How to be an Ethical AI Company (or: trusting the robots with your baby)

So, you’re not a Silicon Valley “AI First” tech company? Your products or services are in other markets, but you recognise that AI can revolutionise your business? You’re also aware of the ethical dilemmas? You want to use AI, but want to get it right. But how?

The pace of technological change, and the required agility to respond, is pressuring the capability of companies to remain true to their ethics. This velocity of innovation, combined with the power it hands to engineers to create undesirable outcomes, has spurred thinking on how an organisation can harness AI tech without risking brand, or worse. From the need for a “Chief Ethics Officer” to commanding the entire organisation to have an “ethical” mindset there is no lack of recommendations, but little to guide companies on practical steps to achieve these goals. In short, there’s a lot of people saying how a company should be, but few describing how they can get there.

So here’s my quick guide: 5 steps to becoming an Ethical AI Company.

Step 1: Learn how to do ethical AI inside the company first.

Forget reshaping the product development process for the time being. Telling your workforce you want them to behave ethically in the design of AI-based services will achieve nothing if your approach to internal business processes is to apply every Machine Learning tool under the sun to scrutinise, optimise and automate with little concern for addressing the ethical considerations.

Develop an internal transformation approach that addresses data bias, diversity in representation, transparency of automated decisions, right to redress, etc. as part of every change project you run. Learn how to execute with internal stakeholders before tackling external. Setting expectations on addressing ethics in forming and shaping the business will make the leap to doing so for products much easier. This will become the culture of the organisation, or “how we do things”.

Step 2: Ramp up your learning capabilities

Chances are there is no “future state” that can be reached where everything stays static and predictable, so forming training around one-off courses to align people to business processes won’t work. Your staff will need agility in addressing ethical concerns during rapidly changing scenarios. Their ability to quickly skill-up on a particular area will be critical to their performance, and the success of your organisation.

Streamline the path to getting trained, remove barriers like strict training budgets, make it easy to access training. Remove any reference to training being a “benefit” in hiring ads, performance reviews, etc. Set the bar high and require staff to learn something new every week. Make researching, learning and discovering part of the daily routine. Re-skill your training team towards being curators of skills, not producers of training materials.

Step 3: Set ethics in scorecards and policy

Set ethics goals in the general business principles with a mandate that these cascade into local policy. Start with realistic ambitions, but be ambitious. Optionally, set a separate code of ethics for automated systems. Set policies that enable people to invest time and budget in acting ethically in the development and use of automated systems. Announce these frequently and repeatedly. Commit to reporting on progress against these policies.

Although a policy on its own won’t effect much change, policy backed by active reinforcement, modelling and communication of behaviours can have a massive impact. Conversely, initiatives which appear contrary to what the company broadcasts as important are hindered from the outset.

Step 4: Empower your leadership, individually and as a group

Set targets for all your leadership team, with strong reinforcement on your HR, IT, product development and procurement leadership. It’s unlikely that individual action by each business area will coordinate into a successful whole, so set a formal role of “Chief Ethics Officer” detached from the legal and audit teams to prevent the role being seen as a legal compliance initiative.

Step 5: Personally sponsor a transformation programme

Take personal accountability of a programme to re-orientate your organisation to deliver AI in an ethical manner. Avoid 2–3 year goal statements with outcomes far enough away to discourage attention. Set goals at 6 months, make these ambitious. Set vision at 5 years, make this aspirational. Make the approach iterative and agile. Concentrate on:

• Introducing ethics into training programmes for all employees
• Ensuring a whistleblower facility is in place and well supported
• Including ethical handling in internal transformation, employee management and internal systems.
• Including ethical consideration controls/phases in R&D, product development and product support processes. Use the business principles as anchors, but challenge the teams themselves to find what works.
• Including review of ethical adherence into your Internal Audit remit.
• Adopting quality standards supporting ethical use of AI.

Benchmark at the outset, and audit on a regular basis with published results. Deliver the transformation itself according to the desired behaviours: ensure diversity, review ethical challenges in its delivery, set a clear tone that ethics are everyone’s responsibility.

Ensure other “in-flight” projects that are forming the organisational design for an AI enabled company include ethical consideration as a priority to prevent mixed messages and conflicting goals.

Pitfalls to avoid

• Avoid neglecting internal use of AI to concentrate on product development. Considering ethics in product design will require a behavioural/mindset change for many engineers. This will be easier if they are already living and breathing this behaviour internally.
• Avoid making this one team’s problem. The change is one that cuts across the organisation.
• Avoid goals for 2–3 years away. An organisation will achieve more in 6 months than they will in 2 years.
• Avoid getting lost in academic discussions. Set clear, achievable goals to improve things and commit to come back and re-examine after these are achieved. Stay agile.