Daniel Dobos

Following the advise from Benjamin Franklin: "Tell me and I forget, teach me and I may remember, involve me and I learn" helps here a lot. Involve the leaders directly in the formation of the education process. Let their curiosity drive the education agenda of real life challenges that allow them to learn the challenges and opportunities directly by own experiences. Taking training, test and validation datasets from their own domain, or ideally even their own organisation allows to see the business value of ML methods directly without the need of learning transfer.

While addressing privacy risks transparently is a good start, but demonstrating that one knows the right tools and techniques to manage and mitigate these risks is even more important. A good principle is to think of real customer data as being radioactive. Stay away from it. Using Privacy Enhancing Techniques (PET) on the training and test datasets is a good starting point. But especially the progress on synthetic data generation makes it also possible to create fully synthetic datasets, which allows to nearly completely eliminate all AI system privacy risks for POCs, MVPs and partially also for developments.

Learnings from the Cybersecurity domain show the high value of common Red Teaming. Many AI regulations, like the White House declaration, ask for common AI Red Teaming between different organisations and entities. Information exchanges about threats, attack vectors, as well as vulnerabilities between organisations from different domains and sectors can reduce dwell times and allows accelerated learnings how to effectively manage and mitigate the additional risks arising from AI systems.

While GPUs and TPUs are the most common hardware tools to accelerate machine learning, these are not the only once. Don't forget about FPGAs and DSPs that can be used for realtime inferences (records around 200 million inferences per second). Or massive parallel CPUs (up to 900k with 44 GB on chip memory) can be a faster and often cheaper alternative to the "classical" GPU solution.

Like for many innovation processes, Benjamin Franklin's quote is worth gold here ad well: Tell me and I forget. Teach me and I remember. Involve me and I learn.- Benjamin Franklin Hackathons, Innovation Sprints and other collaborative innovation events are good examples to generate sustainable and impactful management buy-in.

When I started my research and innovation career, I was sad and frustrated whenever an innovation was not implemented immediately. When the business part was "not prepared for success". When we had to put a successful project "into the drawer". But with time I learned that in most cases it takes only a few months until somebody has the same 'new' idea once again and you can shine by opening the drawer again and implement. Can on spare young colleages from making the painful experience themselves? Resilience is hard to teach by words, but I am happy for all suggestions.