How to ensure ethical deployment of AI implementations?

The significant growth in automation and machine technologies (such as artificial intelligence and machine learning) has undoubtedly brought new scale and service levels to organizations.

One of the advantages we might all expect from AI is the opportunity to eliminate human-led biases and improve discrimination against minority groups. However, if not managed well, AI can further reinforce discrimination by embedding biases in its algorithms.

Today, machines often decide whether we qualify for a mortgage or are monitored by law enforcement agencies or insurance companies seeking to combat fraud. Their influence even extends to deciding which ads you see online - including job ads for high-paying jobs.

There are many organizations where artificial intelligence in automated systems is not well documented or understood. It’s time for automated decision-making to come out of the shadows and take responsibility.

When automated decisions directly or indirectly impact people’s lives, and machines may discriminate in harmful ways, organizations must stand up, take notice, and take action to ensure that AI is implemented as ethically as possible.

Step One

Businesses and government organizations alike should strive to obtain the highest level of protection from any machine technology they deploy. At the outset of any automation project, organizations must conduct a legal, privacy and ethical impact assessment to confirm that the risks are fully understood and can be satisfactorily mitigated. This also ensures that the most appropriate solution is selected to establish an acceptable level of risk while delivering value.

Sign-off on these evaluations should be conducted by a multidisciplinary, objective review team that has veto authority over any problematic aspects of the project, including deployment methods, levels of automation, and opportunities for recourse. Deployment must be a collaborative process between data/tech teams and business leadership teams to implement best practice ethics in data and analytics.

Deployment

The Ombudsman’s report outlines some strong recommendations on good practice in the design and implementation of machine technology. Nonetheless, we believe that all organizations have an obligation to consider at least the following best practices:

• Fairness, transparency, non-maleficence, privacy, respect for autonomy, and accountability Ethical considerations require any implementation of any machine Organizations with technology must ensure that they perform to the highest level of accuracy for all affected groups;
• Have a mechanism to explain any decisions based on the output of the model or system;
• Have detection and processes to mitigate harmful outcomes
• People can give informed consent to participate in the process
• There are mechanisms to challenge any outcomes that are considered unjust.

The development and deployment of any machine technology should be iterative, starting with an ethical review of accuracy based on historical data to ensure consistent performance across the sample population. If some groups perform significantly worse, more data must be sought to ensure adequate representation of all groups.

When risks of harmful consequences are identified, deployment should be equally iterative and cautious, starting with a human-in-the-loop solution to ensure human oversight while gaining confidence in the performance of the model or system.

This is not to say that the human decision-making process is foolproof. It simply provides an opportunity to understand and interrogate the output before deployment. This process should be done with the most trusted operator to reduce the possibility of reintroducing human bias into the process. Additionally, everyone involved in the process should have unconscious bias training.

Once in production, the ongoing accuracy and performance of any machine technology must be continuously measured and monitored. Along with existing KPIs, this performance should be reportable and visible across the organization.

Review

Any organization that implements algorithmic decision-making needs to have an objective ethical review process that includes both quantitative and qualitative considerations. Model performance should be monitored against these ethical metrics to understand performance anomalies for minority groups and any changes in performance over time. The model can then be adapted and adapted on an ongoing basis as part of the operational process.

While implementation may seem daunting, organizations must improve their understanding and implementation of ethical considerations in their artificial intelligence and machine learning projects. Businesses should adopt a “problem-review-measure-improve” approach to managing the performance and impact of their automated decisions to ensure ethical outcomes.

The above is the detailed content of How to ensure ethical deployment of AI implementations?. For more information, please follow other related articles on the PHP Chinese website!