Re: Human-Centric vs. Altman-Centric AI

On Wed, 25 Sept 2024 at 08:18, Michael Robbins <
michael@learningpathmakers.org> wrote:

> We have failed to protect the web. Big AI is eating it and filling it
> with slop. Now we must take a radically different approach:
>
> "Investors seeking to buy into OpenAI’s latest $6bn-plus funding round are
> making an unprecedented bet that the ChatGPT-maker will become the world’s
> dominant artificial intelligence company and be worth trillions of dollars.
>
> The San Francisco-based start-up is finalising a new fundraising valuing
> the company at $150bn. Thrive Capital, Josh Kushner’s venture capital firm,
> has already provided at least $1bn to the company in recent weeks,
> according to people with knowledge of the deal.
>
> OpenAI aims to raise an additional $5bn or more. Apple, Nvidia and
> Microsoft — the three most valuable technology companies in the world — are
> in talks to join the funding round. Others seeking to invest are New
> York-based Tiger Global and United Arab Emirates-backed fund MGX, according
> to multiple people with knowledge of the discussions. The deal is expected
> to close imminently.
>
> However, other leading tech investors, including Andreessen Horowitz and
> Sequoia Capital — Silicon Valley’s top venture capitalists and existing
> OpenAI backers — are sitting out of the round, according to people with
> knowledge of the matter.
>
> Investors in the deal said it was highly unusual in its scale and
> structure. Venture investors such as Thrive and Tiger typically write far
> smaller cheques for less established start-ups, hoping for 10 to 100 times
> their money back.
>
> To achieve such a return with OpenAI, the company would need to grow in
> the coming years to become worth at least $1.5tn; larger than Facebook
> parent Meta and Warren Buffett’s Berkshire Hathaway.
>
> Many are persuaded it will. “We’re talking about the path to building a
> trillion-dollar company,” said a partner at an investment firm that has
> backed OpenAI. “I don’t think this is unreasonable.”
>
> (Via the Financial Times)
>
> Make no mistake. This is what human-centric AI is up against. It’s Game of
> Thrones in the Age of Technofeudalism.
>
> *Quis custodiet ipsos custodes?*
>
> Winter is coming. We can’t combat this with a litany of tech standards and
> solutions. It must begin with people and their “why” for data dignity: to
> build AI agents that help them in school, work, and life.
>
> This is the challenge that I’ll be setting out for this community group. *There
> must be a pathway and purpose for exodus from Technofeudalism and the web
> as we know it.* I’ll be sharing what I’ve discovered for this with the
> group. And recruiting others to join us, to take up where the large
> majority of silent group members have abdicated their participation.
>
> [image: image0.jpeg]
>
>
"OpenAI’s alignment or superalignment team is responsible for safety, and
creating more human-centric AI models."
source:
https://decrypt.co/231294/openai-exec-quit-safety-took-backseat-shiny-products


I have a Google Alert - "Human Centric AI", there's alot of other
examples...  Overall, the combination of SEO and marketing generally,
alongside money - words appear to be employable in ways that mean the
definition of terms, if not clearly defined otherwise - can be made to mean
whatever market leaders want.

as was why i noted 'humanitarian ict',
https://datatracker.ietf.org/meeting/119/materials/agenda-119-hotrfc-sessa-04#h.eno136kpyxqs


as - solutions either support human rights, or designs are unable to, for
whatever reason...

FWIW: I notice, in some community discussions elsewhere, that the better
API reference is via MS.
https://learn.microsoft.com/en-us/azure/ai-services/openai/reference#responses

Certainly, defining an API alternative to OpenAI might be helpful &
good...  but, overall, i sometimes question whether the outcomes will
end-up being in english, because if - say - we were back in the era / time,
when personal computers were emerging,  i could imagine thin-clients to
mainframes competitively selling alternatives, 'cheaper', that declared
themselves to be the same thing - not because its an accurate description,
rather, just because - it doesn't actually matter...  even though, without
personal computers, or indeed in particular also - NeXT - the advent of
'interpersonal computing', the tools used to create the web, wouldn't have
existed.

Tim.H.



>
> On Sep 24, 2024, at 2:02 PM, Timothy Holborn <timothy.holborn@gmail.com>
> wrote:
>
> 
> Hi Michael & others on the list...
>
> Whilst I personally despair the direction wide-scale AI developments have
> taken, and the various related implications; I think it's important for
> various reasons to ensure language does not target people[1], but rather
> problems.  Further, if one 'looks under the hood', I suspect findings
> would show MS more influential[2][3].   This week, there's been some fairly
> significant UN meetings, based on what now exists.
>
> not what could have, should have or might have been made, but was neither
> available to try, nor were the marketing materials available for decision
> makers to consider...
>
> Within this complex scope of so called 'competition', there are also
> competing applied definitions to the notion of 'human centric', and as a
> W3C CG is a community space - i sought input to better understand what the
> meaning of the nomenclature was now to people[4], notwithstanding my own
> strong views[5]...
>
> FWIW: there's a call for submissions re: https://www2025.thewebconf.org/
> which can be found on that site.
>
> A Quick update otherwise; noting, its not specifically 'w3c work', rather,
> just letting people know - perhaps something will fall out of it, that
> ends-up needing w3c processes or would become better, if that occurred...
>
> I think my work is fairly slow-going atm..  nonetheless.  The work on
> defining different types of 'agents' or artificial minds, is a continuing
> topic.  This sheet has some old notes[6] but I haven't really updated it.
> I'm still working through the process of defining what I call 'webizen',
> which I don't think should be anthropomorphised..  Therein, considerations
> about defining them AS ROBOTS, or FOAF:AGENT:SOFTWARE (not legal person)
> and thinking about how to create 'cartridges', for lack of a better term
> now; where they may be loaded / unloaded, called for specific purposes, and
> in-turn how to define them..  Part of these considerations are about the
> psychological aspects, whilst the other part is about the geek processes.
>
> I've not historically been focused on LLMs, as my view was that the
> foundations needed to be done first, then linked to LLMs; but as noted,
> I've got the hardware to do the RD&D, nonetheless, it's only been a few
> months...  I've updated the 'tools' sheet, with some LLM related
> resources[7], there's some other notes - particularly about GPU hardware in
> this sheets doc[8] and both need updating - whilst i've also been thinking
> about how to define a schema & ideally also, enable decentralised discovery
> / updates..
>
> I've got 'models' running locally, and have then got them working on my
> other devices over TailScale[9]  which has then led to investigation about
> replacing the 'default assistant app' on my phone, finding:
> https://github.com/AndraxDev/speak-gpt/  but i haven't got HTTPS working
> yet.  Currently investigating LocalAI[10][11] which is written in go, and
> i'm hopeful, might make it feasible for me - to integrate with the older
> WebizenAI[12] work, that used the old RWW[13] libraries, whilst also
> exploring spatio-temporal supports, as noted earlier[14]. Earlier work,
> considered the use of WebID-TLS / WebID-RSA for devices & things, alongside
> the need for WebID-OIDC for AUTH - but - context was about my attempts (now
> considered to have failed) to define a 'knowledge banking' ecosystem(s). If
> alternatives are running locally, context changes..  So, as I'm working
> through the issues with enabling TLS, I'm also considerate of the NextCloud
> Solid[15] work, and am generally seeking to ensure support for SANs...
> but, its all very experimental atm..
>
> As for fairly simple tests / 'thought experiments'; I've also been toying
> with the idea of how to define a PRIVATE & Permissions Supporting - chat
> app; that uses local LLMs, and has the purpose of translating the language
> people communicate in, to whatever language is preferred by the recipient.
> Then also, considering what functionality might be added into it, to
> support a profile intended to be for children (with guardianship supports);
> in-part, considering proposed laws that claim to be about protecting
> children on social media, by banning the use of social-media by young
> people; but social media or social-web was part of the old definition of
> 'web 3.0'[16]; whereby 'social', refers to socialisation a foundational and
> essential element to the natural function of many species of life, flora,
> fauna & funga, and media being a plural to medium - so, the words they use
> (whilst seeking 'age credentials' / identity wallet integration with social
> media silos[17], that i envisage will lead to payments requirements being
> sought to be resolved) have far broader implications; and, i find the
> situation broadly depressing, as the 'status quo', due to the priorities
> illustrated by others generally, means that there isn't a good way to
> illustrate to these decision makers - the envisaged alternatives, that they
> may act to make unlawful to produce - that could otherwise act to support
> engagement between one-another, rather than dependencies upon global social
> agora that continues to engender harms upon all people, whilst certainly
> not equally, indeed, moreover - dynamically, per idiocracy related
> 'norms'...  IMHO...
>
> The embodiment of these sorts of apps are still envisaged to be progressed
> as a web-extension, but there's also still plumbing related processes being
> investigated.
>
> *NOTE: that in my modelling, i'm assuming users have their own domain (or
> sub-domain); as was part of the
> 'ADP' https://datatracker.ietf.org/meeting/119/materials/agenda-119-hotrfc-sessa-04#h.eno136kpyxqs
> <https://datatracker.ietf.org/meeting/119/materials/agenda-119-hotrfc-sessa-04#h.eno136kpyxqs>
> work..  Therein, for example, users could create Email Alias that
> specifically relates to the relationship (ie: the other person in their
> 'address book'); which, was hoped to help combat spam; but, may also now
> have other implications given the direction of emergent laws, alongside
> interfaces for different kinds of AI Agents.*
>
> The other issue, that I've done SOME work on, but not enough, is in
> seeking to figure out how best to structure content for ingesting into
> LLMs.  They don't appear to support RDF 'natively', rather - designed with
> a focus on JSON; which I think, has implications due to the lack of
> name-space support; and, the means to better integrate Triple/Quad Stores
> with VectorDB sources, I haven't figured out yet.  At least, not well.  I
> experimented with uploading my pathology results archive into an LLM but
> got poor outcomes, understanding that as the system is running on my
> machine - its private. I also tried processing proposed legislation, but
> similarly didn't get very good results - indeed, it seems they may be
> 'preset' to preference US legal structures, which has alot of differences
> to other jurisdictions and related contexts.
>
> But i need to figure out how to process the input docs, into whatever the
> 'best' format is; and am then seeking to progress previous work on 'hyper
> media containers', which would package the files, provide an index in
> whatever the best format ends-up being defined as; and, incorporate also
> the links to original sources, etc...  which may in-turn lead to improved
> solutions that might seek to make software available over IPFS, WebTorrent
> or similar...
>
> Part of the original purposes of importance to me, was the means to more
> easily present complex legal issues to authorised parties...  part of this
> process was also about collection of digital evidence, therefore
> 'verifiable claims'...  I think the function of a personally operated
> 'default assistant app', and related software, might change the way phones
> work - so that, the means in which web-results change seemingly related to
> private conversations - is more - personally controllable, definable,
> etc... doesn't need to share 'the data' to foreign companies /
> jurisdictions; and, make use of the active employment of those sensors for
> purposes that act to better support human rights, as have otherwise been
> set-aside...  or, considered a lesser priority, etc.  Yet, as is, i think,
> consequential to the still nascent state of development in these areas, i
> think there is still a lot of work to do; and, believe (given current
> resources) it'll take some time to advance into something, that can be
> easily deployed on a 'mac studio' or similar...
>
> Another finding, fwiw, there's alot of these LLM applications and they
> usually all want to provide the 'llm store', which, given some of these
> 'models' or 'ai agents', are 10's of GB (GB size, generally associates to
> how many GB of vRAM needed to run them), duplicates on a system is
> undesirable.  It would be better if there was a 'common store', and some
> standard way apps can try to find it; whilst, there's still competition on
> the host software running, that is then expected to load it; and, having
> many of these agents is also undesirable...
>
> AND: I was asked about Computer Vision, CCTV, Video Surveillance policy
> recommendations.  I made a few suggestions based upon the notions of
> ensuring people own their own biometric signatures, decentralised
> discovery, support for 'commons informatics' in a manner that protects
> 'freedom of thought', as to seek to ensure analysis and/or identification
> of objects (macroscopic or microscopic) does not end-up being 'owned' as a
> proprietary interest via a particular jurisdiction (perhaps foreign, from
> source) & various other related considerations, noting Mico-Project[19] and
> Sparql-mm (thought usefully employed with Sparql-Fed) presented a potential
> path, sometime ago..  I don't think Sparql-MM documentation is on W3C atm?
> not sure how it could or should be updated either...
>
> But; Per the Anti-Pattern[3] related realities - there is, at least
> seemingly, a functional process where works seeking to define alternatives
> to 'god ai'[18] projects can be disrupted, as a 'threat' by a now very
> powerful movement, somewhat denoted by the use of JSON specifically, which
> re: "The Software shall be used for Good, not Evil."[21] subjectively
> associates to various complex implications, that are seemingly very
> difficult for many 'geeks' and thereby reasonably impossible for other non
> computer science experts - to be reasonably expected to understand..  OH -
> if / when, the social elements develop, the means for people to form &
> manage agreements between one-another, as is different to platform provided
> 'mandates', also brings about a need to look at how to support rules in
> association to those agreements...
>
> SO, whilst i tried to get UN Parties to help craft the UN Instruments to
> be useful for that purpose; after years of best efforts,  i'll likely
> end-up progressing that work myself..  as much as i had hoped there's be
> interest to build it to be supported via https://metadata.un.org/
>
> certainly, its difficult to both comply with expectations and
> simultaneously, whilst controlling utterances, remain hopeful for the
> future and communicate honourably, derivatives of work intended to promote
> human rights and protect the means for persons to resolve disputes (or
> consequences of harms) peacefully..  but, i also think that these issues
> are indeed common and that it most disaffects those who truly care about
> those sorts of values, as is different to the actions, activities and
> derivatives of gamification... therein, 'reality check tech' would and
> should end-up illustrating facts about how all natural agents are indeed
> imperfect; whereas alternatives, may be delivered using modelling that
> engenders different types of outcomes. Either way, there's causal effects,
> but if 'drama protocols' gain supports in areas previously associated with
> fields of STEM - then, the dynamics of the environment changes, as do the
> outcomes, imho...  But, the consequence may be that some are defined to be
> the problem; whilst others, defined perfect; and whether and/or how this
> acts systemically, may well, by design, become unknowable..  I did
> historically try to define 'Knowledge Age' but it got deleted
> https://en.wikipedia.org/w/index.php?title=Draft:Knowledge_age&action=edit&redlink=1  So,
> perhaps an 'authorised thinker' might want to try again, as this
> 'information age'[22] has many problems...  some like it, others - not so
> much...   but, i'm not sure if a 'knowledge age'[23] is allowed,
> particularly as historical attempts have resulted in failure. As such, I'm
> led to remember the days when i started, as a young person in the 80s /
> early 90s, a kid - getting parts from people selling their old gear 2nd
> hand; finding software, learning - including, as one of the very few geeks
> in the computer room at school...  those old days, of BBS sites, dial-up
> modems - had a different sort of ethos about it..  I think the cost of
> building LLM supporting machines is about the same, as computers back then
> cost about 2k (AUD) so, putting something together nowadays (dual xeon,
> 256gb+ ram, 72gb vRAM, etc) when using 2nd hand parts, is about the
> same...  not for everyone, but i think, something that should be allowed...
>
> in the end, i think the outcomes have the capacity to be safer, better and
> more useful, for people...  human centric[5], supporting the means for
> people to own their own 'thoughtware'... thereby seeking to ensure an
> alternative to digital slavery, where others own the software that defines
> your mind... made accessible to you, via API as the consumable consumer..
> said to be benefiting, because it's 'free'... I disagree, but, because I
> disagree, I'm poor.  I'd prefer alternatives to UBI, where people are paid
> fairly for good & useful work, but that's still something that MAY be made
> achievable in future...  sadly.  Perhaps, as some suggest, it'll take
> decades..  which, seems to me, to be an awful suggestion..   Whilst most
> define whether a person is 'successful' based upon the characteristics of
> their wallet, their 'identity wallet', which therefore makes it often
> dangerous to talk about, if poor..  the PCI goals defined October 24-25 at
> Stanford University, 2015[24] included; "7. Personal information in the
> digital environment is protected by law and controlled by the individual
> owner.", so, whilst its not been realised from that time, others, who may
> not live to see it happen, have considered such ideals as being
> important..  so, i think, there's still a lot of merit in seeking to
> realise the availability of these softs of options; but overall, i guess -
> that's where it's / i'm at...
>
> There's certainly alot that can be pulled apart / deconflated, and sought
> to be progressed...  I continue to believe 'humanitarian ict' is of
> instrumental importance for our humanity...  So, I hope this helps.
>
> 🙏🕊️
>
> Tim.H
>
> *NOTE: This has been just written, no agents, not alot of 'careful
> drafting', there may be errors whether it be due to autocorrect or similar;
> or, just because i made a mistake or oversight, or other unintended error
> or fault.*
>
> [1] https://www.w3.org/policies/code-of-conduct/
> [2]
> https://lists.w3.org/Archives/Public/public-payments-wg/2016Mar/0177.html
> [3]
> https://web.archive.org/web/20180901130552/http://manu.sporny.org/2016/browser-api-incubation-antipattern/
>
> [4]
> https://lists.w3.org/Archives/Public/public-humancentricai/2023Apr/0004.html
>
> [5]
> https://www.w3.org/Search/Mail/Public/search?keywords=%22Human+Centric%22&indexes=Public&resultsperpage=100&sortby=date-asc
>
> [6]
> https://docs.google.com/spreadsheets/d/1rqYC2E2BDIHBADAT7-9CabawkmYBJpBBf1KJO24D7ig/edit?usp=sharing
> [7]
> https://docs.google.com/spreadsheets/d/19IEgvdvwl_EOGhmIFinVQu4OerRojeje8PaZWGvoO4Q/edit?gid=0#gid=0
>
> [8]
> https://docs.google.com/spreadsheets/d/1jDLieMm-KroKY6nKv40amukfFGAGaQU8tFfZBM7iF_U/edit
>
> [9] https://tailscale.com/
> [10] https://localai.io/
> [11] https://github.com/mudler/LocalAI
> [12] https://github.com/WebizenAI
> [13] https://www.w3.org/community/rww/
> [14]
> https://lists.w3.org/Archives/Public/public-humancentricai/2023Jun/0005.html
>
> [15] https://github.com/pdsinterop/solid-nextcloud
> [16]
> https://web.archive.org/web/20130122051820/http://jeffsayre.com/2010/09/15/web-3-0-smartups-the-social-web-and-the-web-of-data/
>
> [17] https://www.w3.org/DesignIssues/CloudStorage.html
> [18] https://www.youtube.com/watch?v=ReB5UHfKCG8&t=666s
> [19] https://www.mico-project.eu/
> [20] https://www.mico-project.eu/portfolio/sparql-mm/
> [21] https://www.json.org/license.html
> [22] https://en.wikipedia.org/wiki/Information_Age
> [23] old notes:
> https://medium.com/webcivics/a-future-knowledge-age-2e3f5095c67
> [24]
> https://peoplecentered.net/wp-content/uploads/2017/10/Principles-of-the-People-Centered-Internet.pdf
>
>
> On Wed, 25 Sept 2024 at 01:36, Michael Robbins <
> michael@learningpathmakers.org> wrote:
>
>> Sam Altman is touting his vision for centralized AI. Our group must be
>> the vanguard for a radically different approach.
>>
>> [image: cover.png]
>>
>> The Intelligence Age <https://ia.samaltman.com/>
>> ia.samaltman.com
>>
>>
>> The human-centric pathway for AI doesn’t use unbelievable amounts of
>> energy and water in the process. But this doesn’t leave control in the
>> hands of businesses and billionaires, so Altman isn’t promoting it. It
>> hinges upon the the decentralization of computing to local devices and
>> building a radically different architecture for AI on a foundation of data
>> dignity.
>>
>> Instead of supercomputing and server farms, the sane and sustainable
>> paradigm is  building ecosystems of AI agents that represent individuals
>> and organizations. This is also the revolutionary pathway we need to remake
>> learning, digital networks, and democracy.
>>
>> The only way to reign in AI out of control is by building representative
>> governance inside the AI ecosystem. We can’t leave it to businesses and
>> technology has outstripped government. Our AI-powered assistants, built
>> with data that we privately secure away from mega corporations, will become
>> the ante for citizenship and democracy in the digital realm.
>>
>> Michael
>>
>