The Future With Artificial Intelligence
Artificial intelligence (AI) is shaping our future and becoming integrated into our lives, both at home and work. With it come many fears about what it means to jobs, humans and our future.
First, let’s put our minds at ease around the fear of tech run amok. We are incredibly far away from a Terminator chasing us. Now, let's focus on the idea that we can positively shape AI to our benefit.
Government’s Role in AI
Most world governmental bodies and politicians are not well-versed in how technologies work. The rules around what, how, why and when for AI are still being defined. This leaves it up to organizations creating AI solutions to draw a line in the sand for all of us. Many business leaders taking the time to understand how AI works are mindful of the power it holds. As AI becomes prevalent in more areas of our lives, guardrails about AI’s use must be established.
We are starting to see governmental engagement on this topic, such as with Singapore, the Organization for Economic Cooperation and Development (OECD) and the EU. The Partnership on AI, comprising 100 partners in 13 countries, has joined together to formulate AI best practices. These efforts aim to bring consistency in identifying areas of care for organizations and service providers.
The Role of Organizations in AI Governance
How do we go from best practices to governance within organizations? What exactly is AI governance, one may ask? AI governance should evaluate and monitor processes led by algorithms for effectiveness while identifying and reducing biases to reduce risk and ensure adequate return on investment for the organization.
Selecting appropriate best practices to implement with the burden of proper AI governance will fall onto the organizations that deploy AI solutions within their processes and functions. Organizations and employees will need to be mindful of incorporating AI governance solutions and make it part of their organizational DNA. Given the level of transformation anticipated with AI, the board of directors should prioritize change management, talent management, investments, results and the risks posed to an organization. The World Economic Forum assembled an excellent Oversight Toolkit to help boards of directors get familiar with various aspects of AI governance.
Governance is not an entirely new concept for organizations who will need to shoulder these responsibilities. However, achieving desired outcomes will require organizations to create a useful and ethical governance model. A well-defined governance framework will prevent AI solutions from amplifying biases while keeping checks and balances so that AI is not misused. There are other benefits to having an AI governance framework implemented. For example, it can help reduce redundant process steps that have been accepted as the norm, reimagine processes or functions altogether, and pivot to a refined organizational model.
Foundational Blocks for AI Governance
Let's look at the fundamental considerations organizations must keep in mind. As an example, Singapore’s approach focuses on:
- Internal governance structure and measurement
- Human involvement in AI-based decision-making
- Operations management
- Stakeholder engagement and communication
We can adapt our existing governance frameworks to measure the value, risk and responsibilities of AI-led decisions. A useful governance model will take into account both buy and build decisions on AI solutions. The process should typically begin before AI solutions are implemented and focus on continuous improvements after roll-out.
Internal Governance Structure and Measurement
When building an algorithm-based AI solution, data becomes the foundation where the availability, quality, breadth and training are critical. However, when buying a market-leveraged AI solution, measure the improved acceptance of AI-led decisions by humans and end-to-end process transformation.
We must track how we measure our existing processes and decision-making. If we can measure our existing processes, it will become easier to identify if an AI-led method is more effective. Suppose it is not; organizations should revisit the approach to determine areas of process inefficiencies. The root cause is usually an inefficient process and not the AI system. A correctly implemented automated process will inevitably be more efficient than a manual human-led one. The key will be to identify the source of friction and remove the barriers.
Human Involvement in AI-based Decision-making
Efficient measuring helps determine the level of human involvement in AI-augmented decision-making. It will make it easier to identify variations between the selection of decisions presented by AI and acceptance by knowledge workers. We should anticipate that initial iterations of AI solutions will have some data reconciliation needs. However, if the acceptance level continues to be lower than expected and does not trend upwards, reviewing decision-making for that particular task may help.
Amazon's recruiting AI application is an excellent example of exposing and amplifying biases in the selection process that did not favor hiring certain types of individuals. In fact, training data available for AI had preferences that led to the AI system to present biased choices. A useful governance model recognizes or questions flaws in existing processes, and the process should build in accountability and ownership to help correct them.
A solid governance approach in a recruiting application scenario empowers the governance model to ask the right questions from an operational management perspective. For example, how have we been making selection decisions? Can we recognize if there are biases that have now become our cultural norms? What are the individual, functional leader, team and organizational responsibilities? How can we correct the identified biases so the AI system does not amplify them?
In Amazon's case, human-led decision-making processes in HR and recruitment were left unchecked. They became the norm, which caused the AI system to amplify the biases. Smart organizations will incorporate checks and balances in the human-led decision-making process reviews.
Consider leveraging independent experts to make AI fit your organization's purpose. They can help clarify the effectiveness of your process compared to your peers and others in the industry. These experts can review the process with an objective eye and recommend changes to combat biases that otherwise might be prevalent in your industry or organization. Until organizational internal governance capabilities mature, reviews can help organizations manage risk and adaption.
Stakeholder Engagement and Communications
For organizations, gaining maturity in the internal governance framework should be a priority for both adoption and continuous use. To get there, stakeholder engagement and communications become a critical consideration. Functional leaders should engage with stakeholders affected by an AI-led process in advance of roll-out decisions.
A functional owner open to partnering to create a meaningful feedback loop and act on recommendations provided will gain trust and support in adapting an AI-supported approach.
Engaging stakeholders to understand which process steps are ineffective and which have room for improvement can help prioritize the team's activities. We should be pulsing if the stakeholders are merely informed or engaged adequately by the process owner. Assess if stakeholders and process owners are working collaboratively to help remove barriers.
Are governance decisions being documented and made accessible for review and auditing? Documenting the decision-making process around biases, ineffective processes, corrective steps and decision rationale will be useful to navigate future situations.
Governance focused on uncovering ineffective processes and areas for improvement will be useful to consider since the AI system presents decisions based on training data.
Think of approaches you already perform for corporate governance, outsourcing governance, and project or program governance. Apply those practices that already work within your organizational culture for AI-enabled processes. The goal isn’t to reinvent the wheel, but rather to leverage what we already know how to do well while adding skilled resources that can bridge the gap to connect the business knowledge and domain expertise while having a thorough understanding of how AI works.
Effective governance can make it easier for process owners to become self-aware and acknowledge and correct process biases before AI applications amplify them. It can also give organizations a set of consistent criteria to apply for compliance and reduce organizational risk.