> Given the latest developments, I would put the odds at fifty-fifty within the next decade.

thing is technological development is not linear, you can't predict future development based on the last n decades. You can't assume an AI winter is not coming because it most likely is

> because it most likely is

Is this not you making a prediction, the very sentence after saying one can not predict the future?

Saying an AI winter is coming isn't saying how long it will be for or what the magnitude will be.

OK, so what is it saying?

That progress won't be steady and predictable.

No. This is like saying that if today is not raining, a rainy day is coming

> because it most likely is

"Ark estimates that Deep Learning has created $1 trillion in market value so far. " -- we might be headed towards a warm winter

(from https://www.nextbigfuture.com/2020/01/ark-invests-big-five-t...)

Value is there until it isn't. A lot of the value here comes from its potential rather than its current use

I neither claimed tech development was linear, nor am I basing future AI progress on the rate of progress in the last decades.

I am basing future AI progress on what AI can do in July of 2020, which I believe already represents substantial progress towards that goal.

How would you measure human level AI?

Since intelligence isn't a well-defined term, the default fall-back would be a measurement by proxy.

Humans are usually tasked with solving an IQ-test as the most common proxy.

A human level AI could do just the same. If the AI scores above 50 points, it can be considered human level (though literally retarded).

The measure of intelligence, François Chollet

https://arxiv.org/abs/1911.01547

Thanks for the interesting read! It was a bit tiresome (a 38 page history lesson, followed by a 5 page introduction and only about 5 pages of actual content), but very informative nonetheless.

The proposed method is very hard to distinguish from a traditional IQ test (or a subset thereof) and the first to parts (~38 pages) basically serve as a justification for this.

In the end the author admits that the proposed method lacks test diversity, has no established validity and has no way of qualifying results.

The essay just stops where it gets interesting (i.e. at the point where the actual science starts), which left me a bit disappointed in the end t.b.h.

It's a great summary of the history and development of the field and the methodologies used therein.

It's in no way a solution to measuring and quantifying general intelligence, though.

I think the only measure that makes sense is https://en.wikipedia.org/wiki/I_know_it_when_I_see_it

ironically being able to do that is a key aspect of intelligence .