Re: Open letter urging to pause AI

po 3. 4. 2023 v 20:27 odesílatel ProjectParadigm-ICT-Program <
metadataportals@yahoo.com> napsal:

> I never expected my post to elicit such an amount of response. But for
> direction's sake and for keeping on topic, let's take a look at the
> takeaways.
>
> - LLMs aren't going away anywhere soon, especially with the insanely crazy
> amounts of money being invested in it, having started an AI arms race of
> sorts;
> - the software industry and software giants like Microsoft and OpenAI
> couldn't care less about regulation and factor in enormous fines as the
> cost of doing business when making huge profits;
> - the FutureofLife Open Letter to Pause Giant AI Experiments, more closely
> looked at has some flaws;
> - Timnit Gebru, Emily M. Bender, Angelina McMillan-Major and Margaret
> Mitchell were not found on the list of signatories to the letter and
> published a rebuke, calling out the letter's failure to engage with
> existing problems caused by the tech, in particular the ideology of
> longtermism that drives much of current AI development.
> Source:
> https://techcrunch.com/2023/03/31/ethicists-fire-back-at-ai-pause-letter-they-say-ignores-the-actual-harms/
> - all efforts to develop AI that meets with ethical guidelines, standards,
> is open, explainable, controllable, trustworthy and safe, whether from the
> UNESCO/UN, standards bodies, professional regulatory bodies and
> professional associations by default are lagging behind;
> - the potential users of a seemingly endless list of applications across a
> wide board of industries of the LLMs are not fully aware of the
> capabilities and potential unintended, or not disclosed secondary impacts;
> - there are plenty of unknown unknowns, like emerging abilities now being
> studied to comprehend and get a grasp on what the LLMs are capable of and
> consequently the unintended, potentially harmful capabilities that in
> certain cases need to be contained, avoided or disabled;
> - the huge carbon and energy footprints of the infrastructure for the
> global wide scale deployment of such technologies.
>
> All in all, in my humble opinion the best approach to tackle this issue is
> to have both professionals inside the industry knowledgeable of the
> problems, dangers and unknowns, together with professionals from the fields
> of software engineering, computer science, mathematics and all fields and
> industry currently using AI as a key technology (e.g. pharmaceutical
> industry, biotechnologies, bio and life sciences, medical technologies,
> natural and geo sciences, chemistry and material sciences) take a good look
> at all the current proposals for guidelines for ethical, explainable, open,
> safe and trustworthy AI, as they relate to their industries, field of
> activities or scientific endeavors and condense these to an actionable list
> for politicians both at national, regional and global body levels to work
> on.
>
> And in the meantime have educators, academics, scientists, engineers and
> professionals from industries using AI technologies stay on topic and
> contribute to an open dialogue about the issues at hand.
>
> The issue is not to stop the development and deployment of AI but to do so
> in a responsible way, based on some agreed upon consensus on guiding
> principles.
>

I'd recommend this video to get a better understanding of the landscape and
the threat model:

https://www.youtube.com/watch?v=fKgPg_j9eF0


>
>
> Milton Ponson
> GSM: +297 747 8280
> PO Box 1154, Oranjestad
> Aruba, Dutch Caribbean
> Project Paradigm: Bringing the ICT tools for sustainable development to
> all stakeholders worldwide through collaborative research on applied
> mathematics, advanced modeling, software and standards development
>
>
> On Saturday, April 1, 2023 at 10:15:41 PM AST, Melvin Carvalho <
> melvincarvalho@gmail.com> wrote:
>
>
>
>
> so 1. 4. 2023 v 23:29 odesílatel Adeel <aahmad1811@gmail.com> napsal:
>
> It is a constructive discussion on the general ethics - it is a core
> competency in every professionalism. Not left to legal speculation whether
> it be the use of transformer models, code of conduct, or selectively
> defined for others, practice what you preach which is the real issue with
> ethics these days.
>
>
> Perhaps calm down a bit and stay on-topic
>
> As an interesting exercise I ran this tread through chatGPT asking: "rate
> this statement as whether or not it is passive agressive on a scale of 1 to
> 10 and why" -- give it a try if you want to see some interesting results
>
> What is came up with passive aggression:
>
> Hugh - 3
> Dan - 5
> Pat - 4
> Adeel - 5,7
> Nathan (control) - 2
>
> Perhaps if run again it would give different numbers.  I expect AI is
> going to be integrated into all sorts of tools, so that we'll get flags and
> pointers in email before sending.  Which will change the nature of mail
> lists forever!
>
> As a side note I chat to new developers about semantic web all the time.
> One recently said to me: "I hope stuff like this can popularize linked data
> without the smugness".
>
> This is actually one of the more pleasant, jolly threads.  But we probably
> can and will do better as a list, when it starts to be common place for AI
> tools to help us.
>
> It is a technology that huge implications for the semantic web.  Because
> the semantic web is about meaning.  There are times when technology moves
> too fast, and has societal implications, so imho, the question is legit.
> Indeed in the AI world this topic has been the big one in the last week,
> with dozens of different views.  It's interesting.  Emad has in any case
> said he thinks gpt v.next will take 6-9 months to train anyway.
>
> The main limitation I have found is that it can do small tasks reasonably
> well, but when you hit the 32k limit, it struggles just as much as we all
> do with larger topics or code bases.  Perhaps that will chain soon with
> plugins tho, we'll see!
>
>
>
> On Sat, 1 Apr 2023 at 22:12, Patrick J. Hayes <phayes@ihmc.org> wrote:
>
> Adeel, greetings.
>
> The code of conduct applies to us all, including of course Dan B. But the
> point here is that Dan has not violated that code, whereas you have, first
> by directly insulting Dan by implying that his posts are dishonest or
> motivated by dark or sinister commercial forces, and then by continuing
> this thread of thinly veiled hostility when there is no reason for it. I am
> not an administrator of this group, but if I were you would now be banned
> from it.
>
> Please find some more constructive way to contribute to the discussion.
>
> Pat Hayes
>
> On Apr 1, 2023, at 1:22 PM, Adeel <aahmad1811@gmail.com> wrote:
>
>
> Hello,
>
> That sounds like a legal speculation, or do you only selectively
> discriminate on group members and their backgrounds when you point that out?
> Does the W3C code of conduct not apply to you after 25+ years of being
> here?
>
> Thanks,
>
> Adeel
>
>
> On Thu, 30 Mar 2023 at 23:33, Dan Brickley <danbri@danbri.org> wrote:
>
> On Thu, 30 Mar 2023 at 20:59, Adeel <aahmad1811@gmail.com> wrote:
>
> Hello,
>
> You can't talk about regulation and compliance in this group, dan doesn't
> like it as google doesn't care about those things.
>
>
> Rather, Google doesn’t delegate to me any permission to speak on its
> behalf on these matters, unsurprisingly enough. Google is also and
> organization of many thousand employees, with quite some range of views. I
> choose not to share several of mine here right now, but I am broadly in
> favour of sensible laws where they have some chance of working.
>
> As in many situations it sometimes makes sense to treat a company as a
> kind of pretend person, and sometimes to look at it as a more complex
> system of forces and interests. Skim
>
> https://www.wired.com/story/inside-google-three-years-misery-happiest-company-tech/
> <https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.wired.com%2Fstory%2Finside-google-three-years-misery-happiest-company-tech%2F&data=05%7C01%7Cphayes%40ihmc.us%7Cd884dfb499ef423f124f08db32ef753e%7C2b38115bebad4aba9ea3b3779d8f4f43%7C1%7C0%7C638159776297688875%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=akFFkIcIXEgvQWjOdtQO4QjZOzAjTOJ82zH1mJwAb%2Bo%3D&reserved=0>
> to see some of what bubbles beneath the surface.
>
> This email list is just an email list. W3C no longer accords it “interest
> group“ status, as it did from 1999 when I migrated the older RDF-DEV list
> to W3C to form the RDF Interest Group on www-rdf-interest@ list. W3C
> doesn’t officially listen here even on RDF topics, let alone areas like
> modern AI and their social impact which aren’t obviously our core
> competency.
>
> Having been here 25+ years I have some instincts about which topics will
> just fill up email inboxes with no ultimate impact on the world and
> benefit. Right now “something must be banned” threads on AI look to me to
> fall in that category.
>
> Cheers,
>
> Dan
>
>
>
>
> Thanks,
>
> Adeel
>
> On Thu, 30 Mar 2023 at 20:22, adasal <adam.saltiel@gmail.com> wrote:
>
> It's out of the bottle and will be played with.
>
> " .. being run on consumer laptops. And that’s not even thinking about
> state level actors .. "
> Large resources will be thrown at this.
>
> It was a long time ago that Henry Story (of course, many others too, but
> more in this context) pointed out that, as to what pertains to the truth,
> competing logical deductions cannot decide themselves.
>
> I just had this experience, and the details are not important.
>
>
> The point is that, in this case, I asked the same question to GPT-4 and
> perplexity.ai
> <https://nam02.safelinks.protection.outlook.com/?url=http%3A%2F%2Fperplexity.ai%2F&data=05%7C01%7Cphayes%40ihmc.us%7Cd884dfb499ef423f124f08db32ef753e%7C2b38115bebad4aba9ea3b3779d8f4f43%7C1%7C0%7C638159776297688875%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=twBLGhgQjubMJb6eq223Hsx8kKJMs8Tiq7jMBfpFyPc%3D&reserved=0>,
> and they gave different answers.
> Since it was something I wanted to know the answer to, and it was
> sufficiently complex, I was not in a position to judge which was correct.
>
> Petitioning for funding for experts, i.e. researchers and university
> professors.
> Although it is absurd to think they would have time to mediate between all
> the obscure information sorting correct from incorrect and, of course, a
> person can be wrong too.
>
> Then there is the issue of attribution ...
> At the moment, perplexity.ai
> <https://nam02.safelinks.protection.outlook.com/?url=http%3A%2F%2Fperplexity.ai%2F&data=05%7C01%7Cphayes%40ihmc.us%7Cd884dfb499ef423f124f08db32ef753e%7C2b38115bebad4aba9ea3b3779d8f4f43%7C1%7C0%7C638159776297688875%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=twBLGhgQjubMJb6eq223Hsx8kKJMs8Tiq7jMBfpFyPc%3D&reserved=0>
> has a word salad of dubious recent publications; GPT -4 has a "knowledge
> cutoff for my training data is September 2021". It finds it difficult to
> reason about time in any case, but these are details.
>
> Others in this email thread have cast doubt on Musk's motivation (give it
> time to catch up) and Microsoft (didn't care for any consequences by
> jumping in now).
>
> So there are issues of funding and control -- calling on the state to
> intervene is appealing to the power next up the hierarchy, but can such
> regulations be effective when administered by the state?
>
> That really just leaves us with grassroots education and everyday
> intervention.
>
> Best on an important topic,
>
>
> Adam
>
> Adam Saltiel
>
>
>
> On Wed, Mar 29, 2023 at 9:39 PM Martin Hepp <mfhepp@gmail.com> wrote:
>
> I could not agree more with Dan - a “non-proliferation” agreement nor a
> moratorium of AI advancements is simply much more unrealistic than it was
> with nukes. We hardly managed to keep the number of crazy people with
> access to nukes under control, but for building your next generation of AI,
> you will not need anything but brain, programming skills, and commodity
> resources. Machines will not take over humankind, but machines can add
> giant levers to single individuals or groups.
>
> Best wishes
> Martin
>
> ---------------------------------------
> martin hepp
> www:  https://www.heppnetz.de/
> <https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.heppnetz.de%2F&data=05%7C01%7Cphayes%40ihmc.us%7Cd884dfb499ef423f124f08db32ef753e%7C2b38115bebad4aba9ea3b3779d8f4f43%7C1%7C0%7C638159776297845108%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=FCqp2o420lD%2FDJA%2BsW2xmUV1SL4eb4FqubE%2FFbQiwv0%3D&reserved=0>
>
>
> On 29. Mar 2023, at 22:30, Dan Brickley <danbri@danbri.org> wrote:
>
>
>
> On Wed, 29 Mar 2023 at 20:51, ProjectParadigm-ICT-Program <
> metadataportals@yahoo.com> wrote:
>
> This letter speaks for itself.
>
>
> https://www.reuters.com/technology/musk-experts-urge-pause-training-ai-systems-that-can-outperform-gpt-4-2023-03-29/
> <https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.reuters.com%2Ftechnology%2Fmusk-experts-urge-pause-training-ai-systems-that-can-outperform-gpt-4-2023-03-29%2F&data=05%7C01%7Cphayes%40ihmc.us%7Cd884dfb499ef423f124f08db32ef753e%7C2b38115bebad4aba9ea3b3779d8f4f43%7C1%7C0%7C638159776297845108%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=n%2BG6aomh5Ad6qwdk9euMYmYvQFLs0dSYFMpwJ%2FquyPo%3D&reserved=0>
>
>
> I may not want to put it as bluntly as Elon Musk, who cautioned against
> unregulated AI which he called "more dangerous than nukes", but when Nick
> Bostrom, the late Stephen Hawking, and dozens, no hundreds of international
> experts, scientists and industry leaders start ringing the bell, is is time
> to pause and reflect.
>
> Every aspect of daily life, every industry, education systems, academia
> and even our cognitive rights will be impacted.
>
> I would also like to point out that some science fiction authors have done
> a great job on very accurately predicting a dystopian future ruled by
> technology, perhaps the greatest of them all being Philip K. Dick.
>
> But there are dozens of other authors as well and they all give a fairly
> good impression what awaits us if we do not regulate and control the
> further development of AI now.
>
>
> I have a *lot* of worries, but the genie is out of the bottle.
>
> It’s 60 lines of code for the basics,
> https://jaykmody.com/blog/gpt-from-scratch/
> <https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fjaykmody.com%2Fblog%2Fgpt-from-scratch%2F&data=05%7C01%7Cphayes%40ihmc.us%7Cd884dfb499ef423f124f08db32ef753e%7C2b38115bebad4aba9ea3b3779d8f4f43%7C1%7C0%7C638159776297845108%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=2QEWpct5NS36zpwh2zyEISddXRBZMJeJtvMcQSJ8HME%3D&reserved=0>
>
> Facebook’s Llama model is out there, and being run on consumer laptops.
> And that’s not even thinking about state level actors, or how such
> regulation might be worded.
>
> For my part (and v personal opinion) I think focussing on education,
> sensible implementation guidelines, and trying to make sure the good
> outweighs the bad.
>
> Dan
>
>
>
>
> Milton Ponson
> GSM: +297 747 8280
> PO Box 1154, Oranjestad
> Aruba, Dutch Caribbean
> Project Paradigm: Bringing the ICT tools for sustainable development to
> all stakeholders worldwide through collaborative research on applied
> mathematics, advanced modeling, software and standards development
>
>
>