00:07I wrote a quick talk that is meant to frame the next panel and as an ex-academic goes into
00:16some of
00:16the history and some of the questions I think we should be asking before we leap headfirst into
00:23AI hype so let me start with some definitions and I want to make sure we're talking about the same
00:32thing when we talk about AI because in this height-based discourse often we're not and too
00:39rarely we fail to make time for these fundamental questions whose answers as we'll see shift our
00:46perspective these are basic questions like what is AI where did it come from and why is it
00:55everywhere guaranteeing promises of omniscience automated consciousness and what can only be
01:01described as magic well first answer first AI is more or less a marketing term it's not a technical
01:10term of art in fact the term artificial intelligence was coined in 1956 by the cognitive and computer
01:17scientist John McCarthy this is about a decade after the first neural network architectures were invented
01:25in subsequent interviews after he coined the term McCarthy is really clear about why he invented it
01:30two reasons first he didn't want to include his academic rival in a convening he was hosting that
01:37summer now his academic rival Norbert Wiener had invented the term cybernetics under which the
01:43field was then organized and of course you don't get academic fame by being a disciple you get academic
01:49fame by forming your own field coining your own term now seconds he wanted grant money and he thought
01:57the term artificial intelligence was shiny enough to get money from the US government and the US military
02:03after the second world war when they were spending a huge amount on cold war dominance via computational
02:11infrastructure so two very familiar reasons but neither of them describes a particular technological
02:18approach now fast forward from then in the course of the terms over 70 year history it's been applied to
02:26a
02:27very vast and very heterogeneous array of technologies that bear little resemblance to each other
02:33today and throughout this history it connotes more an aspiration and a marketing term than a coherent
02:41technical approach and its use has gone in and out of fashion in time with funding prerogatives
02:47and the hype to disappointment cycle so to the second question why is AI everywhere now or why in the
02:56last
02:56decade did it crop up with such renewed force and become the hot new thing that's going to save the
03:02world
03:03well there are very interesting uses of the thing we're calling AI now recognizing patterns in large
03:11amounts of data can help distribute insights and can help us understand our world assuming many conditions
03:19are met but if we're going to be materialist about this answer if we're going to look at political economy
03:25we
03:26need to face the toxic surveillance business model and the big tech monopolies that built their empires
03:32on top of this model now to go back in history again the roots of this business model can be
03:39traced to
03:40the 1990s as scholar Matthew crane's work illuminates in a rush of enthusiasm to commercialize network
03:47computation the Clinton administration laid down the rules of the road for the profit-driven internet in 1996 and in
03:55doing so they committed two original sins sins that we are still paying for today first even though
04:02they were warned by advocates and even agencies within their own government about the privacy and
04:08fundamental human rights concerns that rampant data collection across insecure networks would produce
04:13they put no restrictions on commercial surveillance none private companies were unleashed to collect and
04:21create as much intimate information about us and our lives as they wanted far more than was permissible
04:27by governments and of course governments then piggybacked on top of these companies to gather that data
04:33anyway as the Snowden documents revealed and in the US we still lack a federal privacy law so this is
04:39still the rule of the road now second second sin is that they explicitly endorsed advertising as the business model
04:48of the
04:48commercial internet fulfilling the wishes of advertisers who already dominated TV and print media and didn't want
04:53to lose another market right this combination was and is poison because of course the imperative of
05:02advertising is know your customer in service of identifying the people most likely to be convinced to buy or do
05:09the
05:09things you want them to and to know your customer you need to collect data on them so this incentivized
05:15mass surveillance which now feeds governments and private industry well beyond
05:18advertising with strong encryption such as what signal develops as one of the few meaningful checks on this dynamic so
05:28it's on this toxic foundation that over the course of the 2000s the big tech flat for platforms established themselves
05:35to search social media marketplaces ad exchanges and much more they invested in research and development to enable faster and
05:42bigger data collection processing and to build and maximize computational infrastructures and techniques and
05:48that could facilitate such collection and use such data economies of scale network effects and the self reinforcing dynamics of
05:57communications infrastructures enabled the firms that were early to this toxic model to establish monopoly dominance and the US use
06:07soft power as well to pressure Europe and other jurisdictions into accepting the US model this mass surveillance everything goes
06:15model for commercial platforms which is also
06:18also a comment on the you know Europe stifled innovation by regulation debate which if you look at the history
06:24is not in fact correct so this really brief overview of the history of AI through a material lens helps
06:33explain why the majority of the world's big tech corporations are based in the US with the rest emerging from
06:39China the US got a head start via military infrastructure and neoliberal policies and investment
06:44while China built a self-contained market capable of supporting its own platforms with its own norms for content that
06:51further limited external competition
06:55so here we might pause and reflect maybe ask this is all very interesting Meredith but what does this history
07:02have to do with AI well the answer is it has everything to do with AI because in 2012 right
07:09as the surveillance platforms were cementing their dominance in the US researchers
07:14published a very very important paper on AI image classification which kicked off the current AI boom this paper showed
07:22that a combination of powerful computers huge amounts of data could significantly improve the performance of AI techniques techniques themselves
07:32that were created in the late 1980s
07:36in other words what I'm saying here is that what was new in 2012 was not approaches to AI the
07:43algorithms and the techniques the methods and procedures were very old what changed everything in the last decade was the
07:52staggering computational and data resources newly available to these massive platform companies and thus newly able to animate these old
08:03techniques
08:04or put another way the current AI craze is built on this toxic surveillance business model now this is not
08:11the way it has to be and we can change it but in order to change it we have to
08:16recognize it and understand it clearly this is and while new frameworks and architectures have emerged in the interim the
08:27paradigm of more data more compute equals better AI has not changed
08:33currently there are only a small handful of firms based in the US and China who have the resources to
08:39create and deploy large scale AI from start to finish this is from training to market reach and these are
08:46the cloud and platform monopolies that establish themselves early on the backs of this surveillance business model so everyone else
08:53is life licensing infrastructure they're scrambling for data and they're struggling to find market fit
08:58without massive cloud infrastructure without massive cloud infrastructures through which AI can be sold to customers like Azure API for
09:05instance or massive platforms like Facebook or Instagram into which AI can be integrated for ad targeting or as an
09:14advertiser service and this is why even the most successful AI startups are ultimately ending up as barnacles on the
09:21hull of the big tech ship relying on these actors for infrastructure and the kind of market access capable of
09:28deriving a profit from these very very very capital intensive techniques now it doesn't have to be this way computational
09:38technology is very cool it can do many things and we don't have to accept the paradigm and the framing
09:44that has been handed to us by a handful of platform monopolies selling their surveillance derivatives as intelligence as capable
09:52of guiding
09:53of the underlying governments institutions and our social relationships and our social relationships but again in order to change that
10:01in order to build new models of AI in order to address the pathologies of concentrated power in the hands
10:09of US companies we have to face it and we have to I say atone for those sins of the
10:1790s walk back the surveillance business model and walk back the idea that bigger is better in the 90s
10:23is always a recipe for more performant AI.
10:27So I'm going to leave my talk there
10:29and let that usher in the next panel
10:32where I will then discuss with some AI leaders
10:35and some great AI thinkers
10:36how we might approach this set of problems
10:39and I appreciate you all for hearing me out.
Commentaires