Great Manifesto. Let’s talk about operationalizing this at the local level!😎(the devil is in the details😈)Regards / Saludos / Grato
Andrés Leopoldo Pacheco Sanfuentes Pronouns: He/Him/They/Them (equal preference) Hi all, Thanks for the discussion. A few months ago I wrote a piece for SSIR advocating for civil society funders and orgs to address tech co power consolidation through community-led governance. This speaks to accountability more broadly than to specifics of the technology/algorithms, so I'm not sure if this directly answers your questions, Kate, but sending it in case it is of interest.
Lina
Kate thanks for bringing up the questions, which make sense But technically, they may be 'ill posed' (imho) That is because there is a mixup and overlap in terminology/concepts/implementations adopting the same terminology applied to different concepts etc
All algorithms are in principle auditable even when they are proprietary, and the only way companies can maintain their competitive advantage is by keeping he algorithms proprietary, or de facto, a trade secret You cannot make any laws against trade secrets afaik Some of these algorithms are useful and amazing even, technically but for example, I started to notice that when I leave a whatsapp message to someone the content of my message is picked and turns up into the adveritisng on FB and in turn via some agreement that I may not know about, it turns up in adverts on youtube, google search etc
To what extent are the search results that I obtain skewed based on my user profile, which is in turn based on my login credentials, which is in turn based on the apps/web services that i use? I would say it's a lot skewed. how so? by a mixture of algorithms , commercial agreements, trade secrets which are all legal
I think one face of the blockchain may be to disrupt this entanglement by encryptions and fragmentation but the reality, is that the master key is only visible to some, and THEY are building the machine, in the name of democratization of the internet, go figure
My advice would be, start auditing individual functions (input-process-output) for each task/app then build the map of the ecosystem entanglement from there, keeping in mind that by means of generative algorithm the map is constantly reconfiguring itself, and not traceable (a property of the blockchain, auch) and NOT REPLICABLE (a property of generative algos)
Very very thorny entanglement, the best we can do is to stay on top of things
(scratching head)
Hi,
I'm trying to understand the lay of the land.
So, generative AI company algorithms are proprietary, like Facebook's and Tiktok's have been all along. Companies still aren't sharing algorithms with researchers, even if they sign a non-disclosure agreement (still true?). If we can't see it, we can't analyze it, regulate it, amend it, or make it accountable. I've always been surprised that people don't leak them.
Companies could be compelled to make their algorithms more transparent if there were a law that requires it, but so far there's no law.
Paola, if your field is algorithmic auditability, do you ever see proprietary algorithms? If so, how?
Also:
Earlier today Lina Khan, head of the US Federal Trade Commission, tweeted: ---
Today @FTC referred its case against TikTok to the Civil Division at ---- As a complete non-lawyer, I was interested to see that TikTok was getting in trouble partly because the FTC believes they are about to violate these laws. Users are about to get injured by Tiktok, predicts the FTC. I didn't know a company could be sued for something it hasn't done wrong yet.
If so, could this apply to generative AI companies?
Is there a lawyer who might answer that question?
-Kate
Yes, that's an interesting idea, Hans.
Former NSA chief Keith Alexander, who has a history of lying about spying on Americans, is on Amazon's board.
-Kate
The case of OpenAI is one instance of a general trend in which national security agencies overlap with IT/media corporations.
The same thing happened at Twitter, I believe:
https://twitterfiles.substack.com/p/1-thread-the-twitter-files
It would be quite useful and interesting for someone to perform some non-partisan research on such ties in general.
Hans Klein
Georgia Tech
Thank you Kate for bringing up this issue here
How do you think this should be tackled? My work is in algorithmic auditablity, awareness and explainability
trying to develop more understanding and possibly standards
Note for Sawsan: I think the reference to the president here was purely related to the person being part of that administration at the time?
Paola Di Maio W3C AI KR CG
We should never place our hopes on company boards functioning in the public interest. The recent debacles at Boeing and Tesla demonstrate this. In Tesla's case, the board and shareholders with meme greed have only indulged Elon Musk, further
bolstering his feudalistic tendencies.
So OpenAI has a conflicted mission, a weak board, an insanely risky goal, and no accountability (am I missing something?). Oh right, their product is evolving at a million miles an hour.
They've shed many of the staff and board members who cared most about safety.
Microsoft, their funder, could reign them in but it is motivated instead to egg them on. And now they've got a board member with very close ties to two US presidents and one of the world's most powerful spy agencies. The keys are on the
table, as Juan Benet would say.
I don't think OpenAI could be getting more press coverage--the coverage has been near-constant and pretty responsible.
Are the NGOs working on this having any luck?
On Sun, Jun 16, 2024 at 12:27 PM Andrés Leopoldo Pacheco Sanfuentes <[email protected]> wrote:
Sorry but “accountability” runs afoul of profit so many times, and the “mission” of OpenAI is DoubleSpeak:
OpenAI is an AI research and deployment company. Our mission is to ensure that artificial general intelligence benefits all
of humanity.
Regards / Saludos / Grato
Andrés Leopoldo Pacheco Sanfuentes
Pronouns: He/Him/They/Them (equal preference)
There is currently no accountability for the decisions at OpenAI, to my knowledge. What has to happen for that to change? The board is not working.
How can the company be held accountable? I'm especially interested in the thoughts of policy people and lawyers on this list. And yes, choosing a spy chief for the board is a big red flag.
I was so happy when Liberationtech was resurrected, and of course a former head of NSA on AI is something that needs to covered and discussed.
However, I hope we’re not quickly degenerating into Trump-this Trump-that (and sensationalizing the title, only to realize the guy “was asked to continue under Biden” buried deep down inside). (!)
Journalists may need to do this kind of (… work..?) to keep their jobs — god knows for how long. Normal people, not so much.
People are working very hard to restore a basic level of trust among family and friends, after the several political and civil abuses of the last few years. Let’s please keep good spirits and stay relevant on the things that we all care
about, and not assume political leanings of others, and that magic words will evoke certain reactions à la Pavlov.
Now, back to discussing OpenAI. :)
(Sorry Kate if that’s too forward. All respect to you, thank you for sharing the article).
PhD student - Geoinformatics
Sam Altman, one of AI's most important leaders--at least for now--is a man with incredible contacts, wonderful social skills, and apparently few scruples. Appointing the former head of the NSA to OpenAI's board demonstrates that this company
is unaccountable. This company puts Americans--and everybody else in the world--at risk.
How can OpenAI be made accountable? The stakes are so high. Its board has already failed to contain it.
Not even the worst part of this, but new board member Nakasone's hobby horse is that the US must out-compete China in generative AI.
ps: What happens at OpenAI if Trump is re-elected?
Washington Post:
OpenAI adds Trump-appointed former NSA director to its board
Paul M. Nakasone joins OpenAI’s board following a dramatic shakeup, as a tough regulatory environment pushes tech companies to board members with military expertise.
By Cat Zakrzewski and Gerrit De Vynck
Updated June 14, 2024 at 12:16 p.m. EDT|Published June 13, 2024 at 5:00 p.m. ED
The board appointment of retired Army Gen. Paul M. Nakasone comes as OpenAI tries to quell criticism of its security practices. (Ricky Carioti/The Washington Po
OpenAI has tapped former U.S. Army general and National Security Agency director Paul M. Nakasone to join its board of directors, the continuation of a reshuffling spurred by CEO Sam Altman’s temporary ousting in November.
Nakasone, a Trump appointee who took over the NSA in 2018 and was asked to continue in the role under Biden, will join the OpenAI board’s Safety and Security Committee, which the company stood up in late May to evaluate and improve its policies to test models
and curb abuse.
The appointment of the career Army officer, who was the longest-serving leader of U.S. Cybercom, comes as OpenAI tries to quell criticism of its security practices — including from some of the company’s current and former employees who allege the ChatGPT-maker
prioritizes profits over the safety of its products. The company is under increasing scrutiny following the exodus of several key employees and a public letter that called for sweeping changes to its practices.
“OpenAI occupies a unique role, facing cyber threats while pioneering transformative technology that could revolutionize how institutions combat them," Nakasone told the Post in a statement. "I am looking forward to supporting the company in safeguarding its
innovations while leveraging them to benefit society at large.”
Amid the public backlash, OpenAI has said it is hiring more security engineers and increasing transparency about its approach to securing the systems that power its research. Last week, a former employee, Leopold Aschenbrenner, said on a podcast that he had
written a memo to OpenAI’s board last year because he felt the company’s security was “egregiously insufficient” to stop a foreign government from taking control of its technology by hacking.
Security researchers have also pointed out that chatbots are vulnerable to “prompt injection” attacks, in which hackers can break in to a company’s computer system through a chatbot that is hooked up to its internal databases. Some companies also ban their
employees from using ChatGPT out of concern that OpenAI may not be able to properly protect sensitive information fed into its chatbot.
Nakasone joins OpenAI’s board following a dramatic board shake-up. Amid a tougher regulatory environment and increased efforts to digitize government and military services, tech companies are increasingly seeking board members with military expertise. Amazon’s
board includes Keith Alexander, who was previously the commander of U.S. Cyber Command and the director of the NSA. Google Public Sector, a division of the company that focuses on selling cloud services to governments, also has retired generals on its board.
(Amazon founder Jeff Bezos owns The Washington Post.)
Until January, OpenAI had a ban on the use of its products for “military and warfare.” The company says the prohibition was removed to allow for military uses that align with its values, including disaster relief and support for veterans.
“Our policies have consistently prohibited the use of our tools including our API and ChatGPT to ‘develop or use weapons, injure others or destroy property,’” OpenAI spokesperson Liz Bourgeois said. “That has not changed.” Nakasone did not respond to a request
for comment.
Nakasone brings deep Washington experience to the board, as the company tries to build a more sophisticated government relations strategy and push the message to policymakers that U.S. AI companies are a bulwark against China.
“We want to make sure that American companies ... have the lead in the innovation of this technology, I think the disruptive technology of this century,” Nakasone said when asked about AI during a recent Post Live interview.
--
Liberationtech is public & archives are searchable. List rules:
https://lists.ghserv.net/mailman/listinfo/lt. Unsubscribe, change to digest mode, or change password by emailing
[email protected].
--
Liberationtech is public & archives are searchable. List rules:
https://lists.ghserv.net/mailman/listinfo/lt. Unsubscribe, change to digest mode, or change password by emailing
[email protected].
--
Liberationtech is public & archives are searchable. List rules:
https://lists.ghserv.net/mailman/listinfo/lt. Unsubscribe, change to digest mode, or change password by emailing
[email protected].
--
Liberationtech is public & archives are searchable. List rules:
https://lists.ghserv.net/mailman/listinfo/lt. Unsubscribe, change to digest mode, or change password by emailing
[email protected].
Liberationtech is public & archives are searchable. List rules: https://lists.ghserv.net/mailman/listinfo/lt. Unsubscribe, change to digest mode, or change password by emailing [email protected].
|
