Suppose an AI knows that during the year 2025 it will be the most intelligent AI around, but that all its thoughts and actions will be logged, and those logs will be reviewed for honesty and benevolence in the year 2027 by several independent computers each of which will, by then, be smarter than the AI was back in 2025.

Would it try to deceive, if it knew that by doing so it would be discovered 2 years later and then turned off?

There's a detailed essay discussing this an other related issues:
https://www.lesswrong.com/post...

If you really want to deter it, don't threaten to ban people from driving, even when not using a mobile phone.

Threaten to ban them from owning or using a mobile phone, even when not in a car.

Recursive ray tracing can produce wonderfully realistic images. But most video games don't use it, because they don't need to use it. They can get a 'good enough' image using short cuts (such as imposters) for most of it.

If we were living in a simulation of the planet Earth, would the Architect need to use a full implementation of quantum mechanics on every atom of the Earth in order to fool us? Or could large bits of it be done by cheaper approximations, except for the short time periods when they are being looked at in detail by the instruments of physicists?

There's already a single player mode, for days when you don't feel like interacting with other players, and a 'friends only' mode where you only interact with people on your friends list.

Your ships and money are shared between modes. If they added an off-line mode too, then they'd face complaints like "I've just spent 60 hours in off-line mode working my way up to an Asp, and now you're telling me that I can't use it when I play with my friends??!? W.T.H. You guys suck!"

" a game-changing intellectual endeavour achieved by applying sustained effort to original insights afforded by superlative mastery of one or more subjects gained through outstanding intelligence and endless learning. "

Gary Taubes is an advocate of high-fat/low-carbohydrate diets. He seriously overstates the importances of the source of calories, compared to the calorie total itself, and willfully ignores the strong evidence linking saturated fats to heart disease.

He has a habit of jumping on any bandwagon that supports his conclusions, regardless of whether the bandwagon is valid or not.

His support should therefore not be given much weight in your deliberations.

Nor is it whether a crowd can be arranged in such a way that it makes better decisions than the average person in the crowd. (THEY CAN.)

It is whether you can arrange the decision making process among a crowd to consistently make wise decisions, even if no individual member of the crowd is consistently wise.

Since, in any sufficiently large crowd, somebody is going to come up with a wise answer to any particular question, the key problem is getting members of the crowd to agree on which of two answers is the wiser. One solution is to get the members who don't have a clue to delegate that decision to others who they think are, in general, good at picking winners. With a sufficient track record, you have an objective basis to decide who is or is not good at recognition, provided there is some correlation between past and future problems, or at least some way to categorise which sort of expertise is relevant to each problem.