Eighteen forum members decided to take on the machines at this forum. We had a surfeit of researchers, authors and writers on the subject of artificial intelligence (AI) round the table which made for swift and pointed remarks. Some will remember our 2014 event with Carl Benedikt Frey, "The Future Of Employment: How Susceptible Are Jobs To Computerisation", where the machines would take our jobs. This was worse. The machines might take our lives (sic).
First off, on the limits to AI, was James Tagg, speaking on “Ten Things An AI Can’t Do”. James ranged widely, but the gist of his argument was that limits to Turing machines applied to machinery, while humans had proven they could create things, most pointedly proofs of algorithms, that machines couldn’t, solving Diaphantine equations being one good example. James drew heavily and swiftly on his book, “Are the Androids Dreaming Yet?: Amazing Brain. Human Communication, Creativity & Free Will”, which has had excellent reviews.
As a riposte to James, Trent McConaghy led a discussion on “Are We Neuron-Narcissists?”. Trent reminded everyone of the great advances made in several areas of AI recently. He took the whole brain emulation ideas emphasised in “The Age Of Em” as likely, but moved on to consider a range of technologies that held promise for even more rapid advances. An article circulated before the event, “The Empty Brain - Your brain does not process information, retrieve knowledge or store memories. In short: your brain is not a computer”, was roundly criticised as setting up a straw man argument. One wonders what the group would have made of Elon Musk’s announcement the following week that “The chance we are not living in a computer simulation is ‘one in billions’”. The conclusion of the discussion was that mixed model technology (digital and some other substrate, possibly biological) could possibly achieve intelligence while some of the arguments against pure silicon Turing Machines achieving generalised intelligence might be valid. Trent has set out his thoughts out more extensively in, "The AI Existential Threat, and the Bandwidth++ Scenario".
For a flavour of the quality of the conversation – “The halting argument shows that the set of provable statements cannot be a recursive set. The bag of proof argument, however, only assumes that the set is recursively enumerable. The difference is that for a recursive set, there must exist an algorithm that tell you, eventually, whether an element X is in the set or not in the set. A recursively enumerable set, however, only requires that, if X is in the set, there exists an algorithm that tells you it's in the set. If X is not in the set, then the algorithm need tell us nothing about it (in fact, a set is recursive if and only if it's recursively enumerable and its complement is as well). Many interesting sets are recursively enumerable (but not recursive). For instance, the set of Turing machines that halt is recursively enumerable. You can run every possible Turing machine (via interleaving), and keep a list of every one that halts. This will eventually list any Turing machines that halt (hence the set is recursively enumerable), but for the ones that haven't yet halted, you don't know if they will eventually halt or go on forever (hence the set is not recursive). Matiyasevich's theorem is about this as well. It demonstrates that every recursively enumerable set (alternatively called computationally enumerable) is Diophantine.”
It was probably fair to say that the group was about equally divided into believers in strong AI, disbelievers, and undecideds. Along the way, the group touched on some immediate issues in AI, ranging from automated financial advice (robo-advisors) to the limits of smart contracts in mutual distributed ledger (aka blockchain) technology. As the group headed to Furniture Makers for their 'bank-quet', they took goody-bags with James’s book and our dinner speaker’s book, but also a t-shirt, inspired by Trent, displaying "Recovering Neural Narcissist" in lurid pink scroll with glitter gloss on a black background. Tasteful.
To round things off, our signature cocktail for the day was chosen by an AI, IBM’s Watson, an interesting vodka tonic with cranberries and olives. Having swiftly downed a couple we were delighted to have a fascinating conversation over dinner about Al(bert) Robertson’s first science-fiction work, Crashing Heaven. Al created four classes of AIs in his book, all interacting with humans in co-creating a future. Best quote?
“You’re a pattern of memories running on a dynamic platform that’s constantly renewing itself. The pattern is all that persists, the self looking back on all it has been and knowing itself from that. That’s what makes you you, Jack, not the passing fact of your flesh. And that’s what makes me me. I may be running on a different platform, but the pattern of me is unchanged and I fight hard to protect it. I am Andrea, Jack, I’m the same person as that different person all those years ago, just as you’re the same person as that different Jack who loved me then.”
We've all read bits of similar stuff here and there - Gibson, Egan, Stephenson - but it’s nice when it really does all come together in a great tale, and with a few new bits thrown in, such as the lares & penates status of Stookie Bill. Those who had read it loved this book and will be buying the sequel when it comes out.
Over dinner, Michael started a conversation on the neighbourhood around Austin Friars with its Dutch connections, but then managed to turn somehow to issues of patterns & parsimony. The conversation turned interestingly to shibboleths that might be needed either to identify or even control (kill switches) AIs. Michael closed with an interesting tale of the shibboleth used in London for centuries around the Stalhof or Steelyard to distinguish German-speakers from English-speakers, "cheese & bread". Even the word order was part of the shibboleth. At the end of a convivial dinner, members and guests left happily knowing much less about their own intelligence, but much more about each other.