Saturday, March 12, 2016

Remember this date – 12 02 2012 - the day our species lost to AI - but won

This is worried look of Lee Sedol and he was in the match of his life - he lost. But it wasn't Lee who lost, it was us all or is it really a win for us, the certaors of AlphaGo? The Human v machine sparring that has been going on for some time with checkers, Chess and Jeopardy, were featherweight contests. This was the Big Fight and we, as a species, got thumped. There’s a new breed of champion in the ring and it’s not just smart, it’s a superfast learner, even its own teacher. It’s eats up human expertise for breakfast, then the real game begins, as it uses this experience to play itself, as it’s the only opponent worth playing. Having learned from us, it sucks our experts dry, then transcends their abilities to boldly go where no brain has gone before.
Momentous moment
This is a momentous moment. In less than 2.5 years, since Google snapped up Deepmind for a mere half billion dollars, 2500 years of human experience and expertise at GO has been trounced. But this is only the start. Software that learns is exponentially more powerful than software that has to be written by humans. Given the huge processing power of Google Cloud Services, AlphaGo has one of the greatest engines on the planet under its hood. It also has some of the best algorithms and that’s what matters. Machine learning algorithms are like small Gods. Free from the tyranny of time and space, speed is no limit. They can learn faster than any of us. These algorithms are the new DNA of progress. This machine moves beyond teaching and teaches itself. That’s essentially what humans do as they become expert learners, few in the later years reply any longer on teachers, as we’ve learned to learn for ourselves. AI just moves to abandoning the teacher faster.
A bit of history
Throughout our history as a species we have always benefited from the delegation of the mundane. This has largely been achieved through technology. We conquered the planet through technology. First through stone then metal tools, needles for clothing, tools for agriculture and so on. Then we invented machines that to did the manual work and we moved from the fields to factories. Then we mechanised the factories and moved towards mental work. Now we’re delegating the drudgery of some of that mental work to machines or, more accurately – AI, even more accurately to machine learning.
Teaching and learning
Amid all the hubris that surrounds education and teaching, there’s a deeper problem. Parents know it, learners know it, even teachers, lecturers and trainers know it. Performance has plateaued and everyone is getting a little fraught. Politicians, driven on by the poor foundations, and therefore learning tower of PISA results, demand more testing. Parents, the most conservative of lobbyists, demand more schooling. Teachers scream ‘enough already – we’re exhausted’. Well, isn’t it about time we looked for the sort of solutions that gave us the industrial and information revolutions of the past? Can’t machines solve the problem of teaching?
Teaching trumped by learning?
Could teaching be trumped by a learning machine? Are we beginning to glimpse the possibility of machines that teach themselves to teach? They learn what works, what doesn’t and deliver ever better performance. We see the embryonic evidence for this in adaptive learning systems, that are truly algorithmic, and do use machine learning, to improve as they deliver. The more students they teach, the better they get. They even tech themselves. This is not science fiction. This is real AI, in real software, delivering real courses, in real institutions. The future has been here for some time it’s just not distributed.
Teaching free like search?
Imagine what will happen when these super-teachers are commoditised, delivered from super fast cloud-services and let loose on the web? Teaching and learning will be as free and accessible as Google search. You will not only be able to find things with ease, you’ll be able to learn them with ease. We may see dramatic rises in performance among learners, right across the board, as such systems will be far more sensitive to individual needs, even learning difficulties. Who is likely to deliver such as service? Well Amazon are on the march, Gates has been seriously funding this stuff but Google is the front runner.
Future without teachers?
This may see hopelessly utopian. But could we have a future without teachers? Why not? Teaching is essentially being a conduit. It is a means to an end, not an end in itself. Wouldn’t academics really prefer to do pure research and not teach? Wouldn’t most teachers prefer not to have to mark anything and avoid the stress of the classroom? Couldn’t we dispense with teaching and just have learning?
Probable, improbable or impossible?
Agricultural workers were largely mechanised out of the process by machines. factory workers by robots, secretaries by word processors and It looks likely that we will see the obliteration of drivers, cabbies and truck drivers, through driverless cars. No one predicted that! There’s a lot of evidence to suggest that many professions, even white collar, middle-class professions, may be replaced by smart AI. So what’s so special about teaching? If we can teach millions, of not hundreds of millions at cents per learner, isn’t that desirable?
Remember this date

So remember this date – 12 02 2012 – it sounds almost providential. It may go down in history as the day we lost our several million year long reign as Champions of the World, not to the super-smart Frankenstein we created, but to the machine teachers who help us learn to be better humans.


shackletonjones said...

Hi Donald. It's a fun thing to worry about, I agree. Three areas where I would take issue with your post:
1) AI is learning to do some of the things people do - but it is doing them in a fundamentally different way. AI computation is completely unlike human computation. That means that sometimes it really is better than people (like in the chess/GO) examples, but other times it is effectively crowdsourcing the problem back to people because it can't do it (e.g. language translation). Things in the middle (pattern recognition) AI struggles with. I've written more about that here:

2) A world without teachers - surely that's a typo. You mean 'a world without learning'. The whole point of AI is that it mitigates our need to learn (in the sense of education as opposed to play). The idea of AI super-teachers is bizarre - like the notion of someone trying to memorise everything on the internet. It's there so you don't have to learn it - you just use it. There is a world without learning; it is called 'play'. The things we learn through play (running, jumping, chatting with friends, shaping clay) we do not experience as 'learning'. Like many people I think you assume that AI is somehow here to help us continue doing what we do today. Really it would be better to think of learning as an autonomous agent, shifting from poor hosts (us) to better ones (AI). I have written more about this here:

3) The idea that teaching could be automated exposes a misunderstanding of learning: i.e. a model of learning as knowledge-transfer. The teachers we remember are the ones who care (not going to go over the Affective Context Model again here) and it is unclear exactly how AI might express care (though movies like 'Her' and 'Ex Machina' hint at this). Probably AI can do this - but only by crowdsourcing actual human care, so it's not really being solved by AI. Probably AI, coupled with VR, could provide a optimum level of experience design - but again based on crowd-sourcing.

P.S. in order to post this comment i had to 'select the images with pickup trucks' to prove I am not a robot ;o)

Donald Clark said...

Hi Nick
1. Fully understand this and using several different types of AI myself in projects. There's dozens of different approaches and I've never argued that what AI does is to even mimic the brain. Most of it focuses on solving problems. Incidentally, some aspects of AI really do attempt to do what brains do, but that's a long scientific and philosophical debate. Sure there's crowdsourcing but even underlying that approach is some very smart search, sort and now semantic algorithms at work. On pattern recognition - facial recognition is getting good, fast and other stuff will follow.
2. There is a point here about what we as humans do when most of the mundane tasks are done by machines, and by mundane I mean most professional management jobs. Although, having been involved with AI driven adaptive learning for several years, I've seen it paly the role of teacher across a range of subjects.Far from seeing AI helping us do the things we do today, it certainly will eliminate many of those things - driving and so on. Learning has already shifted, as you say, from us to them, but it's a los more complex than this. The work we've been doing in the US shows that optimum teaching with a good 'teacher and AI gives optimum results.
3. I am not assuming the 'transfer' model.Indeed, I gave a talk on Thursday on your very point - Little Ice has already done HER in rela life - a massive Turning test (affectively).the VR component introduces the emotional aspect - see Henry & Oculus. You have a good point about not replacing humanoid 'teachers' and I agree. We didn;t go faster by copying the bones structure of a cheetah but invented the wheel. This is not about being human, it is about 'learning more effectively. Software already does this and will do it better and better. The difference is that the performance of that software also 'learns' and gets better and better, whereas we remain in the slow lane.