Alphabet Inc. (formerly: Google)

Just finished watching the 3rd match between AlphaGo and Lee Sedol.

As someone who has spent months coding his own program, it is really impressive to see AlphaGo in action. Unfortunately, I was hoping to see AlphaGo losing a match and then figure out how to improve its own game without further human intervention. This would have been much more exciting. Perhaps other professionals might wish to have a go (pun intended) at the alGOrithm. 

On the other hand, AlphaGo allegedly has 1920 CPUs and 280 GPUs. This still sounds like using a very resource intensive, sledge hammer approach. Humans are unfortunately, emotional and get tired rather easily too. Although we are still faraway from creating real intelligence, hopefully, the research can result in some real-life problems being solved.

A masterful 78th move that AlphaGo has no answer for. Ah.... the unpredictability of humans.....

One other interesting point is that Lee Sedol mentioned that AlphaGo appears to prefer to play white. Is this a bias unconsciously built into the programming or it is the Achilles heel of AI? Time will tell. {Under the rules of Go, black plays first}

The end-result of the match. Not bad for Google DeepMind...

Google AI beats South Korean Go pro, ends series 4-1
15 Mar 2016 17:32
[SEOUL] Google's artificial intelligence (AI) programme beat South Korean professional Lee Sedol in the ancient board game Go on Tuesday, recovering from Sunday's loss to win its fourth match in a five-game series.
...
REUTERS

Source: Business Times Breaking News

(15-03-2016, 08:21 PM)CityFarmer Wrote: The end-result of the match. Not bad for Google DeepMind...

Google AI beats South Korean Go pro, ends series 4-1
15 Mar 2016 17:32
[SEOUL] Google's artificial intelligence (AI) programme beat South Korean professional Lee Sedol in the ancient board game Go on Tuesday, recovering from Sunday's loss to win its fourth match in a five-game series.
...
REUTERS

Source: Business Times Breaking News

Yeah, not bad. But not good too, if you consider that there are 15 members in the Google DeepMind team that accompanied AlphaGo to the tournament (they appeared on stage during the press conference). 

Hopefully, we don't have to wait too long before all that algorithm can be condensed into 1 PC...

(15-03-2016, 09:35 PM)HitandRun Wrote:

(15-03-2016, 08:21 PM)CityFarmer Wrote: The end-result of the match. Not bad for Google DeepMind...

Google AI beats South Korean Go pro, ends series 4-1
15 Mar 2016 17:32
[SEOUL] Google's artificial intelligence (AI) programme beat South Korean professional Lee Sedol in the ancient board game Go on Tuesday, recovering from Sunday's loss to win its fourth match in a five-game series.
...
REUTERS

Source: Business Times Breaking News

Yeah, not bad. But not good too, if you consider that there are 15 members in the Google DeepMind team that accompanied AlphaGo to the tournament (they appeared on stage during the press conference). 

Hopefully, we don't have to wait too long before all that algorithm can be condensed into 1 PC...

IBM's Watson, isn't run on 1 PC too. 

Technically speaking it's fifteen guys vs one. It's like using an assistant machine for counting cards in casino.

However it does show that go is not that special of a game and can be broken down to a series of numbers just like chess. So what we call strategy games are just predictable numbers and chance.

Really wonder how much artificial intelligence involved vs just plain brute force computation

Sent from my MotoG3 using Tapatalk

(15-03-2016, 10:07 PM)CityFarmer Wrote: IBM's Watson, isn't run on 1 PC too. 

I think I'm rather optimistic. It should be a matter of time that the hardware catches up and the software improves. Just hoping that it will be faster. 

P.S. I might be biased but I suspect that the sophistication of Deepmind should be several times higher than that of Watson.

(16-03-2016, 06:18 AM)BlueKelah Wrote: Technically speaking it's fifteen guys vs one. It's like using an assistant machine for counting cards in casino.

True (without even considering the rest of the team back in UK). That are further suggestions on how to level the playing field but Deepmind Alphago's team doesn't appear to be interested in a re-match yet. Perhaps there are more serious stuff (like making money  ) on their minds.

(16-03-2016, 06:18 AM)BlueKelah Wrote: However it does show that go is not that special of a game and can be broken down to a series of numbers just like chess. So what we call strategy games are just predictable numbers and chance.

Really wonder how much artificial intelligence involved vs just plain brute force computation

Theoretically, it's possible to simulate that complexity. However, the beauty of AlphaGo is that it was not based a brute force kind of monte carlo simulation. There are neural networks that assisted Deepmind in making decisions. Incidentally, Deepmind even has an ethics committee (being condition of the sale to Google) to govern its future use in real-life applications.

(16-03-2016, 06:18 AM)BlueKelah Wrote: Technically speaking it's fifteen guys vs one. It's like using an assistant machine for counting cards in casino.

However it does show that go is not that special of a game and can be broken down to a series of numbers just like chess. So what we call strategy games are just predictable numbers and chance.

Really wonder how much artificial intelligence involved vs just plain brute force computation

Sent from my MotoG3 using Tapatalk

These fifteen guys are not GO experts.
They are computer scientists and developers.

Against Lee, even a hundred normal GO players working together will probably not be able to beat him.

Even Lee, he had his own team of GO experts for post and pre match analysis of strategy