Decisions in the age of digitalization | Michael Feindt | TEDxKIT
-
0:10 - 0:13Okay, decisions in the age
of digitalization. -
0:14 - 0:17To make one thing very clear
at the very beginning: -
0:17 - 0:21Most important human decisions
can and will never be automated. -
0:21 - 0:25They are done by gut feeling,
under extreme uncertainty. -
0:25 - 0:28I'm not going to talk about these.
-
0:29 - 0:31Don't blame me for that.
-
0:31 - 0:360% automation on decisions like,
-
0:36 - 0:39these are examples from my life:
marrying my wife, -
0:39 - 0:42declining a tax-free,
permanent CERN job offer, -
0:42 - 0:45instead of a KIT professorship,
-
0:45 - 0:47or founding Blue Yonder.
-
0:47 - 0:50In retrospect, these were
all very good decisions, -
0:50 - 0:52but they weren't clear beforehand.
-
0:52 - 0:55So I'm not talking about these things.
-
0:55 - 1:01But I claim that 99% of all operational
decisions in enterprises -
1:01 - 1:03can be automated.
-
1:03 - 1:05And they will be automated.
-
1:05 - 1:09They will also be simultaneously improved
by quite some margin. -
1:11 - 1:13That is what I am going to report on.
-
1:14 - 1:18And these are especially repeated,
regularly repeated, similar decisions. -
1:19 - 1:22Let's take an example from retail.
-
1:23 - 1:26Assume you are a store manager
and you have to decide: -
1:26 - 1:29Of these articles,
how many do I have to order -
1:29 - 1:31so that I have enough tomorrow,
-
1:31 - 1:35but not so many
that I have to throw them away -
1:35 - 1:36once the shelf life is passed?
-
1:36 - 1:40Ten? Zero? 100?
-
1:40 - 1:45That's a decision many people
have to make every day, again and again. -
1:45 - 1:46Or, again, you are store manager:
-
1:46 - 1:50Do you want to decrease
the price of this product today? Yes? -
1:50 - 1:54By 5%? By 10%? By 20%?
-
1:54 - 1:57That's a decision that people
have to make very often. -
1:57 - 2:01Or to send this expensive catalog
to that customer or to that customer? -
2:01 - 2:04Is it worthwhile? Will it be good for him?
-
2:04 - 2:06Will it be good for us?
-
2:06 - 2:09These are questions I'm talking about.
-
2:10 - 2:13Don't you believe the 99%?
-
2:14 - 2:20First, you have to know how most
common decisions are taken in real life. -
2:20 - 2:21And that's about something like that,
-
2:21 - 2:28so 90% of these decisions are:
Do nothing, or do what we always do. -
2:29 - 2:30What people always have done,
-
2:30 - 2:32even before I came here
-
2:32 - 2:39and wanted to bring into this business
everything I have learned at university. -
2:39 - 2:43First, you are told, "No, we do it
this way - ever, always." -
2:44 - 2:49About 9% apply business rules
in one way or another. -
2:49 - 2:52But many of these business rules,
if you really look at them, -
2:52 - 2:53are not very good.
-
2:53 - 2:58Almost none of them
has a real proof of value. -
2:59 - 3:03[For] only about 1% of decisions,
-
3:04 - 3:08[does] somebody sit down,
use his brain, and think about it. -
3:09 - 3:13And even this 1% is far from optimal.
-
3:13 - 3:15For this, you have to know
-
3:15 - 3:18that the human decision-making
system has two systems. -
3:18 - 3:22The so-called system one
is fast and intuitive, -
3:22 - 3:24but it has many biases.
-
3:25 - 3:28System number two is slow and rational,
-
3:28 - 3:29but it is hardly used.
-
3:29 - 3:33Why? Because it's work, it takes
a lot of energy to use it. -
3:34 - 3:39Most of the time we don't use it,
we use system one rules for everyday life. -
3:40 - 3:46And even, and especially, scientists
and experts use their "gut feeling." -
3:47 - 3:51They are not always thinking
and deciding rationally. -
3:52 - 3:54And it's very important to know,
-
3:54 - 3:57system one, the fast one,
cannot speak statistics. -
3:58 - 4:04It cannot really judge risk and chance.
-
4:05 - 4:06That's a big problem.
-
4:07 - 4:10Daniel Kahneman, Nobel Prize winner,
-
4:12 - 4:16has done research his whole life
about how we think. -
4:17 - 4:21His book, "Thinking Fast and Slow"
is really worth reading. -
4:24 - 4:26He describes all of this.
-
4:26 - 4:30He describes how our system one
makes decisions, -
4:30 - 4:33and it is all explained by evolution.
-
4:33 - 4:37So it was right: we are here now
because we did it that way. -
4:38 - 4:43But there are many, many effects,
for example, the IKEA effect: -
4:43 - 4:45We value something higher
-
4:45 - 4:49when we have worked
for it ourselves a bit. -
4:49 - 4:53There are many
so called "cognitive biases." -
4:53 - 4:56And it's very funny to see
-
4:56 - 5:00that somebody, like myself,
who thinks he is very rational, -
5:00 - 5:03has all these biases every day.
-
5:04 - 5:07So, it's like that,
we can't do anything against it. -
5:07 - 5:09So, our question was:
-
5:09 - 5:12How can we make more rational decisions
than we are actually doing? -
5:12 - 5:16The origin of this idea
comes from CERN and KIT, -
5:16 - 5:19so from where I do my research work.
-
5:19 - 5:22The development is now done
at Blue Yonder. -
5:22 - 5:27The basis is really big data,
Bayesian statistics, machine learning, -
5:27 - 5:30data science, stochastic programming,
causality reconstruction. -
5:30 - 5:32In other words, it's really science,
-
5:32 - 5:37how scientists work
in order to get insights from data, -
5:37 - 5:43and then optimize decisions on the basis
of these insights for the future. -
5:44 - 5:47It's a purely scientific
thinking and ethos, -
5:47 - 5:51that is also not true
in most enterprises, as I have learned. -
5:51 - 5:54We are trying to do
this technology transfer -
5:54 - 5:57to reinvent business processes
with scientific methods. -
5:58 - 5:59This is a key.
-
5:59 - 6:02The key to better decisions
in the digitalization era is: -
6:02 - 6:04data and algorithms.
-
6:04 - 6:08So, first, you need
relevant objective data. -
6:08 - 6:11So, it's mostly your data
or your company's data. -
6:11 - 6:15And you need good,
mathematically correct predictions -
6:15 - 6:20for near future events on the basis
of this data, including uncertainties. -
6:20 - 6:22That's a very important point.
-
6:22 - 6:26And this is handled by an area
that is called "predictive analytics." -
6:27 - 6:31Then you also want to know, if you decide
this or that, what does it mean? -
6:31 - 6:34So you need cost functions,
utility functions: -
6:34 - 6:37What is the consequence
of this or that decision? -
6:39 - 6:42And then you want
to optimize this decision -
6:43 - 6:47using the prediction
and the utility function -
6:47 - 6:49to give a recipe for what you should do.
-
6:49 - 6:51That is called "prescriptive analytics."
-
6:51 - 6:54And finally, the King's discipline:
-
6:54 - 6:57If you [have got this] far,
then you can automate it, -
6:57 - 7:01then it's completely automated.
-
7:01 - 7:04When can we use predictive analytics?
-
7:06 - 7:10In between purely random
processes, like the lottery, -
7:11 - 7:16and completely deterministic processes,
like a mechanical pendulum. -
7:16 - 7:20Most processes that we are interested in
are in the continuum between those. -
7:20 - 7:22So they are partly predictable,
-
7:22 - 7:25but there's a lot of randomness
and uncertainty inside. -
7:25 - 7:29Most people think that the world
is very deterministic. That's not true. -
7:29 - 7:31There is some determinism inside,
-
7:31 - 7:34and it's our work
to get this component out. -
7:34 - 7:38But there is always lots
of uncertainty inside, -
7:38 - 7:43and in evaluating the value of a decision,
-
7:43 - 7:45we also need to take
into account the uncertainty. -
7:45 - 7:47That's a very important point.
-
7:47 - 7:50So, that is what
predictive analytics can do. -
7:50 - 7:54Going back to our store manager
and his problem: -
7:54 - 7:57How many articles to order?
-
7:57 - 7:59There may be many influencing factors
-
7:59 - 8:03on how many articles
will be sold tomorrow, -
8:03 - 8:06from weather, weather forecast,
price, promotion, -
8:07 - 8:11day of the week, holidays,
seasonality, and so on. -
8:11 - 8:14But the store manager has
to make his decision based on, -
8:14 - 8:17or take his information from,
many individual events: -
8:17 - 8:22namely many articles, many locations,
many points in time. -
8:22 - 8:27And all together, there are many
such decisions to take. -
8:27 - 8:29The output of predictive analytics
-
8:29 - 8:32should be something
which is not only a value - -
8:32 - 8:35it will never be 80 pieces,
-
8:35 - 8:38because we don't know whether
80 pieces will be sold tomorrow - -
8:38 - 8:42but a distribution
of all possible futures: -
8:42 - 8:4579, 80, 81, 82, and so on.
-
8:45 - 8:49The output is a probability
for each of the possible futures, -
8:49 - 8:50a probability distribution.
-
8:50 - 8:53If you have that for each
of the single cases, -
8:53 - 8:57then you can calculate
what the value of a decision is. -
8:57 - 8:58That's prescriptive analytics.
-
8:58 - 9:02Use the predictions to take decisions
on the basis of utility functions, -
9:02 - 9:05which are somehow known
or have to be found. -
9:08 - 9:12That is still a management decision:
What is a value of this or that? -
9:15 - 9:21We have to decide what is a value,
what we want to optimize. -
9:21 - 9:25Prescriptive applications give recipes,
prescriptions, for every single decision. -
9:26 - 9:31The recipe might be:
Take alternative two, buy 38 tomorrow. -
9:33 - 9:34For some very fine tasks,
-
9:34 - 9:37the mathematical and technical complexity
-
9:37 - 9:39of the whole process,
of prediction and decision, -
9:39 - 9:42can already be outsourced completely,
-
9:42 - 9:48and you can buy it in some cases
as a solution available in the Cloud. -
9:49 - 9:54With predictive analytics, we very often
can attain two mutually exclusive goals. -
9:54 - 9:56That's very interesting,
-
9:56 - 9:58because usually in management decisions,
you cannot do it. -
9:58 - 10:02For example, in purchase
order optimization, -
10:02 - 10:06you can have an increased availability
during the season, -
10:06 - 10:10and simultaneously reduce what remains
at the end of the season, -
10:11 - 10:13for example, in fashion retail.
-
10:13 - 10:18Or you can reduce the out-of-stock rate
in your meat shelves, -
10:18 - 10:21but simultaneously,
you can reduce the meat waste -
10:21 - 10:24at the end of the shelf lifetime.
-
10:24 - 10:29Or you can save 35%
of your marketing budget, -
10:29 - 10:32but still have more customers
to buy your products. -
10:33 - 10:39These are examples that are possible
with such technologies. -
10:39 - 10:43We have also automated
the work of scientists, -
10:43 - 10:48namely of my own science
as a particle physicist. -
10:48 - 10:52So we have seen that
usually a PhD student or researcher -
10:52 - 10:55has to do very similar things,
-
10:55 - 10:59and once you understand it,
you can try to optimize all this. -
10:59 - 11:03We have found that usually
a PhD thesis on particle physics -
11:03 - 11:06can be done by 72 decisions,
-
11:06 - 11:09and they all can be optimized.
-
11:09 - 11:14This was done for one
of the most successful experiments -
11:14 - 11:19in experimental particle physics,
the Belle experiment in Japan. -
11:19 - 11:23All the data having been
collected over ten years -
11:23 - 11:25have been reprocessed with such a program,
-
11:25 - 11:28and the output of it was twice as good
-
11:28 - 11:33as 400 physicists have done,
by hand together, in ten years. -
11:34 - 11:38That was the work of machine learning
or artificial intelligence, -
11:38 - 11:43and three PhD students corresponds
to about 500 normal PhD theses, -
11:43 - 11:46and corresponds to another
ten years of data taking, -
11:46 - 11:49which would have cost 700 million euros.
-
11:51 - 11:55Very often we have too many decisions
to be made by a human. -
11:55 - 11:58For example, 130,000 articles,
five productizes, -
11:58 - 12:03three storage locations, two predictions
40 days in advance, and so on to do, -
12:03 - 12:08and immediately you have some
100 million decisions. -
12:08 - 12:11so it's always good then to automate.
-
12:11 - 12:14Because even if you have
100 million recipes, -
12:14 - 12:17who's going to read all the recipes
and then decide what to do? -
12:17 - 12:22That is where we see that automation
is the ultimate goal of what we want. -
12:23 - 12:27The impact of decision automation
can, for example, be seen here: -
12:27 - 12:30So, this is the out-of-stock rate
at a German supermarket chain -
12:30 - 12:33at constant overall stock level
-
12:33 - 12:35as a function of time.
-
12:35 - 12:39So, the original out-of-stock level
was about 7%, -
12:39 - 12:45and then when the project started,
you see that black line there, -
12:45 - 12:49it got less, that's what we want,
it's good if it's small, -
12:49 - 12:53and then suddenly, one day,
we really automated it, -
12:53 - 12:55and no humans were allowed
to do anything anymore. -
12:55 - 12:59Then you see, the result
was much better, and constantly better. -
12:59 - 13:03So that's the good or bad thing.
-
13:03 - 13:05So, three examples at the end.
-
13:05 - 13:08Automatic fresh-meat replenishment
for supermarket chains -
13:08 - 13:10does work very well,
-
13:10 - 13:15and in some German supermarkets,
it is done automatically. -
13:15 - 13:17Also dynamic pricing:
-
13:18 - 13:22in retail, it's something
that is appearing more frequently. -
13:22 - 13:25With the Internet, it's normal already.
-
13:25 - 13:27It's also coming
in brick-and-mortar retail. -
13:28 - 13:30It's not bad even for the customer, right?
-
13:30 - 13:33Because a retailer can never fight
against the customer, -
13:33 - 13:37the customer has to want
to come and to buy, -
13:37 - 13:40so it does not mean
that the price is always higher; -
13:40 - 13:46it just means the right price,
at the right time, for the right article. -
13:46 - 13:52So most retailers do not fight against
the customers, but against competitors, -
13:53 - 13:57and the competitors also use
methods like these nowadays. -
13:59 - 14:04So finally, another example from research,
-
14:04 - 14:07from the collaboration
of Blue Yonder and KIT, -
14:07 - 14:11we have built such
a decision system onto one chip, -
14:11 - 14:15and this now holds
the world record in decisions, -
14:15 - 14:18it does eight billion
decisions per second, -
14:18 - 14:22and that is for the next generation
of particle accelerators. -
14:22 - 14:28So much data is produced in the sensors,
that we cannot read them out anymore. -
14:28 - 14:33So directly on the sensors,
there's a chip which decides: -
14:33 - 14:37"Okay, in this event,
this part of the response -
14:37 - 14:39seems to be important so we save it,
-
14:39 - 14:41we put it in a computer,
-
14:41 - 14:44and the others are not
even read out anymore." -
14:44 - 14:46So, these are things that are coming.
-
14:46 - 14:49And here, it's very clear,
here it's simply time, -
14:49 - 14:51no human would have the time
to make the decision, -
14:51 - 14:54it has to be extremely fast.
-
14:54 - 14:58What I'm talking about here is not
the far future, but it exists now, -
14:58 - 15:01and for sure, it will evolve further.
-
15:02 - 15:07So, my prediction is that
quite some repetitive routine work, -
15:07 - 15:08but also white-collar work,
-
15:08 - 15:13so it's not only the workers
whose jobs will change in the future, -
15:13 - 15:17but also, like me, white-collar work
will be automated. -
15:17 - 15:19Should we be afraid of it?
-
15:20 - 15:23I think not, it is normal innovation,
it makes us stronger, -
15:23 - 15:30it makes our strategic decisions
more efficient and also sustainable. -
15:30 - 15:33I think that's the way
towards a better future, -
15:33 - 15:37and I am completely convinced
that it is unavoidable. It will come. -
15:37 - 15:38We don't have to discuss it,
-
15:38 - 15:41it's only a question of time,
it will come, for sure. -
15:41 - 15:42Thank you very much.
-
15:42 - 15:44(Applause)
- Title:
- Decisions in the age of digitalization | Michael Feindt | TEDxKIT
- Description:
-
Decisions shape our lives and our world, from love to career to everyday business operations. Can technology help us make better decisions today?
As founder and Chief Scientific Advisor, Prof. Dr. Michael Feindt is the mind behind Blue Yonder – one of the world’s leading companies for predictive applications. During his long years of research at the world’s biggest elementary particle accelerators, he developed the NeuroBayes algorithm, which now serves for purely data-driven forecasts of probabilities of future events in many areas, and therefore as a basis for the automation of operational decisions in the fields of purchasing, distribution, research and development, manufacturing, and finance.
This talk was given at a TEDx event using the TED conference format but independently organized by a local community. Learn more at http://ted.com/tedx
- Video Language:
- English
- Team:
- closed TED
- Project:
- TEDxTalks
- Duration:
- 15:52
Robert Tucker edited English subtitles for Decisions in the age of digitalization | Michael Feindt | TEDxKIT | ||
Ellen approved English subtitles for Decisions in the age of digitalization | Michael Feindt | TEDxKIT | ||
Ellen edited English subtitles for Decisions in the age of digitalization | Michael Feindt | TEDxKIT | ||
Robert Tucker accepted English subtitles for Decisions in the age of digitalization | Michael Feindt | TEDxKIT | ||
Robert Tucker edited English subtitles for Decisions in the age of digitalization | Michael Feindt | TEDxKIT | ||
Robert Tucker edited English subtitles for Decisions in the age of digitalization | Michael Feindt | TEDxKIT | ||
Robert Tucker edited English subtitles for Decisions in the age of digitalization | Michael Feindt | TEDxKIT | ||
Robert Tucker edited English subtitles for Decisions in the age of digitalization | Michael Feindt | TEDxKIT |