We’ve all seen the headlines on the rapid adoption of artificial intelligence (AI) across industries. From improved efficiencies in inventory management to new capabilities in vaccine development, AI has the power to revolutionize the way we work, interact and are entertained.
Less commonly discussed, however, is the importance of implementing an ethical AI supply chain.
Much like other large industries, AI can run into trouble when produced at scale. However, with estimates projecting the space to grow by 17.4% in 2021 alone, we’re at a pivotal point in the technology’s development. By taking the necessary steps to establish standard practices for its creation and implementation now, we can ensure AI is developed safely and used for good for years to come.
Let’s take a look at three ways you can ensure that impact is central to your AI business model.
If you’re familiar with AI, you know that having a diverse and robust data set is critical to its success. That’s why baking impact into your AI business model must start at the earliest stages of algorithm development and data creation.
Despite its increased implementation into the technologies and tools we utilize daily, eight out of 10 AI projects with insufficient training data fail. This is especially concerning when we remember that AI is being used to power critical instrumentation, including the cars we drive and our doctors’ tools to detect and treat diseases.
To ensure data sets are effective and bias-free, AI projects must create teams of annotation and labeling experts. At these early stages of algorithm development, companies have a unique opportunity to involve communities otherwise left out of the digital economy in a high-paying, technical industry. Rather than outsourcing this work abroad for lower-cost labor, AI teams can imbue their supply chain with impact through comprehensive training and employment programs. These programs support employees’ long-term growth and success while helping to break down the tech industry’s oppressive barriers to entry.
Internal Testing And Validation
Once you’ve created a reliable and impact-led training data team, you must validate your data set to check for remaining biases and errors.
In recent years, we’ve watched poorly trained AI fail across some of the most critical sectors of our society. In 2020, for example, we witnessed the wrongful larceny arrest of Robert Julian-Borchak Williams caused by a biased law enforcement facial recognition system in Michigan. To avoid negative outcomes like this and ensure the impact of your AI is positive, projects must receive both internal and external validations before implementation.
External validation requires data set contents to be shared publicly. Without it, internal biases can result in overlooked errors. Currently, very few projects take this crucial step. This was reflected in a study by STAT (subscription required): In looking at 161 AI products that the Food and Drug Administration (FDA) approved, only 73 disclosed in public documents the amount of patient data used in the validation of the product.
To ensure their technology will be used effectively, some companies have decided to undertake peer-reviewed studies to validate their AI. Google, for example, recently utilized this strategy for its dermatology tool.
Once AI is ethically sourced and validated, AI teams must invest in opportunities that not only improve their bottom line but also leverage their high-quality technology for good. Over the past several years, we’ve watched AI teams effectively develop impact-driven AI solutions to combat some of the most pressing issues in our global society.
Examples of this include: Project Guideline, which uses machine learning (ML) to enable visually-impaired people to run independently with the guidance of a cellphone and app (my company supported in creating training data for this initiative); Fake News Challenge, which uses competition to encourage the development of AI, ML and natural language processing tools to help with fake news identification; Allen Coral Atlas, which uses an AI-based tool to clean high-resolution satellite images so that scientists can monitor coral reefs; and the IBM-led initiative Green Horizons, which leverages big data processing, cognitive computing and the Internet of Things (IoT) in its pollution and weather forecasts to aid city planners.
Beyond its obvious technological benefits, if ethically created, validated and implemented, AI can be a powerful tool for good across our global society. To achieve this, companies must complete these steps to ensure their AI supply chain is ethical from inception to implementation.
What impact-driven AI initiatives are you most inspired by?