Millions of people engage with artificial intelligence (AI) in some form every day — most of them unwittingly. It’s at our fingertips through applications people from all over the world use regularly.
For example, Google search prompts, Facebook friend suggestions, autocorrect and predictive texts all run using AI technology. That’s why it’s more important than ever to launch, and sustain, a global effort to improve the transparency of AI in order to shed light on how the technology works, makes decisions and conducts critical applications for industries like healthcare and finance.
I think we can all agree that AI should not be above the law. But, at the same time, AI technologies shouldn’t be aggressively regulated. At least, not until a thorough analysis can be conducted to uncover whether new standards and regulations will serve to protect consumers or, in fact, limit progression. An analysis should focus on the current applications of AI, the potential economic impact of its deployment across new sectors and an assessment of public awareness of the technology.
Simultaneously, those of us in the global tech community developing AI need to jointly address its auditability and transparency. We should be committed to investing in the research and development of emerging AI technologies that many people don’t understand at an algorithmic level. Of course, industry, government and academic stakeholders all have a role to play in helping people understand when AI is most useful, why it improves productivity and, importantly, from where it gathers data to inform human interactions.
Opening the AI “black box”
The key to expanding AI transparency, knowledge and understanding has a few facets; let me explain.
Everyone interacting with AI should be aware that the person on the end of a chat function isn’t human and, likewise, the human shouldn’t pretend to be a bot. Companies utilizing AI technologies for customer and personnel interactions should be clear about what’s happening with the resulting data. People chatting with AI-driven platforms and systems should be able to keep records of their conversations — and site them when a discrepancy, issue or need for confirmation pops up.
People building AI for business and enterprise applications need to responsibly create, source and test diverse data. We need to introduce bias detection testing that identifies if the AI conforms to a standard and agreed testing protocol. Specifically, engineers need to simulate how data sets interact with users across a wide variety of contexts before AI leaves the test lab.
Engineers test products to ensure they aren’t harmful to humans. They conduct product testing for usability, security, scalability and safety to protect users, but we don’t currently test them for the harm they may do to humans socially, ethically or emotionally. Why don’t we as an industry add a bias test to the development cycle to ensure the algorithms in AI applications are not biased and are not causing harm that isn’t immediately or traditionally obvious to the user?
The AI industry should self-govern and establish ethics frameworks in lockstep with corporate boards and executive teams.
AI engineers also need to share best practices and bias elimination discoveries with their peers inside and outside of their organization. They need to ensure that the AI they’re building, and the data it uses, reflects the diversity of the people who use it to gain information. Most importantly, companies need to be able to make the difficult decision to halt progress when technical or coding issues arise that prevent AI from avoiding bias or delivering transparency. By ensuring the technology and data is diverse, objective and bug-free, there will be fewer issues with bias when people attempt to access that information.
The AI industry should self-govern and establish ethics frameworks in lockstep with corporate boards and executive teams. These should involve an opt-in code of principles, like the ethical AI guidelines we released at Sage for organizations developing and applying AI solutions. Working partnerships also should be set up between governments and businesses so they can share security-related information in real time. Achieving a true partnership like this advances transparency immensely and serves as an example for other jurisdictions considering a move to the ‘AI for commercial development, public service and social good’ approach.
Companies considering AI should conduct internal reviews to understand where AI can be applied and how many people need training in order to implement AI-driven services. They should create training and workforce acceleration programs that encourage people with AI-ready development skills to pursue tech-driven careers.
Achieving transparent AI is a global effort
Fundamentally, the tech community needs to define what AI transparency means and work together to apply transparency to AI innovation. We need to stop treating AI as a black box and address the auditability and traceability issues that will lead us down the right path. We need to educate people about AI and illuminate its numerous use cases — from technology to healthcare to transportation to security to home life.
The current lack of AI transparency and education inherently causes mistrust of this important technology. If we can conquer this issue as an industry, we will be able to truly deliver democratized access to information through tech like AI that will bind us as a global tech and user community.
The universal transparency we need to strive for has to happen at the algorithm and data levels. The global AI community needs to work together to stem, then eliminate, bias entirely through proper testing, training and education — something I’ve committed my career to doing. And I’m not alone in this pursuit of building transparent, ethical AI thanks to initiatives like the U.K. Parliament’s Select Committee on AI, NYU’s AI Now Institute and the MIT Media Lab’s AI ethics and governance initiative. There’s a lot more work that needs to be done. But achieving transparent AI will be well worth it.