#aise26 — Public Fediverse posts

Apropos @[email protected] blogpost, after attending the @[email protected] #AISE26 at the inability of participants to communicate (with each other or the stage) thought about making a conference comms software startup that sneakily replicates early twitter (about 2012).

bsky.app/profile/j2br...

RE: https://bsky.app/profile/did:plc:tkbweudpy6tvzjqdiza4z3p5/post/3mqj4m6yqk22w

yet another #AIES26 panel. A Russian is talking about RF laws for LAWS (lethal autonomous weapons) which is hard to take seriously given Putin was claiming there were no humans in the loop controlling Russian AI tanks about a decade ago.

An Egyptian praises his colleagues and says we can't have fragmentation or the superpowers will exploit us. My final #AISE26 thread here https://mastodon.social/@j2bryson.bsky[email protected]/116777451343830502 because it was all pretty big news.

Argh, sounds like #g7 came down on the side of #ageVerification, which is a disaster for children and democratic discourse, cf the hashtag (if you click it you can ask for my previous posts). #AISE26 See also this academic letter on the topic edri.org/our-work/ope...

Open letter: The dangers of ag...

It sounds like the fourth initiative announced by Anne Lazar Sury will be at @[email protected] ? A centre of excellence here for the #AISE26 topic #AISecurityEthics ? Makes me more concerned about the relatively heavy US (and also British) hand at this meeting (cf @[email protected] )

I have always thought we need regions to differentiate, not only because of their different capacities and threats, but also because my greatest concern is regulatory capture by near-global transnational entities. But Egypt's biggest concern was people claiming exceptionalism and evading law #AISE26

This is a "mini panel" not in the programme. There was another such just before where Russia, Brazil, and Egypt just before where Egypt particularly but I think all of them insisted that there should not be regional innovation on AI governance, so it's kind of funny to hear this now #AISE26 1/2

Woah, France also just announced at least four major, expensive initiatives including a centre for excellence on ai governance in the military domain, but the other three for france. #AISE26

New Chinese AI Governance white paper, from two days ago english.scio.gov.cn/whitepapers/... #AISE26 Chinese ambassador just talked about it, seems to carry on from the French/Indian #AIAction philosophy, not the transhumanist/regulatory evasive #AISafety one.

Full text: More Just and Equit...

OK, Chang is also on board with having audits of AI so we can ensure we know who should be accountable. #AISE26

Artem Lomakin: all kinds of tech was innovated & tested in cold war europe, including the internet.

Just realised I'm friends with the transhumanist I didn't name here :-) I have the worst name & face recognition; I recognised him by his policy eventually.

Argh: "We've noticed, AI does make mistakes." AI IS NOT THE ACTOR (I won't name that guy either) #anthropomorphism #AIEthics

Ha, "civil society is part of our infrastructure" says someone shouting out transhumanist British efforts on AI "safety" and related (e.g. genomic "resilience", which AFAIK means ending aging).

#AISE26

Amy Chang, Cisco "human control is too slow to be meaningful" (ignoring some excellent conversations about this yesterday)

Kazuo Noguchi, Hitachi "AGI is considered coming soon"

I'm glad most people have gone home (Friday afternoon in Europe!)

ha: the best of the lot on the previous panel (in terms of admitting others were doing similar things at the same time), Edelman, presented as being from MIT, but left his soft-money job there in 2022 (again, I only know because he's honest on LI). Meanwhile on that platform, Cristina Caffarra reports a big dissing of EU regulatory capacities, trying to call us "middle powers" cf #AISE26

The information age allows all kinds of collusion, but also collective debunking.

https://www.linkedin.com/posts/cristinacaffarra_europe2031-europeantechsovereigntysummit-share-7473623772151918592-EnhA/

The EU is in par with China in terms of AI, and the rest of the world matters a lot too. The only market the US has cornered is "hyperscalers" which are implicated in undermining its democracies, and are not making money. cf for data through 2024 https://mastodon.social/@j2bryson/115570040758750564 #AISE26 #digitalSovereignty #geoeconomics #propaganda

Now they are making a big deal about having focussed on the problem of nuclear launch by AI. Which again, I was in a 2017 meeting about, in the UK (with people from around the world.) https://www.chathamhouse.org/event/mapping-relative-risks-emerging-technologies-pose-nuclear-weapons-systems #AISE26 I'm not meaning to moan too much, I didn't myself propose a talk, and I'm ASSUMING I'm not that special. But there is really something up about only talking about US software, China as an interesting, runner-up hardware power, & otherwise "middle" or "developing" powers.

Tim Rutherford, Xiao Qian, & R. David Edelman talking about US & China 2018 dialog as if those two countries were running everything & were early (though Edelman inserts that other things were going on) BTW, the OECD had just finished a 5 year process of meetings then (I attended 2017 & 2018); the UK had national level softlaw in 2011. I personally was in the UN basement with US, CN, and several EU countries & two of GAFAM in 2019, thanks to the Stanley Foundation https://stanleycenter.org/publications/militarization-of-artificial-intelligence/ #AISE26

OK, I can't complain too much, Boinodiris is now doing a great job of lecturing people about how you can audit all this stuff. She said "liability washing" which is the first time I've heard that phrase IIRC. #AISE26

Governance with teeth: A live
requirements blueprint for AI in security and defence
Speakers:Phaedra Boinodiris, IBM Consulting #AISE26

She didn't know my name when we met, but she just taught my paper... (as did the former director of UNIDIR yesterday already)

Bias one, of course… https://joanna-bryson.blogspot.com/2017/07/three-very-different-sources-of-bias-in.html

Ha, great conclusion – the computer did not make a mistake, the architecture did. And the footprint of that architecture affects everyone in this room. Ashlie Robinson, Jamaican AI Association, #AISE26

Robinson's talk is awesome, too much to blog all, but she just said Jamaica are seeing disinfo undermining climate narratives & concerning efficacy of regional cooperation. Which is ironic given I couldn't get a question about the African Union through #AISE26 slido moderation to a previous speaker.

Collateral code: How military AI governance
failures disproportionately threaten small island developing States
Speaker: Ashlie Robinson, Jamaica AI Association
Jamaica went dark last year, no comms, and had to reach out to the same private foreign actor that was the problem for help. #AISE26

OFFICIAL ANNOUNCEMENT
Kenya to Host the 4th Global REAIM
Summit
April 2027, Nairobi.
The first REAIM Summit in Africa — and the Global South
#UNIDIR #AISE26

Here's my question presently languishing in moderation: "The African Union, like the EU, has been working to put together its own regional AI regulation. How is that effort going, and will it include military applications?" (For the ambassador of Kenya.) I know the AU is making interesting progress (or at least it was last Autumn), so why doesn't #UNIDIR want to talk about it? #AISE26

Did I complain about this yesterday? Really angry that there's no way to discuss speakers, questions, etc at #AISE26 . I can't find conversation on any of the 4 social media I've checked, there are no coffee breaks, they only have slido and there's no way to reply or converse there, and further, they are very heavy handed in their moderation, so you can't even see what people are really wondering about by reading the questions. #twitter is so missed.

#AISE26 panel on responsibility chain, several panelists are military suppliers,
Alex Tsado, Udu Tech: despite necessity of foreign imports to global majority, he sees having local development also increases understanding & therefore control of foreign tech. (One of my students has shown this, too!)

#AISE26 And we're back! Investors talking about their roles in #AIEthics One guy claims there's this new thing #996 9am-9pm for 6 days a week. Every decade I've been alive that's been a thing, how does each generation think they're new at working hard for a decade? Be sure to take time out at some point, boys.

Humans including military subordinates also have many influences, yet commanders have responsibility.
Can't we procure software similarly responsibly commandable, & attribute responsibility to its commander/operator. The EU #AIA has such language, tho sadly not for military AI #AISE26 #AIEthics 2/2

Human-centered proactivity in defence AI: Rethinking agency, judgment and responsibility
Speakers:
Christiana Onoja: Defence Industries Corporation of Nigeria
- Col. Felix Alaita, Kryterion Limited
They say AI responsibility gap emerges from all the many people involved in building it. 1/2 #AISE26

The clock is already running: How quantum breaks the security promises underneath AI governance
Speaker: Maysem Alagamy

Well, yeah. 2 biggest errors @ #AISE26: thinking delegation is abdication (see immediate previous bleet) &, as I said earlier, everyone seems to assume cybersecurity just works.

Now Jingjie He, Chinese Academy of Social Sciences
who points out that the enriched nuclear material was already moved before Trump struck as a convenient example of the importance of remote sensing :-) #AISE26

ouch–terrible take! "the secret sauce of #agenticAI is that humans are not in the loop." Is a General suddenly out of the loop when they delegate to a lower-ranked officer? The General is still accountable, and their agent can still report (& often be interrupted) during execution #AISE26 #aiethics

ai-risk.berkeley.edu/reading-grou... @[email protected] is speaking here now, but I can't find his paper he's citing (please send me the PDF so I can assign it this Fall :-) #AISecurity #AISE26

Reading Group - Berkeley AI Ri...