Google DeepMind – Pebble hanging.

Reply to this post

AlphaGo’s retirement shows that humans are still needed. 

• AlphaGo is hanging up its pebbles after emphatically demonstrating that from here on, machines will be better Go players than humans.
• This move also indicates that despite being one of the most advanced AI’s developed, it will still consume a huge amount of human resources to keep it running.
• Last week AlphaGo crushed the world’s best player Ke Jie 3 to 0 in a convincing display that has left little doubt that human rule of this game is now over.
• DeepMind, the Google owned developer of the AlphaGo, has decided to retire the algorithm and focus on more useful areas such as health, material sciences or clean energy.
• This makes complete sense as DeepMind has proved its point with regards to its AI prowess but also since it published its methodology for AlphaGo, it has already been copied.
• For example, Tencent is very keen to show the it has a strong presence in AI and recently its AI Go player called Jueyi was able to play to a very high standard.
• However, on inspection it appears that Jueyi is little more than a carbon copy of AlphaGo, leading me to completely discount Jueyi as an example of Tencent’s prowess in AI (see here).
• This is possible because AI is a co-operative field and DeepMind has published most of its methodology and results for the creation of AlphaGo in the scientific magazine Nature.
• Most importantly, I think that the retirement of AlphaGo indicates that to keep it going would still require a lot of human time and effort.
• AIs need to be constantly evolved to keep up with how the task for which they have been created is changing.
• Although, AlphaGo was touted as an AIs that could do a lot of learning by itself, the reality was that much of its crucial learning was human supervised thereby consuming resources.
• One of RFM three goals of AI (see here) is the creation of AIs that can build their own models and while there is plenty of evidence that researchers are working hard in this problem, results have been pretty scant to date.
• If AlphaGo could be left to its own devices, there would have been little reason to retire it, but seeing as it will consume resources that can be productively deployed elsewhere, it makes no sense to keep it going.
• This is yet another sign of how nascent AI really is as I think that many of the capabilities which the big ecosystem companies would have us believe are just around the corner, are actually years away.
• I think translators, executive assistants, personal trainers and so on have plenty of time to find other lines of business.