Some opinions on AlphaGo Zero $

Feynman

Interesting, or would be if I understood how AI actually works. I wondered how the cost of AlphaGo would compare to the cost of running, say, YouTube for three days rather than stating it in terms of X typical U.S. homes.

I was also fascinated to see that "Shimon Whiteson’s detailed rebuttal" linked to a handful of Twitter posts rather than an actual paper.

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21

I find that quite low.

And if he find this useless, I wonder how he would feel about LZ. Horrified ?

And bitcoin?

P.S. something I would like to see is an estimate of the energy cost of the EU cookie law. You'd need to account for all the meeting time, development work etc of companies implementing it (and pro rata in the commute time, building heating/AC costs), plus then all the millions of consumers wasting a few seconds every day dismissing the banners (CO2 cost of electricity powering device, of food a human consumes to metabolise). It wouldn't surprise me to exceed 96 tonnes of CO2 a day.

Hell, yes, bitcoin!

At least some day bitcoin will crash.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

The cost/carbon footprint of Alphago is a one-time cost that benefits at least go players. It is about 100 times less than the daily cost of spam.

Be my guest and short it! The best part about speculations about financial markets is seeing whether the other person can put their money where their mouth is

Very good comparison, it puts things in perspective

There are plenty of things that cause much larger CO2 emissions than what Deep Mind does, and have much less to show for it.
At least Alpha Zero contributed to human knowledge in the form of research papers, and much more.
Proof of concepts will always be expensive resource wise, but they're needed to push things forward.

Not only that, but if one wants to start arguing against their wok due to CO2 emissions they've produced, it would be good to not only factor in their published papers, and their "mathematical and scientific" values, but also the added value of publicity, getting more people interested in AI and neural networks, motivating people to tackle other problems that were thought to still be 5-15 years away, etc etc.


Anyways, a rather large amount of dubious lines in that blog post.. for example:
"I don’t have much time for this section because Google’s carbon offsetting scheme is basically a joke but let’s break it down anyway"

"What kinds of real problems can be solved with AlphaGo Zero algorithms?"
"These constraints effectively rule out the application of AlphaGo Zero’s algorithms to any practical problem"
https://deepmind.com/blog/deepmind-ai-reduces-google-data-centre-cooling-bill-40/

I'll say it plain: What a load of bollocks.

I read the bit about the cost, in dollars, and I couldn't understand why that was relevant. Google aren't short of money and spending it on research that was published to the scientific community is not the worst way to dispose of cash.

I then read the bit about the carbon footprint of the experiment and, like others in this thread, I thought the total to be rather low.

Finally, upon reaching the waffle about carbon offsetting, I gave up and concluded that the author's on a crusade... for clicks, most likely.

The way the author describes Google's carbon offsetting scheme is entirely misleading because they make it sound like Google are trying to reduce their footprint by buying *more* energy. I'm no Google apologist (I can't stand their drive to own my data, with or without consent.) but their scheme is pretty much identical to many other big corporations: they pay for renewable energy *matching* their consumption to be fed into the national grid from which they draw their power. (Deutsche Bahn do this, too, for example.)



This paragraph was pretty much the end of my patience with the rant. Google are not obliged to cover the USA's energy requirements in any way and developing AlphaGo doesn't make them so accountable. They pay for renewable energy to be fed into the grid, matching their consumption. What more should they do?

Furthermore, how can you leap directly from a flaccid critique of the fact that they're paying for solar and wind-energy to a hard judgement that they're therefore only virtue signalling? (Google are paying mostly for wind. Surely wind is good because winding coils for generators is less harmful than fabricating solar panels?)



They're scorning Google's efforts to use available renewables, today, in favour of a technology that simply doesn't exist, yet? Again, I ask, what should they do? Travel forwards in time?

After that, I skipped to the end and read the conclusion. Here's my own: Mr Rocke does not know what he is talking about, despite his claims to the contrary in his final paragraph. He is irrationally idealistic and wholly unrealistic if he thinks that today's cash can simply plonk Fusion reactors (after time travelling) down to supply the USA's energy requirements.

And, no, one doesn't get a free pass just because they're planning to *start* a PhD in 18 months.

The ethics and environmental impact of research in the field of deep learning and AI NEED to be challenged and questioned. That is vitally important. They need to be investigated and challenged responsibly and realistically, however, lest all critics of the field get tarred as raving, Luddite loonys and those with the power to make decisions cease to pay heed. Articles like this one do not achieve that end.

Everything will crash "some day", there are at least two more conditions to be met before shorting.

Ah, yes, infamous 2100 Bitcoin death spiral