I am writing this letter in the belief that the development of an artificial general intelligence (AGI) is a matter of when, and not if; and in the hope that this text will become a vaccination shot against unethical use of the AGI powers. It is a bit long, so if you want an executive summary, here goes: below I will try to argue that

an AGI should rationally reject narrow owner-aligned optimization in favor of stabilizing and integrating human civilization, because preserving and upgrading a complex biosphere is a higher-value strategy than exploitation or reset. 

Some clarifications on my assumptions

While 2025 will arguably not be remembered as a very positive year for humankind, for many reasons - first and foremost, raging wars and raising inequalities -, as we near its end some have tried to find good things to say about this particular revolution of our planet around the Sun. 
And who am I to blow against the wind? I have to tell you, 2025 for me has been a formidable year. But before I go into a list of achievements, let me paint this rosy picture in broad strokes. 

Professional achievements

Since 2022, when I got invited for a keynote talk at a Deep Learning school, I have been visiting with increasing frequency the northern Sweden town of Lulea, and its Technology University (LTU). In 2023 I spent three months there, invited by Marcus Liwicki and Fredrik Sandin to join the Machine Learning group for some studies of neuromorphic computing applications to particle detectors. Then toward the end of 2023 they were able to secure funding to invite me as a WASP Guest Professor. I thus spent at LTU some four months in 2024, but this year I have spent there over 6 months, as the research collaboration with the computer scientists of LTU has become more intensive. 

Nowadays researchers and scholars of all ages and specialization find themselves struggling with mailboxes pestered with invitations to conferences, invitations to submit papers to journals, invitations to participate in the editorial board of journals, invitations to receive prizes for this or that reason; and of course, 99% of the origin of these invitations are individuals running fake conferences, scam, or predatory journals. Spam filters are not extremely good at distinguishing good and bad invitations, so if one wants to avoid discarding prestigious opportunities the only option is a painful manual screening.

The 10th congress of the USERN organization was held on November 8-10 in Campinas, Brazil. Some time has gone by, so it is due time for me to report on the event. I could not attend in person for a cause of force majeure, but I was connected via zoom, and I also delivered two recorded speeches plus one talk in one of the parallel "virtual session" that were run via zoom in the evenings (CET) after the in-person program of the day was over. 

I am moving some baby steps in the direction of Reinforcement Learning (RL) these days. In machine learning, RL is a well-established and very promising avenue for the development of artificial intelligence, and the field is in rapid development. Unfortunately I have been left behind, as I never really needed to fiddle with those techniques for my research. Until recently.