Introducing Humans in AI
Agile Ethics for ethically al

Build more responsible AI systems.

download the manual

How it works

Use the HAI Trello board to tackle Agile Ethics in AI one step at a time.

HAI is an 8 step process. Each step includes to-do's which you can assign to your team. You can go in-depth and carry out extra research, or you can discuss each item for as little as 30 mins.

Aim to complete one full cycle DATA-SCAPE before releasing any AI-enabled product. To get started, you need access to:

Learn HAI from this demo board

The HAI process includes 8 core steps which can be adapted to your project. See it in action in this demo Trello board.

picture_as_pdf
Read the Handbook

With knowledge, comes great power - the power to create AI systems that are built on the right values. Read the HAI Handbook first.

Build your own board

Use this template to build your own HAI board. If you're stuck, give us a shout.

Agile Ethics in Trello

Use HAI Agile Ethics with your favorite software development framework.

Agile Ethics are compatible with Agile, Waterfalls and hybrid software development processes. Get your Agile coach, Scrum master or Product Owner to try it!

Build your own agile ethics board in trello

Why use Agile Ethics?

Job augmentation and job creation AI

 

Job augmentation

Re-imagine workbefore it's too late. Reduce technological unemployment and train an AI-ready workforce.

Job creation

 

AI ethics

HAI integrates IEEE's Ethically Aligned Design principles for AI and autonomous systems and the meticulous transparency analysis. Ethics

 

Technology adoption

Learn how to use the Technology Acceptance Model to establish trust in cognitive technologies.

adoption

Good AI in 8 steps

Agile Ethics is an 8 step process, short-coded DATA-SCAPE.

  • (S) Scope of project
    scope of project in AI
    STAGE 1 - (S) scope of project, skill mapping and ethical board

    Given a challenge, is AI the best solution for it? If yes, cover the following three areas with your team:

    • - Measure AI Readiness: attitudes, awareness and aversion towards AI
    • - Put together an ethical board to create an application-specific code of ethics, and to - ensure compliance with AI Ethical pxrinciples and Laws of Robotics
    • - Build skill maps for each job and measure degree of automation
    • - Find out what AI cannot do for your business, and what are the irreplaceable skills that should be nurtured and protected

    Your toolkit includes: skill mapping, meticulous transparency analysis, Principle of Education and Awareness (IEEE), AI Readiness test (AIR)

  • (DA) Data Audit
    data audit and data ethics AI
    STAGE 2 - (DA) Data audit and data ethics

    It doesn't matter how good an algorithm is, if it is trained on data sets that are not reflective of the real world.

    This is usually covered by the DevOps team, which performs data audit that includes an assessment of data sources, data quality, infrastructure, pipelines, data collection and instrumentation, compatibility, all from a technical perspective. With HAI, you also cover:

    • - Data ethics - data relevance and reliability, database completeness, database bias
    • - Data privacy and transparency
    • - Meticulous Transparency Analysis, ensuring that data is relevant to the problem

    Your toolkit includes: data ethics canvas, meticulous transparency analysis,

  • (T) Training
    data training
    STAGE 3 - (T) Training and intelligence editing

    With HAI you can turn a "black box" process into an intelligence editor.

    • Integrate the principles of intelligence editing and transparency into the training stage
    • Ensure algorithmic tractability and compliance with the Principle of Transparency

    Your toolkit includes: intelligence editing, Principle of Transparency, Law of Robotics no. 5 - accountability and traceability

  • (A) Analysis
    benchmark AI
    STAGE 4 - (A) Analysis and reliability

    This stage is when job augmentation starts, not after AI has been trained and deployed. 

    • - Benchmark AI performance and reliability against other AI systems, and against human-level performance
    • - Measure reliability levels, and start creating jobs that focus on AI supervision, training and re-calibration
    • - Measure compliancy bias and build levers to correct it
    • - Measure perceived usefulness of AI and trust for adoption purposes

    Your toolkit includes: compliancy bias testing, job re-writing toolkit, Law of robotics No.3 - safety of use

  • (*) Feedback
    STAGE 5 - (*)Feedback - stop, restart or go

    Estimate output of training against business KPIs, and compare with initial scope of project. In Feedback mode product owners can re-start the DATA cycle (Data Audit, Training, Analysis) or progress to the next stage.

    • Evaluate safety of AI system, and estimate the nature of delegation in the company - what actions should be supervised, approved by or performed only by humans?
    • Job re-allocation process: start up-skilling workers for AI training, supervision, calibration or delegation roles
    • Conduct Wizard of OZ experiments to find loopholes in the safety and delegation plan.

    Your toolkit includes: skill mapping, meticulous transparency analysis, Principle of Education and Awareness (IEEE), AI Readiness test (AIR), Wizard of OZ

  • (C) Calibrate
    STAGE 6 - (C) calibration and human-ai interaction

    Calibration is when AI becomes easy to use for humans, instead of human-like. Conduct design thinking exercises to understand what human-AI interaction should look like.

    • Follow design thinking practices to increase human-AI interaction and adoption
    • Wizard of OZ exercises to gauge what feels appropriate and psychologically safe to users - this is the beginning of creating a company's code of ethics for that AI application
    • Conduct Wizard of OZ experiments to find loopholes in the safety and delegation plan
    • Compliance with the 4th Law of Robotics - adequate disclosure (AI is an artefact, and it must not be designed human-like)

    Your toolkit includes: 4rd Law of Robotics, Wizard of Oz, design thinking, user testing

  • (A) Augment
    STAGE 7 - (A) Augmentation and job re-writing

    Augmentation through AI is about making hard technologies easy to use, and giving easy wins a hard re-think. Once moderately upskilled people focus using AI to create new business models, new jobs arise.

    • - Extract complex cognitive and social skills from the skills maps completed in the beginning
    • - Support further development of these creative, cognitive or social skills, post AI deployment
    • - Complete upskilling, training or onboarding program for users and workers
    • - Allow people working frequently with AI or in AI augmented jobs to experiment with new business models

    Your toolkit includes: skill mapping, workforce upskilling, job re-writing, user onboarding and training

  • (PE) People and Environment
    STAGE 8 - (pe) people and environment - job creation, impact on human wellbeing

    The last stage is for responsible businesses to evaluate the impact of AI on the company, workforce and people. Criteria for evaluation include: impact on human wellbeing, new jobs created or re-written, strategic and profit gains for the business, new business models or services discovered, innovation.

    • - Has the business gained a competitive advantage as a result of using AI
    • - Re-apply the AI Readiness test - is the adoption rate of AI higher than initial estimates - both for workers and end-users
    • - Has worked been transformed positively in terms of mitigating the risk of unintentional technological unemployment?
    • - Has exposure to easy to use AI led to innovative new services or business models?

    Your toolkit includes: AI Readiness test, Principle of Human Benefit, Job creation markers, IEEE's Wellbeing standards

  • S- (DATA) - CAPE
    AI design process in 8 stages - HAI
    (1-8) HAI Process - AI design process

    HAI is comprised on 8 stages, and at each stage the team is guided through applying AI ethics, AI adoption best practices, and helping people re-write or create new work models. Each stage includes a toolkit and set of activities, making it easier for AI to be ethical and beneficial by design.

    • (1) - (S) Scope of project 
    • (2) - (DA) Data Audit 
    • (3) - (T) Training
    • (4) - (A) Analysis 
    • (5) - (*) Feedback
    • (6) - (C) Calibrate 
    • (7) - (A) Augment
    • (8) - (PE) People and environment 

Want to build "good" AI?

Start with HAI - Agile Ethics process for AI.

Build your Agile EThics board

HAI for Agile Ethics is free to use, adapt and experiment with. We only ask that you credit method appropriately and link back to this resource.

Copyright © 2018. Intellectual property belongs to authors.