Wondering how humanity’s last champion of Go, one of the most complex games of strategy ever devised by our species, fared against’s Google all-conquering overlord?
Current world champion Ke Jie was already on the ropes last week after being ground out in the first game by AlphaGo, the machine-learning powered AI from Google’s DeepMind laboratories. And that didn’t factor the fact that AlphaGo had already trounced Ke Jie, and practically everyone else, earlier in the year under the online account of “Master”.
Appropriate.
But in the final two matches, the gap between humanity and AlphaGo became explicitly pronounced. It was one predicted that it would take a decade before AlphaGo would be capable of beating professional Go players.
Back in the real world, where Ke Jie resigned against AlphaGo after a four hour battle in the third game, the Chinese world champion bleakly told the South China Morning Post that he would never be able to beat AlphaGo in his lifetime, and that he “made several bad moves, moves that I regret, because I wanted to play well too much”.
Oh dear.
On the bright side, DeepMind announced that AlphaGo would be “stepping back from competitive play”. It’ll be used predominately in the Go world going forward as a teaching and analysis tool, since there’s nothing left to achieve besides sending an entire generation of Go professionals into a spiral of manic depression. DeepMind will also be publishing an academic paper on how they improved AlphaGo’s efficiency, and how it can be adapted for other applications.
Maybe one application for AlphaGo could be finding ways to cheer Ke Jie up. Bloke looks like he needs it.
Comments
12 responses to “Humanity”
In news outside of the beginning of Skynet’s takeover, the first gameplay reveal for Redhood in Injustice is out. Looks quite rad.
In before Terminator DLC for Injustice
(deadset how good would that be though)
Would honestly take Arney over Subzero any day.
A.I is the next logical step in evolution, right? There’s no rule that says life HAS to be cellular (ignoring religious hokey pokey).
Most Sci-Fi tends to say that machines have little to no evolutionary capacity and that they will eventually try and recreate humanity in order to learn how to change. From a practical standpoint, this makes sense because an AI is just programming, not a complete simulacrum for a biological system. It still needs a processor, electricity and some kind of shell. It also needs to protect itself against corruption and deterioration just like we need to protect against viruses and the effects of aging.
Cellular evolution works because mutations in genetics give rise to new pathways of possible stronger organisms. As a machine though you need to decide what those mutations should be so you are always limited by the capacity of your ability to think of them. You could try randomised algorithms but again, you’re just imitating cellular evolution.
I’d say the existence of machine learning and genetic algorithms show that evolving machines are already here. While certainly inspired by biological life, machines do not inherently require biology to exist or improve – they can be just as capable of self-improvement from purely random “mutations”, or more directed changes.
Of course machines still currently require very specialised support environments to function, and are nowhere close to biology’s sheer depth and diversity yet. But there’s no evidence to suggest that this will never be the case, and the remarkably rapid improvement of neural learning algorithms suggest that, at least in software, AI that can “think” in useful ways may be far closer than we thought, even if this does not much resemble human thought.
Yeah but evolution only really look so good because it gets such a long time to work on the problem. Biological evolution is great for eventually coming up with innovative new forms, but it’s not especially good at optimising a design.
Remember, evolution is about coming up with the best individual, to fit a specific niche, using the least amount of energy, and with the least amount of change.
Evolution is a slacker. It does the absolute bare minimum to get the job done.
In other words, maybe having two thumbs and four fingers would have given us a real benefit, but one thumb was good enough so there’s no pressure to evolve a second thumb. Evolution can’t jump over the “meh, near enough” dip in a cost-benefit graph.
Also, we can’t sit around for 5 million years waiting for evolution to say “maybe the robots could go to space or whatever?”.
So maybe the solution is to figure out biological evolution’s “algorithm”, apply that to the machines, and run “evolution.bat” at a massively accelerated rate.
Then, when evolution comes up with the weird, out-of-the-box solutions, we then use our intelligence to refine them – like, if we’d been able to tweak evolution’s solution for the human eye, we would have moved the optic nerve so it doesn’t create a blind spot. Etc.
Humanity had a good run.
The machines have won. Humanity should just let it go.
Sidestepping the usual apocalyptic opinions on AI, it’s interesting to think that our next evolution could come from creating an intelligence greater than our own, and then having it teach us how to be better as a species.
Yeah. I think augmented intelligence is the key to survival as a species. We’ve travelled so deep into the most important scientific fields that it now requires decades of education in order to be in the position to potentially make progress. With fewer and fewer people who posess the creative type of mind even being able to comprehend the problem, the less likely we are to actually solve it.
With AI doing the heavy lifting and humans handling the creative, outside the box thinking we could achieve some great things. Just look how far the library of information on the internet has pushed us. Imagine that with the capacity to do more than just reference work.
I, for one, welcome our new AI overlords.