Welfarist Control Theory

Walenstadt (Grid2050)

Large amounts of rooftop solar production.

In a sunny day, the grid cannot receive the entire production, and curtailment is necessary.

 

Curtailment = loss of return of investment

 

Who do we curtail?

  • Renewable energy should not be wasted.
     
  • The law requires to minimize total curtailment of green energy.
     
  • A free market would yield the same outcome.
\min_{x \in \mathcal X} \sum_i x_i

yields feasible grid state

  • The grid is a public good, paid by taxes.
     
  • Inequitable access worsens social disparity.
     
  • The law requires to treat all customers in the same way.
\min_{x \in \mathcal X}\max_i x_i
  • The grid is a public good, paid by bill surcharges.
     
  • Consumers are financially heterogeneous.
     
  • Uncertain return hinders investments.
\min_{x \in \mathcal X}\max_i \frac{x_i}{P_i}

Such a simple question...

Utilitarian

minmax

minmax

SOCIAL MANDATE

Operating a critical infrastructure

  • "right" to basic use
  • efficient use of infrastructure
  • "equal" access
  • market for premium service

ENGINEERS

Design control / optimization / decision rules that fulfil the mandate despite

  • unknown arrival of demand
  • risk of non-fulfilment
  • computational limits
  • privacy of users' preferences
  • large scale

Society

Engineers

?

NCCR applications

  • energy
  • mobility
  • "public commons"

Guesswork...

With further automation of decisions that are relevant for society, control engineers need to make increasingly decisive assumptions regarding what society wants.

 

Right now, some design choices may not be driven by responsible and ethical design, but rather by industry norms and academic conventions.

Example: the old faithful sum of costs, is widely used to pursue efficiency but often assumes higher levels of risk neutrality and interpersonal comparability than are permissible.

Example: allocation decisions, i.e., regulated access to any scarce resource like highway lanes, public transit, road space, water, energy, grid capacity, user attention, right to pollute, ...

SOCIAL MANDATE

Society entrusts control engineers to develop methods that meet specifications on

  • efficiency / performance
  • fairness
  • priorities / guarantees

ENGINEERS

Design control / optimization / decision rules that fulfil the mandate despite

  • dynamics and feedback
  • risk
  • heterogeneous systems
  • limited information
  • complexity

CERTIFICATES

Individuals and stakeholders can

  • access certificates
  • test fulfilment of mandate
  • obtain explanations
  • prequalify solutions ("grid code")

Society

Engineers

Theory for measurement of social mandate

Theory for certification

Session 1:

Social choice theory basics and Impossibility

Session 4:

Dynamic social choice theory

Session 2:

Who decides what is fair and how?

Society

Engineers

Theory for measurement of social mandate

Theory for certification

Session 3: 

Mechanism design and extracting preferences

Learnings from the course ...guaranteed ;) 

  • Learn a language to discuss social choice
  • Apply results from mechanism design

    Readings
    (untapped set of results that are relevant for control, optimization, engineering, etc., generally...)
     
  • Translate those concepts into the language of control and optimization

    Coursework
    (treading on the research frontier)

What is the ultimate Moonshot?

  • A principled framework to mine
    the social mandate
    for control applications
     
  • To identify the appropriate optimization problem
    (i.e. selection of social cost function)
     
  • Creating certificates guaranteeing alignment of mandate and practice
  • Evaluation of systems
  • Design of systems
  • Policy making

Format of a debate

Central paper: provides a common ground for the debate.
 

Together with the readings, we will state a motion/position, and divide the class into three teams:

Affirmative, negative and the audience

 

Preparatory coursework

The affirmative and negative teams will:

  • research the literature starting from our suggestions
  • translate the readings in the language of control, optimization, learning
  • collect evidence and examples in application domains
  • prepare constructive speeches

In class

Debate!
 

Example: Every automated decision should minimize a weighted sum of agents' costs.

 

  1. Teams will present prepared constructive speeches (first affirmative, then negative)
  2. Discussion within teams
  3. Rebuttals by the teams
  4. Closing, "vote/winner"
  5. Synthesis and consolidation by the audience