empathy.

- They concern business processes which often produce very different
outcomes given very similar initial conditions;
> I guess the power of a great design process is to simplify this
situation. I think some case studies are needed to bottom out this
point.

- They are based, to some extent, upon intuition and judgement on the
part of the decision-maker.
> Absolutely, that's what we're paid for - having good intuition and
good judgement, which we can apply with confidence thanks to experience
and powerful processes.

- Ben

14 Jan 2005 - 9:40am
Mike Baxter
2004

<mike> [IxD thinking] is best thought of as a decision NET rather than a
decision TREE and as
far as I'm concerned it is a much better model of how we intuitively think
as designers</mike>

<ben> but I still *feel* that decision tree discipline
can be appropriate to a lot of IxD problems ...

... Maybe iteration is a habit?
</ben>

Okay, it's Friday afternoon, let's step this up a gear :)
I think iterative thinking is absolutely fundamental to IxD. One of the
things that transforms the way design students think is when they realise
the perils of hierarchical thinking and begin to think iteratively about the
problems they are trying to solve. Even as world-weary professionals, who
amongst us hasn't been caught out by treating a high-level decision as fixed
(or decided upon and forgotten) only to find (too late?) that challenging
and questioning that decision would have opened up the possibility of much
better design solutions. It is the repeated questioning of all the key
decisions/assumptions that gives rise to innovative design.
So, to come back to the main point of the thread, decision trees are a
representation of hierarchical thinking. As such they are not only a
significant mis-representation of how we as professional designers actually
think, they are also dangerously misleading for the less experienced because
they lead towards the type of hierarchical thinking that leaves high-level
decisions unchallenged, thereby constraining design solutions
inappropriately.
Mike

Syndicate content Get the feed