Some thoughts on the recent @NewYorker article, 'Who should stop unethical AI?' [a thread!] https://www.newyorker.com/tech/annals-of-technology/who-should-stop-unethical-ai

First, @luke_stark has an excellent thread that highlights some important criticisms of this piece (especially concerning what is not mentioned in the article, like the firing of Timnit Gebru) : [1] https://twitter.com/luke_stark/status/1361308435585851392?s=20

Looking more at the content though, Michael Kearns being quoted as saying AI/ML research is at 'a little bit of a Manhattan Project moment', besides being hilariously understated, reminded me of Bertrand Russell's 1946 essay 'Philosophy for Laymen'. [2]

(This was actually the first work of philosophy that I read when I started my formal education!) [2.5]

In it, he suggests that, while science helps solve problems involving 'mastering natural forces', when it comes to insuring the increase of human happiness / well-being, 'the laboratory can give no decisive guidance'. [3]

He goes on to say that something more than just technical skill is required here; namely wisdom.

And, you wanna know who claims to be real good at wisdom? That's, right: philosophers! They love wisdom! [4]

So, he suggests that philosophy should be part of 'a general education'. And, this is something appears to be echoed in the present day. (Of course, philosophy is not a solution to all of AIs ethics problems, but it is one of many relevant fields of study that can be useful.) [5]

Personally, I took NeurIPS' 'broader impact statement' seriously last year when I submitted a paper. I think this is an excellent exercise in self-reflection. But, it is unsurprising to hear that some take it to be a joke. [6]

In some sense, it is a joke: it is asking people to utilise a skill which they have probably not cultivated. But then it is not a joke because it's a bad idea, but because it is asking a lot (if taken seriously). I think this is actually really difficult to do in earnest. [7]

That said, it is reassuring to hear from @IasonGabriel that the statements received were surprisingly good, although maybe that is a low bar? In any case, he would know better than me, and so there is good reason to believe that is true. [8]

Broader impact statements and other similar mechanisms are one of (yet to be) many possible tools that might be used to aim the field toward ethical research. But what is a tool without training in how to use it? [9]

Imagine handing someone the keys to a forklift and saying 'the onus is on you to figure out how to use it'.

The point is for changes here to be effective, they probably have to be systemic. And that's not an easy task. [10]

Getting back to Russell : 'it supplies an antidote to the anxieties and anguish of the present, and makes possible the nearest approach to serenity that is available to a sensitive mind in our tortured and uncertain world'. And that's probably what we need. [Fin]