giovanni sileno

@gsileno

researcher in a broader sense. working on automated regulatory systems at UvA. interested in artificial/natural cognition, social constructions and dynamics.

Amsterdam, The Netherlands
Vrijeme pridruživanja: listopad 2011.

Tweetovi

Blokirali ste korisnika/cu @gsileno

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @gsileno

  1. proslijedio/la je Tweet
    17. sij

    if you need AI to tell the teacher when students are bored, you have a problem with teacher recruitment, not with AI.

    Poništi
  2. 20. sij

    9/ Even from a philosophical standpoint, if our aim is to pursue rationality (rational systems, rational institutions, etc.), it is rather obvious that this won't be obtained by empirical means only.

    Prikaži ovu nit
    Poništi
  3. 20. sij

    8/ AI/ML capitalizes too much on optimization, but the contexualization phase (e.g. settling what is "right") might be even more problematic, for instance w.r.t. the social environment, for its high variability. There is a technological gap here [see e.g. our paper on normware].

    Prikaži ovu nit
    Poništi
  4. 20. sij

    7/ Focusing on the current mainstream methods, in supervised ML, by construction, the model is made to satisfy the training samples. What is “right” is set during training. But what to do if what is “right” change after training? For instance, if this is specified by law?

    Prikaži ovu nit
    Poništi
  5. 20. sij

    6/ Still,(most of the times, an hype cycle results in concrete achievements. They just become infrastructure: invisible, but necessary for a next wave to come.

    Prikaži ovu nit
    Poništi
  6. 20. sij

    5/ The hype cycle of raising expectations, illusions and then delusions seems to be systemic. Eacht time the "novel" methods reach a plateau. The problem of intelligence looks again more general/complex than it seemed shortly before.

    Prikaži ovu nit
    Poništi
  7. 20. sij

    4/ Each spring started in practice from 5 ingredients: societal needs, acceptable theoretical paradigms, strong advocates, initial unexpected successes, adequate computational technologies. These trigger and are reinforced by adequate funding, starting a hype cycle.

    Prikaži ovu nit
    Poništi
  8. 20. sij

    3/ Although each time the mainstream topic eclipsed all the rest, the subsequent advances were enabled only because fundamental research in the others somehow continued. [Diversity in research should be really enforced as a policy, just as in evolutionary computation.]

    Prikaži ovu nit
    Poništi
  9. 20. sij

    2/ Historically, many forms of intelligence (or rather, methods to (re)produce appropriate behaviours) have been explored through a cycles of springs (and winters). Deep learning is just the last of a series.

    Prikaži ovu nit
    Poništi
  10. 20. sij

    1/ Most disciplines emerge around specific domains of knowledge (e.g. biology - living forms, physics - laws of the universe, law - legal systems and justice, etc.). AI is instead defined by a purpose: conceiving artificial systems that are "intelligent".

    Prikaži ovu nit
    Poništi
  11. 20. sij

    A (slightly updated) selection of considerations on the "History of AI, Current Trends, Prospective Trajectories", I gave today for the Winter Academy on AI and International Law . Thanks for inviting me again! Main messages:

    Prikaži ovu nit
    Poništi
  12. 8. sij

    Maintaining non-computationally driven procedural knowledge is highly critical to prepare to any event in which technology might suddenly stop working. Do "modern" countries have any strategic backup?

    Poništi
  13. proslijedio/la je Tweet
    21. pro 2019.

    Fascinating case for many reasons, including reasons relating to trias politica

    Poništi
  14. 8. pro 2019.

    Statistical biases reflect frequencies in the data. In contrast, discriminatory biases reflect the role of the algorithms determining decision-making structures, that in turn produce social discriminations. They're rather orthogonal, even without looking at people.

    Poništi
  15. 3. pro 2019.

    "expected utility theory implicitly assumes that individuals can interact with copies of themselves, effectively in parallel universes (the other members of the ensemble)", an ergodic view on utility can leave the individual in only one world:

    Poništi
  16. 13. stu 2019.

    As long as a cyber-physical system eventually interacts with humans, its technical infrastructure will eventually derive its semantics from the social interactions it enables, and for this reason it can always be used for purposes unintended at design time.

    Poništi
  17. 9. stu 2019.

    Said that, if we judge under this lens our contemporary societies, their "intelligence" with respect to global warming is rather limited. Note that whether it is or it is not caused by humans is not relevant here, but how much are we preparing to what is coming.

    Prikaži ovu nit
    Poništi
  18. 9. stu 2019.

    Yet, this definition is particularly interesting because it does not have any technological flavour. It can be applied to evaluate the (practical) intelligence of forms of life, as well as of organizations, of societies, and of human individuals.

    Prikaži ovu nit
    Poništi
  19. 9. stu 2019.

    At second sight, however, the problem is just postponed. Who defines what's "correct"? And for who? in which other contexts is it still correct? Does it take into account the short and the long terms? etc.

    Prikaži ovu nit
    Poništi
  20. 9. stu 2019.

    Intelligence is one of those many words used often but not easy to define. The field of artificial intelligence started from a simple pragmatic choice: intelligence is (selecting and) performing correct behaviour.

    Prikaži ovu nit
    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·