What is responsible AI and how to overcome implementation challenges

Organizations today do see the value of responsible AI. But there’s a big difference between knowing what’s right, and getting it done.

According to IBM, over 80% of orgs have not met or exceeded “their stated principles and values” when it comes to ethical AI.

Source: IBM

This, of course, gives the few that do a massive advantage, as responsible AI becomes more important for companies that are innovating technology today.

For organizations wanting to take the next steps towards responsible AI, now is definitely the right time. But first, let’s explore the basics of responsible AI and why it matters.

Definition and importance of responsible AI

To properly understand responsible AI, let’s begin with a definition, and then move on to a more in-depth discussion.

What is responsible AI?

Responsible AI is exactly as it sounds: using artificial intelligence in a responsible, legal, ethical way. For IT organizations, responsible AI ensures that system usage doesn’t lead to:

• Discrimination
• Injustice
• Breaking the law
• Physical or emotional damage
• Data and privacy breaches, among other things

Why is it important to consider ethics and responsibility in AI development?

AI is a very powerful technology that is spreading fast into almost every aspect of societal functioning. Like other technologies, it can contribute to a better world or it can become a dangerous tool and make existing problems worse.

Companies that develop AI in an ethical and responsible way are protecting their customers’ emotional, physical, and data safety. With this responsible approach, these organizations gain the trust and privilege to keep operating, growing, and contributing via AI to the betterment and development of our world.

What’s the impact of AI on society and the potential risks of irresponsible AI usage?

While the benefits of AI are enormous, so are the potential risks and downsides. When used irresponsibly, AI could:

• Exacerbate existing social inequalities. For example, AI could allocate better loans or medical care to certain groups of people over others, or give source credits to certain groups and not others, thus maintaining inequality.
• Put people at physical, emotional, or financial risk by exposing their data.
• Expose organizations to data breaches and the serious consequences, such as large fines, legal implications, and reputational damage.

Core principles of responsible AI

So what defines responsible AI? What are the guardrails that contrast ethical, responsible AI from risky and dangerous AI development? Let’s break down the core principles of responsible AI and what they mean.

Transparency

Transparency means ensuring that AI systems are understandable and their decision-making processes are explainable, including which data they’re using. This helps organizations understand their own actions in AI development, and the impact on customers, verifying ethical processes and assuring long-term business health.

Fairness

According to Deloitte’s Transparency and Responsibility in Artificial Intelligence report, bias can run deep — in statistics, data, and the way these are measured.

Source: Deloitte

This highlights the urgent need to create accountability for AI systems and the humans behind them.

Responsible AI works to prevent biases in AI algorithms and promote equality among users. In an interview with Atera, Sarah Bird, chief product officer of responsible AI at Microsoft, explored three levels of fairness:

• Providing equal service, no matter the customer’s language, accent or dialect.
• Allocating resources, like loans and medical care, without discrimination and bias.
• Avoiding stereotypes and demeaning content generation.

Accountability

Establishing clear accountability for AI actions and outcomes is a key aspect of responsible AI. This includes human supervision and setting processes for testing and adjusting the system throughout the development process.

Accountability also includes keeping the humans behind the ‘machine’ accountable — with training and stakeholder reviews, for example.

Privacy

Privacy is a leading concern in the IT industry, and also flows into the responsible AI process.  Alongside breach prevention, it’s important to verify that AI systems do not expose private information, or use private or breached data to serve users.

Safety and security

Finally, ensuring AI systems are safe, secure, and reliable is a core principle of responsible AI. It is concerned with privacy, data, and asset protection — but also with protecting the emotional and physical integrity of users.

For example, demeaning and triggering content, or breaching information that would make someone susceptible to emotional or physical abuse are less-talked about (but no less important) safety aspects of responsible AI.

Challenges in implementing responsible AI

AI systems are incredibly complex. Often, there’s a need for a cross-company effort to overcome some of the key challenges in leading responsible AI. The following challenges may hamper efforts to implement responsible AI, however, with some out-of-the-box thinking and strategizing, they can be overcome.

Identifying and mitigating biases in data and algorithms

In tackling the incredibly difficult challenge of bias in AI, Microsoft developed a 20-page document delineating hate speech guidelines. According to Bird, the company required an expert linguist to mitigate the document before developing the technology for Bing Chat. The AI technology now scores the content for hate speech and enables real time adjustments.

Yet, bias is so deeply rooted in humans (including those in charge of AI systems) that supervision upon supervision is likely required, plus feedback from users, to overcome it for truly responsible AI.

Balancing innovation with ethical considerations

Sometimes, companies are faced with the choice between innovation and responsible AI, and need to strike the right balance. There are endless ways this can arise, for example:

• Developing a technology that drives better medical outcomes, but requires access to sensitive patient or hospital information to be effective.
• Creating a software that lets every small business owner act as a designer, thus gaining a competitive advantage. But when the business owner asks for a photo of a woman, the instant result never reflects the diversity of the population.

In cases like these, responsible AI demands a deeper exploration about how to innovate AI to achieve an ethical outcome.

Ensuring compliance with regulatory standards and guidelines

This encompasses multiple aspects, including:

• Compliance in the massive data aggregation required to feed AI algorithms
• Keeping the data safe once it’s being used by your AI system
• Creating protection from system misuse by ill-intent players

In addition to these broader compliance issues, each organization must also prioritize industry-specific regulatory compliance to realize their responsible AI goals.

Strategies for responsible AI

We’ve covered a lot of theoretical ground about what responsible AI might look like. Now it’s time to take purposeful action towards responsible AI. Here are some strategies to guide you:

Adopt a human-centric approach to AI development

Consider what people need to make the most of your AI-based system — but also what could hurt their experience or wellbeing. Put humans in charge of system supervision, and provide training to help them overcome their own bias.

Implement robust data governance frameworks

This is a threefold effort:

• Determine guidelines for responsible AI, such as data privacy, transparency, and fighting bias in the system.
• Put technical controls in the system to prevent it from acting irresponsibly, like generating hate speech or sharing private data.
• Prioritize human supervision.

Monitor and evaluate AI systems regularly

Define real time alerts in your system to scale your monitoring for responsible AI. However, be sure to keep human technicians on top of your systems to stay on the extra-safe side.

Set evaluation goals and regular accountability reviews with managers, as well as processes for developing and testing fixes for what doesn’t work.

Engage cross-company teams and stakeholders

According to Sarah Bird, Microsoft’s chief product officer of responsible AI, when Microsoft started working with GPT 4 technology, it first brought together experts from different fields to get to know the system’s strengths and challenges. Cross-company collaboration can reduce blindspots and offer more creative, innovative solutions.

But for responsible AI to become an essential part of your operations, you need management stakeholders that can advocate for it at the C-suite level. Lee Hickin, Microsoft’s AI technology lead in Asia, told Atera that he’s already seeing “chief AI officers, chief AI risk officers [and] chief data ethics officers,” especially in highly sensitive industries.

Educate AI practitioners about ethical and safe AI practices

Even with the best intentions, humans are not immune to bias, so the AI systems they operate are not immune either. Additionally, many data breaches are due to human error — even among professionals working in areas like IT and marketing, where you might expect there would be increased awareness.

Source: Tessian

Therefore, educating anyone involved in AI operations at your company is key. Hickin suggests rewarding this education by putting employees’ new insights into real work problems. This will give a sense of meaning and encourage them to want to learn even more.

Responsible AI isn’t easy, but it’s worth the effort

Responsible AI helps develop a more equal world, with greater care for emotional, physical, and data safety. It also protects companies from losing sensitive information to data breaches, which comes with steep fines and causes serious damage to reputation and customer trust.

Done right, responsible AI can be one of your strongest pillars in the marketplace. Implement responsible AI with robust data governance frameworks, engage cross-company teams and stakeholders, monitor and evaluate AI systems regularly, and mostly — remember the human element.

Although AI is “artificial”, it was created by humans for humans. So in addition to the rules, technologies, and tactics, always put your people at the forefront of your AI development. This is the ultimate key to responsible AI.