WEBVTT

00:00:00.000 --> 00:00:02.415 align:middle line:90%
[SQUEAKING]

00:00:02.415 --> 00:00:04.347 align:middle line:90%
[RUSTLING]

00:00:04.347 --> 00:00:05.313 align:middle line:90%
[CLICKING]

00:00:05.313 --> 00:00:11.745 align:middle line:90%


00:00:11.745 --> 00:00:13.120 align:middle line:84%
RICHARD DE NEUFVILLE:
Another one

00:00:13.120 --> 00:00:14.828 align:middle line:84%
I want to talk to you
about, because it's

00:00:14.828 --> 00:00:17.830 align:middle line:84%
very common in certain
forms of engineering,

00:00:17.830 --> 00:00:20.830 align:middle line:90%
is a robust design.

00:00:20.830 --> 00:00:23.170 align:middle line:84%
It sounds better if I
have a robust design, that

00:00:23.170 --> 00:00:24.230 align:middle line:90%
seems very positive.

00:00:24.230 --> 00:00:27.310 align:middle line:84%
We wouldn't want a
weak design, would we?

00:00:27.310 --> 00:00:29.860 align:middle line:90%
So it's a very positive affect.

00:00:29.860 --> 00:00:33.010 align:middle line:84%
What it means technically
in the way it's used

00:00:33.010 --> 00:00:36.070 align:middle line:84%
is a product whose
performance is

00:00:36.070 --> 00:00:39.440 align:middle line:84%
minimally sensitive to
factors causing variability.

00:00:39.440 --> 00:00:42.250 align:middle line:84%
So for example,
it basically says

00:00:42.250 --> 00:00:47.200 align:middle line:84%
that the standard deviation
around a average performance

00:00:47.200 --> 00:00:49.540 align:middle line:90%
is smaller.

00:00:49.540 --> 00:00:56.810 align:middle line:84%
And that is a useful
thing in many cases.

00:00:56.810 --> 00:00:58.960 align:middle line:84%
So here I have the
green and the red.

00:00:58.960 --> 00:01:03.670 align:middle line:84%
And the green is more robust,
its standard deviation is less.

00:01:03.670 --> 00:01:08.150 align:middle line:84%
And that's useful if you're
trying to do various things.

00:01:08.150 --> 00:01:10.990 align:middle line:84%
For example, if
you're trying to tune

00:01:10.990 --> 00:01:13.810 align:middle line:84%
in to a signal, a radial
signal or whatever,

00:01:13.810 --> 00:01:18.430 align:middle line:84%
you want something that you can
really hone in on and keep on.

00:01:18.430 --> 00:01:20.110 align:middle line:84%
Or if you're fitting
parts together,

00:01:20.110 --> 00:01:25.630 align:middle line:84%
a standard example would
be fitting a steel axle

00:01:25.630 --> 00:01:29.350 align:middle line:84%
into a steel wheel for
making-- for railroad cars,

00:01:29.350 --> 00:01:31.900 align:middle line:84%
and you want to have
the tolerances as small

00:01:31.900 --> 00:01:35.030 align:middle line:90%
as possible so it fits well.

00:01:35.030 --> 00:01:41.280 align:middle line:84%
But as a general thing, this is
not what you necessarily want.

00:01:41.280 --> 00:01:43.430 align:middle line:84%
For example, if
we're in a startup,

00:01:43.430 --> 00:01:47.030 align:middle line:84%
we want to limit the
downside, for sure,

00:01:47.030 --> 00:01:49.700 align:middle line:84%
but we want the standard
deviation to the upside

00:01:49.700 --> 00:01:52.520 align:middle line:90%
to be as high as possible.

00:01:52.520 --> 00:01:56.540 align:middle line:84%
That is, we'd like to say, all
right, our good design is not

00:01:56.540 --> 00:01:59.640 align:middle line:84%
one that is this
tight design here,

00:01:59.640 --> 00:02:02.450 align:middle line:84%
but one that eliminates
the downside, yeah,

00:02:02.450 --> 00:02:05.600 align:middle line:84%
but that stretches
off the upside.

00:02:05.600 --> 00:02:07.310 align:middle line:84%
And it would be
even better if this

00:02:07.310 --> 00:02:10.370 align:middle line:84%
had a higher
standard-- a greater

00:02:10.370 --> 00:02:13.130 align:middle line:84%
standard deviation more
towards the upside.

00:02:13.130 --> 00:02:17.340 align:middle line:84%
And in fact, that is the
notion of a lot of flexibility

00:02:17.340 --> 00:02:20.030 align:middle line:84%
design is you're trying
to do two things.

00:02:20.030 --> 00:02:24.260 align:middle line:84%
As with the garage case, you're
trying to limit the downside,

00:02:24.260 --> 00:02:26.900 align:middle line:84%
limit the losses, but
give as much chance

00:02:26.900 --> 00:02:29.790 align:middle line:84%
as reasonably possible
for the upside.

00:02:29.790 --> 00:02:33.920 align:middle line:84%
So the notion of robust
design being a good design

00:02:33.920 --> 00:02:36.890 align:middle line:84%
is very valuable
in some situations,

00:02:36.890 --> 00:02:40.850 align:middle line:84%
but really isn't what
we want in general.

00:02:40.850 --> 00:02:43.490 align:middle line:84%
That is, we want to
limit the downside,

00:02:43.490 --> 00:02:47.030 align:middle line:84%
maximize the variability of
the standard deviation caused

00:02:47.030 --> 00:02:48.420 align:middle line:90%
by the upside.

00:02:48.420 --> 00:02:53.190 align:middle line:84%
So once you think about
it in those terms,

00:02:53.190 --> 00:02:54.600 align:middle line:90%
it's fairly obvious.

00:02:54.600 --> 00:03:00.230 align:middle line:84%
But in fact, it's not
necessarily the obvious thing

00:03:00.230 --> 00:03:01.730 align:middle line:90%
if you talk about--

00:03:01.730 --> 00:03:06.080 align:middle line:84%
if you look at design books
and others that talk about what

00:03:06.080 --> 00:03:09.090 align:middle line:84%
they want and we really ought
to have a robust design,

00:03:09.090 --> 00:03:11.480 align:middle line:90%
it has a limited applicability.

00:03:11.480 --> 00:03:15.350 align:middle line:84%
But for system design
under uncertainty,

00:03:15.350 --> 00:03:17.990 align:middle line:84%
it is not the most
useful guidance,

00:03:17.990 --> 00:03:21.610 align:middle line:90%
in my opinion at least.

00:03:21.610 --> 00:03:27.150 align:middle line:84%
So in general-- and I will get
more into these next time--

00:03:27.150 --> 00:03:30.150 align:middle line:84%
what we we're
looking for is a way

00:03:30.150 --> 00:03:34.380 align:middle line:84%
of showing the dimensions
of choice that we'd have.

00:03:34.380 --> 00:03:38.880 align:middle line:84%
So in this particular
one, it refers to the case

00:03:38.880 --> 00:03:41.880 align:middle line:84%
by Rania Hassan doing
satellite design.

00:03:41.880 --> 00:03:47.640 align:middle line:84%
And she had various
alternatives-- a rigid fleet

00:03:47.640 --> 00:03:51.840 align:middle line:84%
where the elements could move
and then other alternatives--

00:03:51.840 --> 00:03:57.060 align:middle line:84%
and they had some measured
performance-- expected

00:03:57.060 --> 00:04:01.290 align:middle line:84%
value, the standard deviations,
such as for robustness,

00:04:01.290 --> 00:04:06.180 align:middle line:84%
the value of flexibility,
the amount of cost

00:04:06.180 --> 00:04:09.550 align:middle line:84%
that you'd have to pay,
the present values,

00:04:09.550 --> 00:04:15.270 align:middle line:84%
the maximum possible gain,
the maximum possible loss,

00:04:15.270 --> 00:04:18.450 align:middle line:84%
all elements that
you might think of.

00:04:18.450 --> 00:04:21.670 align:middle line:84%
And just to illustrate
that, in general,

00:04:21.670 --> 00:04:24.060 align:middle line:84%
we would somehow want
to display or make

00:04:24.060 --> 00:04:27.810 align:middle line:84%
obvious the range of
parameters and to think

00:04:27.810 --> 00:04:29.910 align:middle line:90%
about what is the best--

00:04:29.910 --> 00:04:31.410 align:middle line:90%
the preferred trade-off.

00:04:31.410 --> 00:04:36.510 align:middle line:84%
And so in this particular case,
we've surrounded in red boxes

00:04:36.510 --> 00:04:38.850 align:middle line:84%
the one that is the
best by each measure.

00:04:38.850 --> 00:04:42.420 align:middle line:84%
And then you can see that,
without going into the detail,

00:04:42.420 --> 00:04:45.180 align:middle line:84%
that this particular one
seems the best overall

00:04:45.180 --> 00:04:48.150 align:middle line:84%
but is not necessarily the
best in all circumstances.

00:04:48.150 --> 00:04:50.190 align:middle line:84%
And it's perfectly
reasonable to expect

00:04:50.190 --> 00:04:53.610 align:middle line:84%
that people could look at this
or think about this and say,

00:04:53.610 --> 00:04:58.470 align:middle line:84%
hmm, there are trade-offs
between different designs,

00:04:58.470 --> 00:05:00.150 align:middle line:90%
which ones do I choose?

00:05:00.150 --> 00:05:03.750 align:middle line:84%
And that choice
is not one that's

00:05:03.750 --> 00:05:07.080 align:middle line:84%
meaningful to be optimized,
because how many units

00:05:07.080 --> 00:05:10.830 align:middle line:84%
of standard deviation offset
how many units of expected value

00:05:10.830 --> 00:05:14.460 align:middle line:84%
is not something that
is mathematically

00:05:14.460 --> 00:05:16.810 align:middle line:90%
defined in any useful way.

00:05:16.810 --> 00:05:22.860 align:middle line:84%
So what is the takeaway
for the method?

00:05:22.860 --> 00:05:27.540 align:middle line:84%
It's that expected value is
not a sufficient measure.

00:05:27.540 --> 00:05:30.940 align:middle line:84%
I've said this before, but
I want to emphasize it.

00:05:30.940 --> 00:05:33.150 align:middle line:84%
Although that we've
used it a lot as a way

00:05:33.150 --> 00:05:38.250 align:middle line:84%
to get into the material
of dealing with uncertainty

00:05:38.250 --> 00:05:42.690 align:middle line:84%
in an initial way, we
want to think about it

00:05:42.690 --> 00:05:43.710 align:middle line:90%
in a more subtle way.

00:05:43.710 --> 00:05:46.320 align:middle line:84%
So to think about the maximum
and minimum that happen,

00:05:46.320 --> 00:05:49.360 align:middle line:84%
we want to be able to
compare alternatives.

00:05:49.360 --> 00:05:53.640 align:middle line:84%
Some of these things we can't
show graphically that much,

00:05:53.640 --> 00:05:57.420 align:middle line:84%
such as the capital
expenditures or the benefit-cost

00:05:57.420 --> 00:06:01.060 align:middle line:84%
or various measures, the
value of flexibility.

00:06:01.060 --> 00:06:04.120 align:middle line:84%
So there are different
ways to think about it.

00:06:04.120 --> 00:06:07.650 align:middle line:84%
And what I want to
leave you with for today

00:06:07.650 --> 00:06:13.380 align:middle line:84%
is that valuation of a system
design over its many dimensions

00:06:13.380 --> 00:06:14.790 align:middle line:90%
is not at all obvious.

00:06:14.790 --> 00:06:17.640 align:middle line:84%
There's a lot of
subtleties about it,

00:06:17.640 --> 00:06:24.420 align:middle line:84%
as Flynn alluded
to very correctly.

00:06:24.420 --> 00:06:30.150 align:middle line:84%
And we need to think
about ways in which we

00:06:30.150 --> 00:06:33.390 align:middle line:84%
as analysts and
managers can identify

00:06:33.390 --> 00:06:35.700 align:middle line:84%
the dominant solutions,
the ones that

00:06:35.700 --> 00:06:39.300 align:middle line:84%
are clearly better than
others, and discard the others

00:06:39.300 --> 00:06:41.760 align:middle line:84%
and then think about
what the trade-offs are

00:06:41.760 --> 00:06:43.060 align:middle line:90%
along that curve.

00:06:43.060 --> 00:06:46.050 align:middle line:84%
If you want the dominant curve,
the Pareto optimal curve,

00:06:46.050 --> 00:06:48.630 align:middle line:84%
you have to think about what
are the ones that we want

00:06:48.630 --> 00:06:52.290 align:middle line:84%
to keep and explore as
to the relative merits

00:06:52.290 --> 00:06:54.900 align:middle line:84%
and to help the people
we're working for,

00:06:54.900 --> 00:06:58.980 align:middle line:84%
either our bosses, or ourselves,
or a more general public,

00:06:58.980 --> 00:07:02.160 align:middle line:84%
as to what is the
preferred design

00:07:02.160 --> 00:07:05.540 align:middle line:84%
or where there's a
negotiation between them.

00:07:05.540 --> 00:07:10.000 align:middle line:90%