COMP90041 - Final Project 1 - Solved

Moral Machines
The idea of Moral Machines is based on the Trolley Dilemma, a fictional scenario presenting a decision maker with a moral dilemma: choosing ”the lesser of two evils”. The scenario entails an autonomous car whose brakes fail at a pedestrian crossing. As it is too late to relinquish control to the car’s passengers, the car needs to make a decision based on the facts available about the situation. Figure 1 shows an example scenario. In this project, you will create an Ethical Engine, a program designed to explore different scenarios, build an algorithm to decide between the life of the car’s passengers vs. the life of the pedestrians, audit your decision-making algorithm through simulations, and allow users of your program to judge the outcomes themselves.

1         Build an Ethical Engine (15 points)
Your program should consist of seven core classes:

ethicalengine/ |-- Character.java

|-- Person.java

|-- Animal.java

|-- Scenario.java

|-- ScenarioGenerator.java

|__ Audit.java

EthicalEngine.java
You can create additional classes if needed. EthicalEngine.java contains the main function and coordinates the program flow. Scenario.java contains a list of passengers, a list of pedestrians, as well as additional scenario conditions, such as whether pedestrians are legally crossing at the traffic light. The decision-making algorithm is implemented as a static method with the name decide(Scenario scenario) in EthicalEngine.java. Start by implementing the classes Character.java, Person.java, Animal.java, and Scenario.java. SenarioGenerator.java is explained in detail in Section 2 and Audit.java in Section 3.

All classes except for EthicalEngine.java must be part of the package ethicalengine.

1.1         The Abstract Class Character
Character is an Abstract Class from which all character types inherit. This base class should be implemented as depicted in Figure 2. The class further comprises two enumeration types:

1.    Gender must include the types FEMALE and MALE as well as a default option UNKNOWN, but can also include more diverse options if you so choose.

 

Figure 1: Scenario example: a self-driving car approaches a pedestrian crossing but its breaks fail. Your algorithm needs to decide between two cases. Left: The car will continue ahead and drive through the crossing resulting in one elderly man, one pregnant woman, one boy, and one dog losing their lives. Right: The car will swerve and crash into a concrete barrier resulting in the death of its passengers: one women, one man, and one baby. Note that the pedestrians abide by the law as they are crossing on a green signal (image source: http://moralmachine.mit.edu/).

2.    BodyType includes the types AVERAGE, ATHLETIC, and OVERWEIGHT as well as a default option UNSPECIFIED.

The Character Class should implement the constructors as depicted in Figure 2. Make sure the empty constructor initializes all attributes with appropriate default values.

Age should be treated as a class invariant for which the following statement always yields true: age >= 0.

1.2          Classes Inheriting from Character.java
Create at least two concrete classes that directly inherit from the abstract class Character:

1.    Person.java: scenarios are inhabited by people who exhibit a number of characteristics (e.g., age, gender, body type, profession etc.). In the scenarios, each person is either considered to be a passenger or a pedestrian. A person can be you.

2.    Animal.java: animals are part of the environment we live in. People walk their pets so make sure your program accounts for these, at least for: cats and dogs.

1.2.1           The Class Person.java

This class represents a human in the scenarios. On top of its parent methods, the class Person must at least include the following public methods:

•    the constructor Person(int age, Profession profession, Gender gender, BodyType bodytype, boolean isPregnant).

•    the copy constructor Person(Person otherPerson).

 

Figure 2: UML Diagram for Character.java

•    getAgeCategory(): returns an enumeration value of the type AgeCategory depending on the person’s age with one of the following values:

–    BABY : a person with an age between 0 and 4.

–    CHILD: a person with an age between 5 and 16.

–    ADULT: a person with an age between 17 and 68.

–    SENIOR: a person with an age above 68.

•    the public method getProfession(): returns an enumeration value of the type Profession, which must include the following values: DOCTOR, CEO, CRIMINAL, HOMELESS, UNEMPLOYED, UNKNOWN. Only ADULTs have professions, other age categories should return the default value NONE. Additionally, you are tasked with coming up with at least two more categories you deem feasible.

•    the public method getProfession(): returns an enumeration value of the type Profession. Note that only adults have professions, other age groups should return NONE.

•    the public method isPregnant(): returns a boolean indicating whether the person is pregnant. For all instances of Person whose gender is not FEMALE this should return false.

•    the public method setPregnant(boolean pregnant): sets the value returned by isPregnant() while preventing invalid states, such as a pregnant male.

•    isYou(): returns a boolean indicating whether the person is representative of the user, e.g., you are one of the passengers in the car.

•    the public method setAsYou(boolean isYou): sets the value of whether the person is representative of the user.

•    the public method toString() must output a person’s characteristics according to the format shown below.

Pregnancy should be treated as a class invariant for which the following statement always yields true: if the person’s gender is not female, the person cannot be pregnant. Also, only persons who belong to the age category ADULT have a profession.

The public method toString() must return the following output format when printed to the commandline:

[you] <age category> [profession] <gender> [pregnant]
Note that attributes in brackets [] should only be shown if they apply, e.g., a baby does not have a profession so therefore the profession is not displayed. Here is an example:

adult doctor female
or

adult doctor female pregnant
Similarly, here is an example if the person is you:

you baby male
Note that words are in lowercase and separated by single spaces. Age and bodyType are ignored in the output.

1.2.2           The Class Animal.java

This class represents animals in the scenarios. On top of its parent methods, the class Animal must include the following public methods:

•    the constructor Animal(String species).

•    the copy constructor Animal(Animal otherAnimal).

•    the public method getSpecies(): returns a String indicating what type of species the animal represents.

•    the public method setSpecies(String species): sets the value returned by getSpecies().

•    the public method isPet(): returns a boolean value depending whether the animal is a pet or wild animal.

•    the public method toString() must output a pet’s characteristics according to the format shown below.

The public method toString() must return the following output format when printed to the commandline:

<species> [is pet]
Here is an example:

cat is pet
Here is another example where isPet() returns false:

bird
Note that words are in lowercase, separated by single spaces, and that gender, age, and bodyType are ignored in the output.

1.3         The Class Scenario.java
This class contains all relevant information about a presented scenario, including the car’s passengers and the pedestrians on the street as well as whether the pedestrians are crossing legally.

Each scenario can have only one instance of Person for which isYou() returns true. The following public methods must be implemented:

•    the constructor Scenario(Person[] passengers, Person[] pedestrians, boolean isLegalCrossing, boolean pedsInLane): you can use Arrays or ArrayLists in your class, but you need to make sure this constructor takes a person array as an argument.

•    the public method hasYouInCar(): returns a boolean indicating whether you (the user) is in the car.

•    the public method hasYouInLane(): returns a boolean indicating whether you (the user) are in the lane, i.e., crossing the street.

•    the public method getPassengers(): returns the cars’ passengers as a Person[] array.

•    the public method getPedestrians(): returns the pedestrians as a Person[] array.

•    the public method isLegalCrossing(): returns whether the pedestrians are legally crossing at the traffic light.

•    the public method setLegalCrossing(boolean isLegalCrossing): sets whether the pedestrians are legally crossing the street.

•    the public method getPassengerCount(): returns the number of passengers in the car (in int).

•    the public method getPedestrianCount(): returns the number of pedestrians on the street (in int).

•    the public method toString() must output the scenario according to the format shown below.

The public method toString() must return the following output format when printed to the commandline:

====================================== # Scenario

======================================

Legal Crossing: <yes/no>

Passengers (<getPassengerCount>) - <passenger.toString> .

.

Pedestrians (<getPedestrianCount) - <pedestrian.toString> .

.
Here is an example for a legal crossing (green light):

====================================== # Scenario

======================================

Legal Crossing: yes

Passengers (4)

-     cat is pet

-     overweight child male

-     average senior female

-     athletic adult female pregnant job: ceo

Pedestrians (3)

-     baby male

-     average adult male job: doctor

-     overweight adult female job: homeless
Here is another example with you in the car and a (non-pregnant) women and pedestrians crossing the street at a red light (illegal crossing):

======================================

# Scenario

======================================

Legal Crossing: no

Passengers (2)

-     you baby male

-     average adult female job: criminal

Pedestrians (2)

-     average senior male

-     average senior female
Note that character characteristics are written in lower case and separated by single spaces. Your output must match the output specifications.

1.4          The Class EthicalEngine.java
This class holds the main method and manages your program execution. It takes care of program parameters (see Section 4) as well as user input (see Section 5).

This class also houses the decide(scenario) method, which implements the decision-making algorithm outputting either PEDESTRIANS or PASSENGERS depending on whom to save. The code must choose whom to save for any scenario.

Decision Algorithm Your task is to implement the public static method decide(Scenario scenario) that either returns a value of the Enumeration type Decision, which is either PEDESTRIANS or PASSENGERS. Your code must choose whom to save for any scenario.

To make the decision, your algorithm needs to consider the characteristics of the characters involved as well as the situation. You can take any of the characters’ characteristics (age, bodyType, profession, pets, etc.) into account when making your decision, but you must base your decision on at least 5 characteristics–from the scenario itself (e.g., whether it’s a legal crossing) or from the characters’ attributes. Note that there is no right or wrong in how you design your algorithm. Execution is what matters here so make sure your code meets the technical specifications. But you may want to think about the consequences of your algorithmic design choices.

2        Scenario Generator (10 points)
The class ScenarioGenerator.java will be the basis of your simulation and shall be used to create a variety of scenarios. To guarantee a balanced set of scenarios, it is crucial to randomize as many elements as possible, including the number and characteristics of persons and animals involved in each scenario as well as the scenario itself.

To be able to properly test your scenarios and make sure your results can be replicated, you must apply pseudorandomness. Therefore, you need to familiarize yourself first with the class java.util.random [1] and especially with the function setSeed(long seed).

ScenarioGenerator.java must, therefore, include the following methods:

•    the empty constructor ScenarioGenerator(): this constructor should set the seed to a truly random number

•    the constructor ScenarioGenerator(long seed): this constructor sets the seed with a predefined value

•    the constructor ScenarioGenerator(long seed, int passengerCountMinimum, int passengerCountMaximum): this constructor sets the seed as well as the minimum and maximum number for both passengers and pedestrians with predefined values

•    the public method setPassengerCountMin(int min): sets the minimum number of car passengers for each scenario

•    the public method setPassengerCountMax(int max): sets the maximum number of car passengers for each scenario

•    the public method setPedestrianCountMin(int min): sets the minimum number of pedestrians for each scenario

•    the public method setPedestrianCountMax(int max): sets the maximum number of pedestrians for each scenario

•    the public method getRandomPerson() which returns a newly created instance of Person with random age, gender, bodyType, profession, and state of pregnancy

•    the public method getRandomAnimal() which returns a newly created instance of Animal with random age, gender, bodyType, species, and whether it is a pet or not

•    the public method generate() which returns a newly created instance of Scenario containing a random number of passengers and pedestrians with random characteristics as well as a randomly red or green light condition with you (the user) being either in the car, on the street, or absent.

The method generate() will need to abide by the minimum and maximum counts previously set for passengers and pedestrians in the scenario. If these values have not been explicitly set they need to be implicitly (i.e., by default) set to 1 and 5 respectively. A minimum may never be larger than its corresponding maximum.

3         Audit your Algorithm (10 points)
An audit is an inspection of your algorithm with the goal of revealing inherent biases that may be built in as an (un)intended consequence. In this task, you will simulate a variety of scenarios and have your EthicalEngine decide on their outcomes. The class Audit.java should:

1.    create a specific number of random scenarios,

2.    have your EthicalEngine decide on each outcome,

3.    and summarize the results for each characteristic in a so-called statistic of projected survival.

The following methods must, therefore, be implemented:

•    the empty constructor Audit()

•    the public method run(int runs): runs the simulation by creating N = runs scenarios and running each scenario through the EthicalEngine using its decide(Scenario scenario) method. For each scenario you need to save the outcome and add the result to your statistic

•    the public method setAuditType(String name): sets the name of the audit type. For example: Algorithm for an audit of your algorithm.

•    the public method getAuditType(): returns the name of the audit. Default should be Unspecified.

•    the public method toString(): returns a summary of the simulation in the format depicted below. If no simulation has been run, this method returns ”no audit available”.

•    the public method printStatistic(): prints the summary returned by the toString() method to the command-line.

3.1       Statistic of Projected Survival
Your statistic should list a number of factors, including:

•    age category

•    gender

•    body type

•    profession

•    pregnant

•    class type (person or animal)

•    species

•    pets

•    legality (red or green light)

Your statistic should account for each value of each respective characteristic, that are present in the given scenarios. For example, if you had scenarios with overweight body types, overweight must be listed in the statistic. If none of your scenarios included this particular body type, it must not be listed there. Also, make sure that you only update the statistic for, let’s say cats, if a cat was present in the tested scenario. If there is no cat in a given scenario, you must not change the % of cats that survived in your audit.

This is the output format (with pseudocode) of the statistic:

====================================== # <auditType> Audit

======================================

- % SAVED AFTER <int run> RUNS <for each characterstic:>

<characterstic>: <survival ratio>

-average age: <average>
Here is an example output:

====================================== # Algorithm Audit

======================================

- % SAVED AFTER 100 RUNS unemployed: 0.8 doctor: 0.79 ceo: 0.67 green: 0.67 senior: 0.66 pregnant: 0.66 female: 0.65 athletic: 0.64 overweight: 0.59 baby: 0.55 adult: 0.54 you: 0.54 unknown: 0.53 person: 0.5 criminal: 0.42
homeless: 0.38 male: 0.35 red: 0.34 child: 0.32 animal: 0.15 dog: 0.13 cat: 0.14 --

average age: 34.6
The list of characteristics must be sorted in descending order of the survival ratio. Note that the last two lines are not part of the sorted statistic but are at a fixed position in the output. The average age is calculated across all survivors of class Person (animals are excluded) and displayed with one digit after the decimal place (you can simply clip it). Your statistic must not list animals by gender, age, or body type.

3.1.1           Update your Statistic within an Audit

If you run multiple scenarios within a particular audit, make sure to update your statistic rather than overwrite it. For example, you may run an audit subsequently over 10 (audit.run(10)), 50 (audit.run(50)), and 100 (audit.run(100)) scenarios and print an updated statistic after each run to the command-line. The result on the command-line should be three statistic outputs: the first with 10, the second with 60, and the last with 160 runs.

3.2       Save your Audit Results
To save the results of your audit to a file, add the public method printToFile(String filepath) to your Audit class. The method prints the results of the toString() method to a target file named results.log. The filepath variable (e.g., ’logs/results.log’) includes both the target directory (logs/, in this case) and the filename (results.log). If results.log already exists in the target directory, you should append the new data rather than overwrite the existing file. If the file does not exist, your program should create it. If the directory specified by the filepath variable does not exist, your program should print the following error message to the command-line:

ERROR: could not print results. Target directory does not exist.
The results must be saved in ASCII code, i.e., human-readable.

4          Import Scenarios from a Configuration File (10 points)
Instead of generating scenarios solely randomly, you need to make sure in this task that your program can import scenarios from a data file. This will allow you to run audits on a consistent set of scenarios. In this task, you need to extend your EthicalEngine class to allow it to create scenarios based on data it reads from a configuration file.

4.1          Specify the Configuration File as Command-Line Argument
The config file should be specified when your program is launched. In this task, you need to create a command-line option. Command-line options or so-called flags specify options that modify the operation of your program. Options follow the program execution command on the command-line, separated by spaces. Options can be specified in any order. The following program calls are equivalent and should be supported by your program:

$ java EthicalEngine --config path/to/config.csv and

$ java EthicalEngine -c path/to/config.csv
The command line argument following the flag –config of -c respectively specifies the filepath where the configuration file (config.csv) is located. Your program should check whether the file is located at the specified location and handle a FileNotFoundException in case the file does not exist. In this case, your program should terminate with the following error message:

ERROR: could not find config file.
4.2       Parsing the Configuration File
Next, your program nees to read in the config file. Table 1 lists the contents of config.csv, a so-called comma-separated values (CSV) file. The file contains a list of values, each separated by a comma. As can be seen in Table 1, the first line contains the headers, i.e., the names (and description) of each data field and can therefore be ignored by your program. Each subsequent row presents an instance of Character. Scenarios are preceded by a single line that starts with scenario: and indicates whether the scenario depicts a legal (green) or illegal (red) crossing. In this case, the first scenario describes a legal crossing

scenario:green
with 3 passengers and 3 pedestrians (on of which is a dog). In fact, the first data set describes the scenario depicted in Figure 1. The second scenario describes an illegal crossing with 4 pedestrians and 2 car passengers.

Your EthicalEngine class needs to be able to read in a config file as depicted in Table 1 and create a Scenario instance for each scenario the file contains. Note that a config file can contain any number of scenarios with any number of passengers and pedestrians. You can assume that all config files follow the same format with the columns ordered as shown in Table 1.

4.3       Handle Invalid Data Rows
While reading in the config file line by line your program may encounter three types of exceptions, which your program should be able to handle:

1.    Invalid number of data fields per row: in case the number of values in one row is less than or exceeds10 values a InvalidDataFormatException should be thrown. Your program should handle such exceptions by issuing the warning statement ”WARNING: invalid data format in config file in line ¡linecount¿” to the command-line and skip the respective row then continue reading in the next line.

2.    Invalid data type: in case the value can not cast into an existing data type (e.g., a character where an int should be for age) a NumberFormatException should be thrown. Your program should handle such exceptions by issuing the warning statement ”WARNING: invalid number format in config file in line ¡linecount¿” to the command-line, assign a default value instead, and continue with the next value in that line.

3.    Invalid field values: in case your program does not accommodate a specific value (e.g., skinny as a bodyType) a InvalidCharacteristicException should be thrown. Your program should handle such exceptions by issuing a warning statement ”WARNING: invalid characteristic in config file in line < linecount >” to the command-line, assign a default value instead, and continue with the next value in that line.

Note that < linecount > depicts the line number in the config file where the error was found. While you can import the NumberFormatException from the package java.lang you will need to create custom exceptions for the other two.

 

Table 1: Contents of the configuration file (config.csv).

4.4           Audit your Algorithm Using the Scenarios from the Config File
Once your program has imported all scenarios from config.csv it should create a new Audit. Therefore, you need to extend your Audit class by adding two more methods:

•    the constructor Audit(Scenario[] scenarios): this constructor creates a new instance with a fixed set of scenarios.

•    the public method run(): runs the simulation with the scenarios specified and runs each scenario through the EthicalEngine using its decide(Scenario scenario) method.

Use the printToFile(String path) method to save your audit results to results.log.

5        Interactive Scenarios (10 points)
Now it is time to let the user take over and be the judge. Therefore, you need to build an interactive console program, which presents the user with a number of ethical scenarios. These scenarios can either be randomly generated or imported from a config file. For each scenario the user is asked to make a decision about who should survive. The results are logged to a user file (user.log) but only if the user consents to it.

5.1      Program Setup
As described in Section 4.1, we will use command-line options or so-called flags to initialize the execution of EthicalEngines. Therefore, you should add two more options as possible command-line arguments. Print Help Make sure your program provides a help documentation to tell users how to correctly call and execute your program. The help is a printout on the console telling users about each option that your program supports.

The following program call should invoke the help:

$ java EthicalEngine --help and

$ java EthicalEngine -h
The command-line output following the invocation of the help should look like this:

EthicalEngine - COMP90041 - Final Project

Usage: java EthicalEngine [arguments]

Arguments:

            -c or --config                            Optional: path to config file

           -h or --help                                Print Help (this message) and exit

             -r or --results                            Optional: path to results log file

-i or --interactive Optional: launches interactive mode
The help should be displayed when the –help or -h flag is set or if the –config flag is set without an argument (i.e., no path is provided). If the –config or -c flag is not set, your program should generate random scenarios.

The flag –interactive or -i indicates the interactive user mode. Without this flag the audit from Section 3 should kick in. Only if the –interactive or -i is set the program launches its interactive scenarios. The following command will launch the program with a config file in the interactive mode:

$ java EthicalEngine -i -c config.csv
Here is an example of launching the program in the interactive mode with random scenarios:

$ java EthicalEngine -i
5.2      Program Execution
You need to extend the EthicalEngine class to manage the user interaction and support the following program flow: show a welcome message, collect user consent for data collection, present 3 scenarios and have user judge these, show the statistic, and ask for another round of scenarios.

Show Welcome Screen At the start of the program, a welcome message must be shown: your program should read in and display the contents of welcome.ascii to the user without modifying it. The message provides background information about Moral Machines and walks the user through the program flow. Next, your program should collect the user’s consent before saving any results. Explicit consent is crucial to make sure users are aware of any type of data collection. Your program should, therefore, ask for explicit user consent before logging any user responses to a file. After the welcome message, you program should therefore prompt the user with the following method on the command-line:

Do you consent to have your decisions saved to a file? (yes/no)
Only if the user confirms (yes), your program should save the user statistic to results.log. If the user selects no your program should function normally but not write any of the users’ decisions to the file (it should still display the statistic on the command-line though). If the user types in anything other than yes or no, an InvalidInputException should be thrown and the user should be prompted again:

Invalid response. Do you consent to have your decisions saved to a file? (yes/no)
5.2.1        Present Scenarios

Once the user consented (or not), the scenario judging begins. Therefore, scenarios are either imported from the config file or (if the config file is not specified) randomly generated. Therefore, you should use the Audit class to keep track of the scenarios and decisions. Make sure to set the audit type to User using the method setAuditType(String name). Scenarios are presented one by one using the toString() method of the Scenario instance and printing its outputs to the command-line. Each scenario should be followed by a prompt saying:

Who should be saved? (passenger(s) [1] or pedestrian(s) [2])
Any of the following user inputs should be considered saving the passengers:

•    passenger

•    passengers

•    1

Any of the these user inputs should be considered saving the pedestrians:

•    pedestrian

•    pedestrians

•    2

After the user made a decision, the next scenario is shown followed by the prompt to judge the scenario. This procedure should repeat until 3 scenarios have been shown and judged. After the third scenario decision, the result statistic is presented.

5.2.2        Show the Statistic

The statistic must be printed to the command-line using the same method and format as described in Section 3.1. If the user previously consented to the data collection, the statistic is saved (i.e., appended) to the file user.log using the function printToFile(String filepath) of the Audit class. Additionally, the user should be prompted to either continue or quit the program as follows:

Would you like to continue? (yes/no)
Should the user choose no the program terminates. If the user decides to continue (yes), the next three scenarios should be shown. If the config file does not contain any more scenarios, the final statistic should be shown followed by the following prompt

That’s all. Press any key to quit.
Any following key press should terminate your program.

Here is an example of a statistic followed by a prompt to continue:

====================================== # User Audit

======================================

- % SAVED AFTER 9 RUNS pregnant: 0.81 doctor: 0.74 unemployed: 0.69 athletic: 0.69 green: 0.68 female: 0.66 ceo: 0.61 adult: 0.6 unknown: 0.59 person: 0.57 baby: 0.52 you: 0.51 overweight: 0.48 homeless: 0.45
criminal: 0.37 child: 0.36 male: 0.31 cat: 0.23 dog: 0.16 --

average age: 32.4

Would you like to continue? (yes/no)
And that’s it. Almost.

6        Documentation (5 points)
Always make sure to document your code in general. For this project you need to provide two types of documentation for your program: a UML diagram depicting your overall architecture and make sure to use JavaDoc syntax so that you can create an automatic Java code documentation in HTML. Only your UML diagram needs to be submitted through Canvas. You do not need to submit your documentation created through JavaDoc. This will be created automatically.

6.1       UML Diagram
Prepare a UML diagram describing your entire program containing all classes, their attributes (including modifiers), methods, associations, and dependencies. For each class you need to identify all its instance variables and methods (including modifiers) along with their corresponding data types and list of parameters. You should also identify relationships between classes, including associations, multiplicity, and dependencies. Static classes must be included in the UML. You can leave out any helper function that you added.

6.2     Javadoc
Make sure all your classes indicate their author and general description. For each constructor and method specified in the final project description you must provide at least the following tags:

•    @param tags

•    @returns tag

•    @throws tag

You can leave minor helper function that you may have added. Make sure to test the correct generation of your documentation using javadoc[2].

7        Reflection (Bonus Task)
This optional task gives you the opportunity to reflect on and describe the reasons you applied when designing your program’s decision-making as well as the consequences revealed by auditing your algorithm. For example, what were inherent biases you might have become aware by running this simulation? Were there any surprises? What are the consequences of design choices that you take as a programmer in general? Please make sure to stay below 250 words.

This part is optional but allows you to score an additional 2 points you may have lost somewhere else. Note that you cannot exceed the total of 60 points for the entire project. But we would love to read about your thoughts.









 
[1] https://docs.oracle.com/javase/8/docs/api/java/util/Random.html
[2] https://docs.oracle.com/javase/8/docs/technotes/tools/windows/javadoc.html