Maersk

Global understanding on the direction and concerns around AI, especially around agentic AI around usage of data and autonomous decision making in multi-layer complex settings, like supply or value chains, where one decision may contaminate and strongly influence across industries due to heavy interdependencies. Agreement on what needs to be done from governments and private sector together to not only test these models and agents but also ensure the data they are using is correct, meaningful and unbiased, in order to ensure the decisions suggested are as fair and neutral as possible.

• Safe, secure and trustworthy AI
• AI capacity-building
• Transparency, accountability, and human oversight
• Open-source software, open data and open AI models

Safe, secure and trustworthy AI is a major concern and should be for all, across all industries and users, strongly linked to transparency, accountability across the companies and teams that build and develop the models, with a strong governance and human oversight to not only influence the direction and limits but also identify and deal with deviations that pose a threat to general safety, fairness and trust. Capacity-building and open source software are also relevant to ensure we can all develop across nations and emerging economies or smaller companies, so there is no enlargement of the gap between larger organizations or richer economies and the smaller ones or emerging markets. The more open source models are it is also easier to ensure accountability by testing them openly, so any deviation should be identified and corrected faster.

AI Governance, especially for multi-tiered usage in agentic AI, as the models can be trained for self-optimization but when they are competing with each other for optimization in a complex network they may end up making decisions that are contra productive for the greater good or strongly create biased results in a geopolitical scenario, especially when stronger models may be competing with weaker ones. So there needs to be some international legislation on decision making processes across global value chains. Also, the same with AI deception, highlighted by UN Scientific Council, what are the ways to detect and correct when this happens, especially in scenarios of growing complexity, with thousands of pieces of information or decisions being shared and made everyday, autonomously.

In supply chains, we are still in a very nascent stage in terms of agentic AI, but we expect a big increase in the upcoming years, especially once the data needed for it to work are cleansed and structured in ways that can expand its access by AI (which is the stage most companies are at now). Once industries start experimenting more with agentic AI and replacing human decision with AI decision across different parts of the supply chain, this may create a temporary sense of increase of efficiency and automatic improvement but that would eventually come at the cost of other supply chains, given the high level of interdependencies across suppliers in same or even different industries (like the chips manufacturers), limited capacity of transportation volume in the market or constant changes and disruptions, normally announced to all customers at the same time and where humans would try to make the best decisions, but now that could be replaced by AI with hidden rules that may conflict with each other.

I believe it can open the door to understand different perspectives and points of view from private sector, academia, governments or even end users and also expand the knowledge and awareness of both the challenges but also what may be some of the collective solutions for these challenges. It can also make tech companies more accountable to what they are building, but putting their work on a global stage and showing the impact of what and how is being built may have across our ways of working and living.

It should build a very strong partnership with the industry and the private sector, who will be not only some of the biggest builders of the next phase of AI, but also one that will have a very strong impact in the way we manufacture, transport and sell products, impacting all users across the globe, as the next industrial revolution. It is also important we create mechanisms to share what is already being done and make companies more aware of not only how they should contribute but also what they should be paying attention to and do not compromise on, in the race to developing their AI processes internally and with their partners.

AI Dialogue should be open to the public and to industry representatives, but necessarily as spokespersons for their companies but representing their industry, challenges or knowledge (in case they are not necessarily aligned to their companies official approach). There should be made collective groups across different themes (industries or geographies) that can act as thinktanks of what is being build and implemented, have access to the base of the AI (data, sources, models) and see what needs to change as it scales across use cases and industries, as this is currently being decided only by tech companies and governments and I think they do not have the necessary information and level of interest to solve the issues as the industries players or collective society would have.

Clearly industry and private sector across different company sizes and types of industries, but also civic society. However, I can see that AI to final consumer is already very much mainstream now, so there is sufficient knowledge and opinions, with AI in a B2B setting is still very new and we are not aware of what will happen, how and who will be making those decisions. So voices coming from industry sectors that have strong real-life knowledge of impact of the decisions, technology being used or that could be used and can understand the implications of the shifts across geopolitical and international trade decisions and balance.

Short workgroups of experts across different industries working together testing "What if" scenarios, including shifts in manufacturing, international trade tensions, sourcing areas, business models and how that would influence product availability in food products, FMCG, lifestyle, chemicals or others. Having several experts that understand the flow dynamics and the impact of AI decision making in those dynamics across product sectors and planning for what could be done and how and what should not be done and why, to avoid having to fix problems after they occur.

A global platform with practicioners and experts dedicated or interested in working together in certain topics would be an interesting take, similar for example to innovation platforms that launch ideas or challenges that all involved can participate. From a police perspective, it would be interesting to also define what agentic AI can use from an global perspective while interacting with other agents and how many levels of agents there can be and what decisions can be done with human intervention or oversight. From a practice perspective, all new use cases should have experts from the AI workgroups testing and seeing how the solution works before it can be considered "safe" to the scaled and used for other companies, so potentially could be interesting to come with a "AI safety" certification so users can know which AI providers and tools to trust.