When not to use Machine Learning?

• If clear specifications are available
• Simple heuristics are good enough
• Cost of building and maintaining the system outweighs the benefits (see technical debt paper)
• Correctness is of utmost importance
• Only use ML for the hype, to attract funding

Examples?

Consider Non-ML Baselines

• Consider simple heuristics -- how far can you get?
• Consider semi-manual approaches -- cost and benefit?
• Consider the system without that feature
• Discuss Examples
  • Ranking apps, recommending products
  • Filtering spam or malicious advertisement
  • Creating subtitles for conference videos
  • Summarizing soccer games
  • Controlling a washing machine

When to use Machine Learning

• Big problems: many inputs, massive scale
• Open-ended problems: no single solution, incremental improvements, continue to grow
• Time-changing problems: adapting to constant change, learn with users
• Intrinsically hard problems: unclear rules, heuristics perform poorly

Examples?

When to use Machine Learning

• Partial system is viable and interesting: mistakes are acceptable or mitigatable, benefits outweigh costs
• Data for continuous improvement is available: telemetry design
• Predictions can have an influence on system objectives: systems act, recommendations, ideally measurable influence
• Cost effective: cheaper than other approaches, meaningful benefits

Examples?

Discussion: Spotify

Big problem? Open ended? Time changing? Hard? Partial system viable? Data continuously available? Influence objectives? Cost effective?

AI as Prediction Machines

The economic lense

• predictions are critical input to decision making (not necessarily full automation)
• decreased price in predictions makes them more attractive for more tasks
• increases the value of data and data science experts
• decreases the value of human prediction and other substitutes
• decreased cost and increased accuracy in prediction can fundamentally change business strategies and transform organizations
  • e.g., a shop sending predicted products without asking
• use of (cheaper, more) predictions can be distinct economic advantage

Predicting the Best Route

Predictions vs Judgement

Automation with predictions

• Automated predictions scale much better than human ones
• Automating prediction vs predict judgement
• Value from full and partial automation, even with humans still required
• Highest return with full automation
  • Tasks already mostly automated, except predictions (e.g. mining)
  • Increased speed through automation (e.g., autonomous driving)
  • Reduction in wait time (e.g., space exploration)
• Liability concerns may require human involvement

Automation in Controlled Environments

The Cost and Value of Data

• (1) Data for training, (2) input data for decisions, (3) telemetry data for continued improving
• Collecting and storing data can be costly (direct and indirect costs, including reputation/privacy)
• Diminishing returns of data: at some point, even more data has limited benefits
• Return on investment: investment in data vs improvement in prediction accuracy
• May need constant access to data to update models

Where to use AI?

• Decompose tasks to identify the use of (or potential use of) predictions
• Estimate the benefit of better/cheaper predictions
• Specify exact prediction task: goals/objectives, data
• Seek automation opportunities, analyze effects on jobs (augmentation, automate steps, shift skills, see taxis)
• Focus on steps with highest return on investment

Cost Per Prediction

What contributes to the average cost of a single prediction?

Examples: Credit card fraud detection, product recommendations on Amazon

Cost Per Prediction

• Useful conceptual measure, factoring in all costs
  • Development cost
  • Data aquisition
  • Learning cost, retraining cost
  • Operating cost
  • Debugging and service cost
  • Possibly: Cost of deadling with incorrect prediction consequences (support, manual interventions, liability)
  • ...

AI Risks

• Discrimination and thus liability
• Creating false confidence when predictions are poor
• Risk of overall system failure, failure to adjust
• Leaking of intellectual property
• Vulnerable to attacks if learning data, inputs, or telemetry can be influenced
• Societal risks
  • Focus on few big players (economies of scale), monopolization, inequality
  • Prediction accuracy vs privacy

Organizational Objectives

Innate/overall goals of the organization

• Business
  • current revenue, profit
  • future revenue, profit
  • reduce risk
• Non-Profits
  • Lives saved, animal welfare increased
  • CO2 reduced, fires averted
  • Social justice improved, well-being elevated, fairness improved

accurate models themselves are not a goal

AI may only very indirectly influence such organizational objectives, influence hard to quantify, lagging measures

Breaking Down Processes

Break overall goals along processes

• Break workflow into tasks
• Identify decisions in tasks
• Evaluate benefit of AI for prediction or automation
• Evaluate the influence of improving some tasks on process

Maintain mapping from task-specific goals to system goals

Leading Indicators

Measures correlating with future success, from the business' perspective

• Customers sentiment, customers liking the products (e.g., through surveys, rating)
• Customer engagement, regular use, time spent on site, messages posted
• Growing user numbers, recommendations

indirect proxy measures, lagging, bias

can be misleading (more daily active users => higher profits?)

User Outcomes

How well the system is serving its users, from the user's perspective

• Users receive meaningful recommendations, enjoying content
• Users making better decisions
• Users saving time due to system
• Users achieving their goals

easier and more granular to measure, but only indirect relation to organization objectives

Model Properties

Quality of the model used in a system, from the model's perspective

• Model accuracy
• Rate and kinds of mistakes
• Successful user interactions
• Inference time
• Training cost

not directly linked to business goals

What is Measurement?

• Measurement is the empirical, objective assignment of numbers, according to a rule derived from a model or theory, to attributes of objects or events with the intent of describing them. – Craner, Bond, “Software Engineering Metrics: What Do They Measure and How Do We Know?"

• A quantitatively expressed reduction of uncertainty based on one or more observations. – Hubbard, “How to Measure Anything …"

Everything is Measurable

• If X is something we care about, then X, by definition, must be detectable.
  • How could we care about things like “quality,” “risk,” “security,” or “public image” if these things were totally undetectable, directly or indirectly?
  • If we have reason to care about some unknown quantity, it is because we think it corresponds to desirable or undesirable results in some way.
• If X is detectable, then it must be detectable in some amount.
  • If you can observe a thing at all, you can observe more of it or less of it
• If we can observe it in some amount, then it must be measurable.

But: Not every measure is precise, not every measure is cost effective

On Terminology:

• Quantification is turning observations into numbers
• Metric and measure refer a method or standard format for measuring something (e.g., number of mistakes per hour)
  • Metric and measure synonmous for our purposes (some distinguish metrics as derived from multiple measures, or metrics to be standardizes measures)
• Operationalization is identifying and implementing a method to measure some factor (e.g., identifying mistakes from telemetry log file)

The Maintainability Index

$max(0, (171 - 5.2 ln(HV) - 0.23 CC - 16.2 ln(LOC)*100/171))$ $<10$: low maintainability, $\leq 20$ high maintainability

Decomposition of Measures

Often higher-level measures are composed from lower level measures

clear trace from specific low-level measurements to high-level metric

Specifying Metrics

• Always be precise about metrics
  • "measure accuracy" -> "evaluate accuracy with MAPE"
  • "evaluate test quality" -> "measure branch coverage with Jacoco"
  • "measure execution time" -> "average and 90%-quantile response time for REST-API x under normal load"
  • "assess developer skills" -> "measure average lines of code produced per day and number of bugs reported on code produced by that developer"
  • "measure customer happyness" -> "report response rate and average customer rating on survey shown to 2% of all customers (randomly selected)"
• Ideally: An independent party should be able to independently set up infrastructure to measure outcomes

Correlation vs Causation

Correlation vs Causation

• In general, ML learns correlation, not causation
  • (exception: Bayesian networks, certain symbolic AI methods)
• To establish causality:
  • Develop a theory ("X causes Y") based on domain knowledge & independent data
  • Identify relevant variables
  • Design a controlled experiment & show correlation
  • Demonstrate ability to predict new cases

Confounding Variables

Confounding Variables

• To identify spurious correlations between X and Y:
  • Identify potential confounding variables
  • Control for those variables during measurement (randomize, fix, or measure + account for during analysis)
• Examples
  • Drink coffee => Pancreatic cancer?
  • Degree from high-ranked schools => Higher-paying jobs?

Challenge: The streetlight effect

• A type of observational bias
• People tend to look for something where it’s easiest to do so
  • Use cheap proxy metrics, that only poorly correlate with goal?

Risks of Metrics as Incentives

• Metrics-driven incentives can:
  • Extinguish intrinsic motivation
  • Diminish performance
  • Encourage cheating, shortcuts, and unethical behavior
  • Become addictive
  • Foster short-term thinking
• Often, different stakeholders have different incentives!

Make sure data scientists and software engineers share goals and success measures

On Incentives: University Rankings

Measurement Validity

• Construct: Are we measuring what we intended to measure?
  • Does the abstract concept match the specific scale/measurement used?
  • e.g., IQ: What is it actually measuring?
  • Other examples: Pain, language proficiency, personality...
• Predictive: The extent to which the measurement can be used to explain some other characteristic of the entity being measured
  • e.g., Higher SAT scores => higher academic excellence?
• External validity: Concerns the generalization of the findings to contexts and environments, other than the one studied
  • e.g., Drug effectiveness on test group: Does it hold over the general public?

Successful Measurement Program

• Set solid measurement objectives and plans
• Make measurement part of the process
• Gain a thorough understanding of measurement
• Focus on cultural issues
• Create a safe environment to collect and report true data
• Cultivate a predisposition to change
• Develop a complementary suite of measures

Key Challenge: Telemetry

• Some goals can be quantified easily, others not so much
• Be creative in data sources (telemetry)
  • Existing logs and measures
  • Logging additional information
  • User surveys, feedback mechanisms
• Often only proxy measures, sometimes delayed, often only samples

How to Measure System Success?