sharon malzberg shared a presentation about artificial intelligence: tutorial for beginners

88
The 10 Principles of Applied Artificial Intelligence 1 The 10 Principles of Applied Artificial Intelligence How to Implement AI in Your Software Solution

Upload: sharon-malzberg

Post on 28-Jan-2018

210 views

Category:

Technology


1 download

TRANSCRIPT

The 10 Principles of Applied Artificial Intelligence 1

The 10 Principles of Applied Artificial IntelligenceHow to Implement AI in Your Software Solution

The 10 Principles of Applied Artificial Intelligence 2

Artificial intelligence (AI) is rapidly moving from the laboratory and into business and consumer applications.

The 10 Principles of Applied Artificial Intelligence 3

The result is a fundamental shift in how software is built, and what it’s capable of doing.

The 10 Principles of Applied Artificial Intelligence 4The 10 Principles of Applied Artificial Intelligence 4

The sort of AI we mean isn’t the artificial

general intelligence of science fiction.

The 10 Principles of Applied Artificial Intelligence 5

Instead, we’re talking about the specialized uses of artificial narrow intelligence that are already being used to power some of today’s most successful technology businesses.

The 10 Principles of Applied Artificial Intelligence 6

In this SlideShare, we’re going to take a look at the principles governing how that AI should be applied to your business.

The 10 Principles of Applied Artificial Intelligence 7The 10 Principles of Applied Artificial Intelligence 7

It’s important to point out that applied artificial intelligence

(applied AI) has a lot in common with applied analytics.

The 10 Principles of Applied Artificial Intelligence 8

That’s because in some ways applied analytics is a subset of applied AI.

The 10 Principles of Applied Artificial Intelligence 9The 10 Principles of Applied Artificial Intelligence 9

But our applied AI principles extend the

applied analytics model with…

The 10 Principles of Applied Artificial Intelligence 10

ü An increased focus on the full automation of business processes.

The 10 Principles of Applied Artificial Intelligence 11

ü The use of richer, rapidly evolving machine learning techniques.

The 10 Principles of Applied Artificial Intelligence 12

ü Better articulated performance objectives and model accuracy.

The 10 Principles of Applied Artificial Intelligence 13

ü Appropriate use of human judgement and automated predictions.

The 10 Principles of Applied Artificial Intelligence 14

ü Continuous learning and adaptation through feedback loops.

The 10 Principles of Applied Artificial Intelligence 15

ü Approaches to limit the impact of model error and bias.

The 10 Principles of Applied Artificial Intelligence 16The 10 Principles of Applied Artificial Intelligence 16

But the biggest difference of all may be in its potential for

business impact.

The 10 Principles of Applied Artificial Intelligence 17

While applied analytics helps optimize processes, applied AI has the long-term goal of total process automation.

The 10 Principles of Applied Artificial Intelligence 18The 10 Principles of Applied Artificial Intelligence 18

With that context in mind, let’s take a look

at the principles of applied AI.

The 10 Principles of Applied Artificial Intelligence 19The 10 Principles of Applied Artificial Intelligence 19

Understand all processes.1

The 10 Principles of Applied Artificial Intelligence 20

As with analytics, it’s important to start with the processes.

The 10 Principles of Applied Artificial Intelligence 21

That’s because applied AI has the potential to affect every process through augmentation or automation.

The 10 Principles of Applied Artificial Intelligence 22

So consider customer, partner and competitor processes, as well as those you control today.

The 10 Principles of Applied Artificial Intelligence 23The 10 Principles of Applied Artificial Intelligence 23

And even entirely new processes that

weren’t possible before.

The 10 Principles of Applied Artificial Intelligence 24

Document the decisions and actions for each.

The 10 Principles of Applied Artificial Intelligence 25

And understand which parts of the process are driven by humans, and which could be optimized, augmented, automated or re-invented.

The 10 Principles of Applied Artificial Intelligence 26The 10 Principles of Applied Artificial Intelligence 26

2 Prioritize the most valuable processes.

The 10 Principles of Applied Artificial Intelligence 27

Identify the processes where better outcomes or greater efficiency would have the highest impact.

The 10 Principles of Applied Artificial Intelligence 28

Think about how you might extend those processes further.

The 10 Principles of Applied Artificial Intelligence 29The 10 Principles of Applied Artificial Intelligence 29

And go beyond analytics.

The 10 Principles of Applied Artificial Intelligence 30

Shift your thinking from from injecting insights into existing processes…

The 10 Principles of Applied Artificial Intelligence 31

…to automating entire business processes.

The 10 Principles of Applied Artificial Intelligence 32

Start with the desired outcomes that would have the highest impact within your customers’ processes.

The 10 Principles of Applied Artificial Intelligence 33The 10 Principles of Applied Artificial Intelligence 33

Then identify where processes are short on instrumentation (data gaps) and integrations

(connectivity gaps).

The 10 Principles of Applied Artificial Intelligence 34The 10 Principles of Applied Artificial Intelligence 34

Design frictionless integrations.3

The 10 Principles of Applied Artificial Intelligence 35

Try to design solutions that are as automated, natural and complete as possible.

The 10 Principles of Applied Artificial Intelligence 36The 10 Principles of Applied Artificial Intelligence 36

To do so, take an outside-in approach.

The 10 Principles of Applied Artificial Intelligence 37

Think about how your AI-enabled process or solution will be integrated (and used).

The 10 Principles of Applied Artificial Intelligence 38

Then identify interdependencies with other processes such as...

The 10 Principles of Applied Artificial Intelligence 39

How other software processes (or humans) will access its capabilities.

The 10 Principles of Applied Artificial Intelligence 40

The more integrated your applied AI solution is, the greater the value.

The 10 Principles of Applied Artificial Intelligence 41The 10 Principles of Applied Artificial Intelligence 41

Integrate human judgement as required. 4

The 10 Principles of Applied Artificial Intelligence 42

Understand the roles humans play in your solution.

The 10 Principles of Applied Artificial Intelligence 43

Humans may be sources of data, trainers, reviewers or users.

The 10 Principles of Applied Artificial Intelligence 44The 10 Principles of Applied Artificial Intelligence 44

Having humans in the loop may be a user, market or regulatory requirement.

The 10 Principles of Applied Artificial Intelligence 45

So even if full automation is possible, human oversight may be necessary.

The 10 Principles of Applied Artificial Intelligence 46The 10 Principles of Applied Artificial Intelligence 46

Understand performance objectives. 5

The 10 Principles of Applied Artificial Intelligence 47

Know the model performance levels required to add business value.

The 10 Principles of Applied Artificial Intelligence 48The 10 Principles of Applied Artificial Intelligence 48

Model performance includes error rates, consistency, bias and explainability.

The 10 Principles of Applied Artificial Intelligence 49

Determine the cost of accuracy vs. the business impact of false predictions.

The 10 Principles of Applied Artificial Intelligence 50

Avoid models where the impact of false predictions is greater than the business can tolerate.

The 10 Principles of Applied Artificial Intelligence 51The 10 Principles of Applied Artificial Intelligence 51

Consider whether the explainability of the model

is a requirement or not.

The 10 Principles of Applied Artificial Intelligence 52

If it is, complex methods like deep learning may be unsuitable.

The 10 Principles of Applied Artificial Intelligence 53The 10 Principles of Applied Artificial Intelligence 53

Match modeling techniques to requirements. 6

The 10 Principles of Applied Artificial Intelligence 54

Machine learning techniques are changing fast.

The 10 Principles of Applied Artificial Intelligence 55The 10 Principles of Applied Artificial Intelligence 55

Ensure your team stays current.

The 10 Principles of Applied Artificial Intelligence 56

But don’t get carried away with the latest and greatest.

The 10 Principles of Applied Artificial Intelligence 57

Aim for the simplest model that can do the largest part of the job.

The 10 Principles of Applied Artificial Intelligence 58The 10 Principles of Applied Artificial Intelligence 58

Ensure you test model performance (against

requirements) and compare approaches.

The 10 Principles of Applied Artificial Intelligence 59

And have a strategic plan for evaluating new and more sophisticated models.

The 10 Principles of Applied Artificial Intelligence 60The 10 Principles of Applied Artificial Intelligence 60

Capture relevant data to support your models. 7

The 10 Principles of Applied Artificial Intelligence 61

Collect comprehensive, aggregated and diverse data from your target business processes.

The 10 Principles of Applied Artificial Intelligence 62The 10 Principles of Applied Artificial Intelligence 62

You will use this experience data

to train your models.

The 10 Principles of Applied Artificial Intelligence 63

Include input, context, action and outcome data from end-to-end processes in real time.

The 10 Principles of Applied Artificial Intelligence 64The 10 Principles of Applied Artificial Intelligence 64

And derive general and edge case experience data, using simulation techniques

if necessary.

The 10 Principles of Applied Artificial Intelligence 65The 10 Principles of Applied Artificial Intelligence 65

Manage quality for error and bias.8

The 10 Principles of Applied Artificial Intelligence 66

Machine learning is only fair and objective if it’s free of bias.

The 10 Principles of Applied Artificial Intelligence 67

Implement processes to monitor for bias and other errors such as inaccurate predictions.

The 10 Principles of Applied Artificial Intelligence 68The 10 Principles of Applied Artificial Intelligence 68

Bias may come from existing business

processes and their data.

The 10 Principles of Applied Artificial Intelligence 69

And remember that any human input into AI training or model evaluation predictions could be biased itself!

The 10 Principles of Applied Artificial Intelligence 70The 10 Principles of Applied Artificial Intelligence 70

Build fault tolerance as part of model integration.9

The 10 Principles of Applied Artificial Intelligence 71

Like all software, machine learning models will have errors, behave unexpectedly and sometimes even fail.

The 10 Principles of Applied Artificial Intelligence 72The 10 Principles of Applied Artificial Intelligence 72

And some machine learning models will

change over time with new data.

The 10 Principles of Applied Artificial Intelligence 73

Automation will also amplify the impact of errors.

The 10 Principles of Applied Artificial Intelligence 74The 10 Principles of Applied Artificial Intelligence 74

Traditional approaches to error detection and

elimination will be inadequate.

The 10 Principles of Applied Artificial Intelligence 75

So increase system dependency through fault tolerance.

The 10 Principles of Applied Artificial Intelligence 76The 10 Principles of Applied Artificial Intelligence 76

• Rule-based circuit breakers

• Human and AI-based monitoring and audits

• Check-and-balance redundancy

• Mandatory human review

• Kill switches

• Least-harm actions

Consider techniques like:

The 10 Principles of Applied Artificial Intelligence 77The 10 Principles of Applied Artificial Intelligence 77

Defend your legal and ethical stance.10

The 10 Principles of Applied Artificial Intelligence 78

Understand and articulate the overall benefits of artificial intelligence to your domain.

The 10 Principles of Applied Artificial Intelligence 79The 10 Principles of Applied Artificial Intelligence 79

Those might include:

• Efficiency gains

• Eliminating repetitive tasks

• Reducing risk of error

• Helping humans solve the unsolvable

The 10 Principles of Applied Artificial Intelligence 80

Frame ethical and legal discussions in the context of the overall benefits.

The 10 Principles of Applied Artificial Intelligence 81The 10 Principles of Applied Artificial Intelligence 81

And support your arguments with facts.

The 10 Principles of Applied Artificial Intelligence 82

Consider external factors such as legislation as well as customer sentiment.

The 10 Principles of Applied Artificial Intelligence 83

In fact, regulation may restrict automation levels to reduce risk and consequence.

The 10 Principles of Applied Artificial Intelligence 84

Don’t gloss over the replacement of human effort with AI automation.

The 10 Principles of Applied Artificial Intelligence 85The 10 Principles of Applied Artificial Intelligence 85

And be open and transparent.

The 10 Principles of Applied Artificial Intelligence 86

Applied AI is already fundamentally changing the nature of work and life.

The 10 Principles of Applied Artificial Intelligence 87The 10 Principles of Applied Artificial Intelligence 87

Understand all processes.

Prioritize the most valuable processes.

Design frictionless integrations.

Integrate human judgement as required.

Understand performance objectives.

4

7

8

9

1

2

3

Match modeling techniques to requirements.

Capture relevant data to support your models.

Manage quality for error and bias.

Build fault tolerance as part of model integration.

Defend your legal and ethical stance.

Big, long-term strategic changes lie ahead. To deal with them, remember these principles:

6

5

10

The 10 Principles of Applied Artificial Intelligence 88The 10 Principles of Applied Artificial Intelligence 88

Want to learn more about artificial intelligence?Visit georgianpartners.com