29 January 2024

Greetings all, and a happy Wednesday! I hope ye all had a lovely Data Privacy day yesterday. I certainly did.

It seems the AI & Human Rights world has woken up after some new year’s slumber, there is a lot of news this week.

The big (massive? seismic?) news this week is obviously the release of DeepSeek, an open source AI model, that was built using far less, less fancy chips than is the norm. The New York Times has a good explainer. The price comparisons I heard were approximately $5.6 million to train DeepSeek, versus $100m to $1billion for the more ‘traditional models’. If there is anything concrete to come from this, there has to be at least some hope that AI development will have a much less horrendous environmental impact. Of course, disruption is never plain sailing, and in addition to the usual hallucinations, DeepSeek also appears to have a penchant for propaganda/censorship and some data safety concerns. (shocked gasp).

Speaking of propaganda (and seamless segueways) UNIDIR are hosting their first conference on AI, Security and Ethics. More details, including on paper submission, etc., on the links. Looks like its shaping up to be a very interesting event.

Also, on data safety concerns (oh! He’s on fire!) this is a really nice explainer on some of the privacy risks associated with AI models. I found it super helpful.

We welcome (!?) the return of the naughty facial recognition corner this week. Ireland is apparently including live facial recognition (and retrospective) in its programme for work, despite LFR previously being ruled off the table. There are, of course, concerns. I know it’s not facial recognition, but there are reports that police in the UK are trialling odour and gait tech. Is that a whiff of concern i sense? (you can thank chatGPT for that one). The naughty corner (we used to call it the bold step) is slightly crowded this week, with reports on how Microsoft depend its ties with the Israeli military to provide tech support during the Gaza war, which raises (at a minimum) some business and human rights concerns. No bold step would be complete, however, without Mark Zuckerberg. Meta announced the end of fact-checking on Facebook, Instagram, and Threads. Along with a side order of more hate speech in the name of free speech. WITNESS have a comment. Interesting to see how those oversight board lads get on at Rightscon. Oh, and Trump revoked a Biden era executive order on addressing AI risks.

Does that sound like a lot of not very good policy? Fear not! If you’re in the UK, an AI tool will now allow ministers to get a ‘vibe check’ on whether MPs will like policies. I have to say. Calling it a vibe check is a vibe all of its own.

Well thats it from me this week. I’ve a few more hours in the office, and then away for a long weekend under blue skies. Yay.

Huge thanks to Ameneh Deshiri for help with the newsletter over the last while.

I’ll leave ye this week with ‘Five Hundred Miles’, not by the Proclaimers (sorry), but by Mamman Sani. The Synth King of Niger. It’s beautiful i think. And so are you all. Have a lovely week.