The Blurb On The Back:

Recent startling successes in machine intelligence using a technique called ‘deep learning’ seem to blur the line between human and machine as never before.  Are computers on the cusp of becoming so intelligence that they will render humans obsolete?  Harry Collins argues we are getting ahead of ourselves, caught up in images of a fantastical future dreamt up in fictional portrayals.  The greater present danger is that we lose sight of the very real limitations of artificial intelligence and readily enslave ourselves to stupid computers: the ‘Surrender’.

By dissecting the intricacies of language use and meaning, Collins shows how far we have to go before we cannot distinguish between the social understanding of humans and computers.  When the stakes are so high, we need to set the bar higher: to rethink ‘intelligence’ and recognise its inherent social basis.  Only if machine learning succeeds on this count can we congratulate ourselves on having produced artificial intelligence.  

Thanks to the Amazon Vine Programme for the review copy of this book.

You can order ARTIFICTIONAL INTELLIGENCE by Harry Collins from Amazon UK, Waterstone’s or Bookshop.org UK.  I earn commission on any purchases made through these links.

The Review (Cut For Spoilers):

Harry Collins is a sociologist and Distinguished Research Professor at Cardiff University’s School of Social Sciences.  This thought-provoking book takes a deep dive into what we mean by ‘intelligence’ and what it takes to pass the Turing Test, arguing that despite extraordinary developments in artificial intelligence, the Singularity is not at hand but we are in danger of fooling ourselves that it is and thus surrendering to ‘stupid’ machines.

I picked this up after reading the recent news reports about an ex Google engineer who claimed that a Google AI had achieved sentience.  I’ve been reading about AI off-and-on for the past few years as it’s a subject that I find interesting (even though I don’t have the scientific/engineering background to understand all of it).  

Collins comes at this from s sociological perspective (which is something I do have some familiarity with) and as someone who was deeply embedded within the physicist community exploring gravitational waves (to the point that he passed a Turing Test on it such that other physicists thought he was an expert in it) he is able to discuss what we mean by language and knowledge.  His basic arguments are as follows:

– no computer will be fluent in a natural language, pass a severe Turing Test and have full human-like intelligence unless it is fully embedded in normal human society; and

– no computer will be fully embedded in human society as a result of incremental progress based on current techniques.

Essentially his point is that computers are not sufficiently socialised into society to be able to pass themselves off as human.  They do not “understand” in a way that people do but instead are trained in pattern recognition that doesn’t work when confronted with the contextualisation and natural corrections that humans make when communicating with each other.  Collins runs through what he means by this, taking into account claims made by technologists such as Kurzweil about consciousness and to explain why he believes that AI is not analogous to the human brain.  

Collins also warns about how the way in which humans are enamoured with the vast progress in machine-learning also makes us susceptible to falling prey to what are actually quite ‘stupid’ computers that we think are solving difficult problems when in fact they haven’t done so at all.  His view is that this is a particular danger because engineers have not resolved the issue of social context and socialisation with the result that the AI is not sufficiently embedded within society to fully understand it and can only become accustomed to it as what it considers to be mistakes are corrected on a case-by-case basis (rather than understanding from the off that they are not necessarily mistakes).  What’s important is that Collins is not saying that this is impossible completely, just that it’s highly unlikely given where the technology currently is.

In constructing his arguments, Collins runs through what impossibility claims are and what Ai consists of and where it currently is (bearing in mine this book was published in 2019).  He then goes on to look at language and how we ‘repair’ misspellings and mistakes in order to make sense of sentences, which is something at AI cannot possibly do.  He moves on to examine what we mean by “context” and how humans make sense of context when they communicate and then goes on to look at imitation games, how the Turing Test works and how you can build a strenuous one to stress test at what stage AI is at and where it might go.  I thought these sections were particularly interesting because prior to reading this I didn’t know that there were different strata of Turing Test and what each represented and Collins draws on his own experience of being embedded within the gravitational wave group of physicists, which really works well to draw out his points.  Also interesting is how Collins takes into account AI programmes such as those that won at Go, Deep Blue and the programme that won at Jeopardy and how he demonstrates that clever those these machines are, they’re not proof that the Singularity is within reach.

All in all, I found this to be a really interesting read that furthered my own understanding of what AI is, where its deficiencies are and what the possibilities are going forward.  If you have an interest in AI and machine learning then I definitely think that it’s worth a look.