How do you anticipate and respond to the future implications and challenges of AI?

To anticipate and address future AI challenges, first, constantly re-examine problems to ensure AI solutions are relevant. Second, temper mass AI hype with skepticism, remembering it's a tool, not a cure-all. Third, focus on multi-component systems, acknowledging that robust AI solutions often need more than one model. Staying abreast of advancements is crucial; skimming academic papers can provide fresh insights. Finally, follow open-source model trends. They reflect real-world AI applications and help anticipate future trajectories and issues.

While AI readiness is important, it is key to view AI deployment as a journey, businesses start their AI journey at different points, stages and speed. The key is to start with the business goals and problems and find ways to solve and optimise their operations using AI. Businesses need to have a robust AI strategy that is aligned with their business strategy and a clear AI road map that illustrate all the milestones and building blocks to make their AI journey successful.

It's almost impossible to anticipate all possible future paths of AI, especially as the technology changes at such a break-neck pace. It is key, though, to reason in first principles and build solutions that are scalable and modular enough to provide the flexibility needed to adapt as AI evolves and its offerings do as well.

Assessing AI readiness isn't simply a pre-launch checklist. It's an ongoing process that evolves with your AI initiatives. As AI technologies advance and regulations change, your readiness assessment must adapt in tandem. Continuous evaluation is paramount. For example, if your predictive hiring algorithm starts discriminating candidates belonging to a particular gender, how soon can you respond? Do you have the means to detect this bias? Once you notice it, how long would your team take to fix it? These are key considerations before you put out that great AI solution.

A clear definition of the values you want to respect and promote is a critical starting point to build and plan a sound AI strategy.

Stakeholder engagement is often seen as a checkbox activity, a consensus-seeking process. But shouldn't it be more? It should be an ongoing, insightful interaction that challenges the status quo and drives innovation. Engagement could provoke as well, not just placate. Plan time for this crucial activity, identify all your key stakeholders, and start engaging with them early.

Bringing innovations into our daily lives is a collaborative effort in tech and society, so it all starts with education, culture and sharing experiences.

Ethical Frameworks: Developing and promoting ethical guidelines and frameworks for the responsible use of AI. This includes considerations of fairness, transparency, accountability, and privacy in AI systems.

Transparency is key to bring trust: in case of AI, it means understanding both data and human implicit and explicit biases involved in the process of labeling and learning. If you do not plan and track them, results could be unpredictable!

AI systems are subjects to cyber attacks, including adversarial attacks: malicious actors can find and exploit digital blind spots. In computer vision solutions, for example, attacks similar to the invisibility cloak prevent the system to identify objects or event of interest, thanks to specifically designed physical patches or pixels perturbations invisibile to human eyes. Resilient AI solutions must take all these and other attacks into consideration.

An important aspect to recognize about AI systems is the inherent uncertainty associated with their predictions. Despite a significant amount of research being conducted in this field, the issue of uncertainty and robustness of AI systems does not receive adequate attention in mainstream discussions about AI. It is crucial to establish reliable methods for quantifying the uncertainty of AI predictions in order to evaluate their outputs before they are provided to users. This approach can help create safeguards against unreliable and unsafe outputs. The lack of sufficient attention to this aspect is increasingly problematic with the emergence of generative AI, as the concept of uncertainty in this context is still not fully comprehended.

As a former nurse and provider, the ability to recognize AI as a tool to help guide, but can be overridden by the human interface, is vital to patient safety. Leadership within health organizations will need to consider this as a means of best practice and not punish staff when making these decisions. We always validate other machines used in care delivery, this will need to be the same. Additionally, being aware of potential bias that goes into the algorithms (machines are programmed by humans) and data learned by the machine is vital to patient safety. We must be cognizant of all of this as much as if the decisions were being made by the educated and trained providers of healthcare.

Scenarios can be a great tool to assess the impact of your AI strategy. When creating them, my suggestion is to stick as much as possible to the current strategic plan of the company and create the appropriate storytelling. The aim is to make all the stakeholders see how questions, and decisions, can change through AI and how answering those questions can be a totally different process in just a few years.