This post is the result of the reflection I’ve had reading five McKinsey articles that explore artificial intelligence (AI), leadership and their intersections in a rapidly changing world.

Building Trust in AI through Explainability

The article “Building AI trust: The key role of explainability” argues that trust is crucial for adopting AI. The lack of transparency in AI systems, i.e. the lack of clear insight into how algorithms work or how they are fed, raises concerns about possibly generating inaccurate and biased results, undermining trust.

I had already analyzed this phenomenon in my last post “AI and Leadership (powered by HBS)“, and is one of the explanations for the phenomenon called “algorithm aversion”.

Explainability of AI (XAI) thus emerges as an imperative for building trust with humans by making AI decision-making processes understandable to users. Trust in AI comes from understanding the results of AI-powered software and how – at least at a high level – they are created.

The article proposes a human-centered approach to AI explainability, adapting explanations to the needs of different stakeholders, such as executives, AI governance leaders, affected users, business users, regulators/auditors and programmers.

Various XAI techniques are presented, such as post-hoc methods (analyzing models after training) and ante-hoc methods (intrinsically explainable models, such as decision trees), and global explanations (understanding decision-making in all cases) and local explanations (focusing on specific decisions).

More than a regulatory issue (which can and should be provided for in AI regulation), XAI is going to be a predominantly market issue. Humans will want proof of the reliability of AI, however convenient it may be.

The Rise of AI and the Role of Leadership in the Age of Cognition

In the article “Direct from Michael Dell: Leadership lessons and the future of AI“, Michael Dell, CEO of Dell Technologies, shares his vision of the transformation driven by AI. Dell argues that AI is driving a shift from computing to cognition, unlocking the power of data.

“About 60 years ago, we started this journey of calculating computers, and basically that’s what we’ve been doing ever since. Now, with advances in AI and generative AI, we’re moving towards cognition and intelligence, and unlocking the power of data with all this innovation.”

Dell highlights the importance of managing data effectively, considering that most data is created and used in the physical world.

Dell compares this transition to a point on Schumpeter’s creative destruction curve, but with exponential acceleration. The ability to reason using all the world’s knowledge on any subject, synthesizing it into results and new understandings, will drive scientific discovery, productivity and human well-being.

Dell envisions a future where AI will make humans happier, healthier and more successful, improving everything we set out to accomplish. His conviction is based on the growing importance of data in all aspects of modern life. 

Whether it’s the development of autonomous vehicles, the discovery of medicines or the creation of new business models, everything interesting and innovative in the world evolves around data. For Dell, humanity’s most complex problems require greater computing power and more data, and it is in this context that AI plays a crucial role.

Although this may seem like an overly optimistic outlook, I can’t help but see myself in it, since the adoption of AI in our work context has allowed us to create more and more, with exponentially richer results and leading to what I call “exponentially rewarding work”.

However, this does not mean that we have to be careful when adopting AI...

Implementing Generative AI with Speed and Safety

The article “Implementing generative AI with speed and safety” explores the risks and opportunities of generative AI (gen AI). McKinsey estimates that gen AI could add up to 4.4 trillion dollars to the global economy. However, implementing gen AI requires a responsible approach to mitigate risks such as inaccurate results, biases in training data, misinformation and the potential for malicious use.

“While many business leaders are determined to capture this value, there is growing recognition that the opportunities of gen AI are accompanied by significant risks.” As a professor of mine wisely said more than two decades ago, “the problem is not with the tools, but with how we use them”.

The article describes various risks:

• Accuracy and Reliability: Generative AI can produce inaccurate or “hallucinatory” results, especially when trained with incomplete or biased data. It is essential to implement validation mechanisms to guarantee the reliability of the results. This is a risk I’ve experienced “on the fly” in a mega-project I’m working on, in which AI is helping us to draw up pedagogical roadmaps for e-learning solutions. Training an algorithm requires care and patience, but, above all, teamwork between subject specialists and pedagogical specialists, in this case. The critical scrutiny of AI outputs is a job that requires human intervention.

• Bias and Discrimination: Generative AI models can perpetuate and amplify the biases present in the training data. This can lead to discriminatory results and undermine justice and fairness. Yet another moral imperative that must be safeguarded and audited by humans.

• Data Privacy and Security: Generative AI can put data privacy and security at risk, especially when dealing with sensitive information. Models can be vulnerable to attack or manipulation, leading to unauthorized disclosure of data.

• Intellectual Property: The use of copyrighted data in training generative AI models raises concerns about intellectual property infringement. The output generated may inadvertently infringe existing intellectual property rights.

• Cybersecurity: Generative AI systems can be vulnerable to cyberattacks aimed at manipulating results or stealing data. Robust cybersecurity protection is essential to mitigate these risks.

• Third-Party Risk: Dependence on third-party AI tools can introduce risks related to data security, privacy and compliance. It is essential to carefully evaluate AI providers and implement appropriate control mechanisms.

• Malicious Use: Malicious actors can exploit generative AI to create deepfakes, spread disinformation or conduct other harmful activities. It is crucial to implement safeguards to prevent and detect malicious use. I had the opportunity to learn about the possible impacts of this malicious use in a debate I recently had with José Rodrigues dos Santos at APG’s 55th International People Conference, about his latest novel “The Chaos Protocol”.

• Ethical and Social Issues: Generative AI raises broader ethical and social issues, including the potential for job displacement, the impact on human creativity and the need for clear ethical guidelines.

The effective management of these risks requires a holistic approach that includes the implementation of technical control mechanisms, the definition of clear governance policies, the promotion of a responsible AI culture and the involvement of multidisciplinary teams.

In addition, it is crucial to keep abreast of evolving AI regulations and adapt practices accordingly. Only through a proactive and comprehensive approach can companies harness the potential of generative AI, minimizing risks and ensuring ethical and responsible implementation.

The Role of Guardrails in AI Risk Mitigation

The article “What are AI guardrails?” introduces the concept of “guardrails” as security mechanisms to ensure that AI tools operate within the organization’s standards, policies and values.

AI guardrails help ensure that an organization’s AI tools, and their application in the business, reflect the organization’s standards, policies and values. The article describes guardrails as mechanisms that ensure AI compliance with organizational policies, values and standards, preventing risks such as inaccurate information (“hallucinations”), toxic content, and privacy and regulatory violations.

The article presents a taxonomy of guardrails, each addressing a specific risk:

• Appropriateness Guardrails: filter out toxic, harmful, biased or stereotype- based content.

• Hallucination Guardrails: Ensure that AI-generated content does not contain factually incorrect or misleading information.

• Regulatory Compliance Guardrails: Validate that the content generated complies with regulatory requirements.

• Alignment Guardrails: Ensure that the content generated aligns with the user’s expectations and does not deviate from its main purpose.

• Validation Guardrails: Verify that the generated content meets specific criteria and can direct the content to a correction cycle.

In this case, what we have is AI itself helping to create mechanisms to regulate the use of AI. Once again, I would stress that what matters here is that human intervention is never dispensed with, either in the role of creating the regulatory rules or in auditing the application of the guardrails.

AI and Agile Leadership: Learning to Let Go

The article “Will artificial intelligence make you a better leader?” explores the relationship between agile leadership and AI. Both require the ability to “let go” and trust unknown processes. AI, by its very nature, requires a leap of faith, as well as embracing ignorance and radically reframing problems. Inner agility and AI may seem like strange bedfellows, but when you consider crucial facts about the latter, you can see its potential to help you lead with clarity, specificity and creativity.

The article recounts a real-life case of a CEO who, facing internal conflicts within the team and difficulties in production efficiency, turned to AI to gain unexpected insights.

By analyzing large amounts of data, AI revealed hidden patterns of poor communication between departments, leading to significant improvements in product launch times and costs.

Once again, what we have here is AI as an accelerator of the reflection process of leaders, through insights that would previously have been very time-consuming to obtain or even “invisible” due to the complexity or sheer volume of data.

But critical thinking and creativity are skills that human leaders need much more in this new paradigm of work and leadership. Instead of being a generator of “mental laziness”, AI can and should be used as an “intellectual provocateur” that should lead us to lead in a more enlightened, creative and sharp way.

In conclusion, the growing importance of AI in the business world and its impact on leadership are absolutely unavoidable and should be the subject of deep and broad reflection, both for each of us and in the public sphere.

Trust, explainability, data management, risk mitigation and agile leadership are crucial elements for responsible and successful AI implementation. As AI evolves, leaders must adapt their leadership styles to embrace uncertainty, promote collaboration and harness the power of AI to drive innovation and growth

What a brave new world awaits us! Shall we?

Listen a podcast about this topic here.