Human extinction outlook

[John Clark]

I ran across an interesting graph in Tuesday's issue of the journal Nature, one AI researcher is quoted as saying:  

"AI systems are doing things that seemed impossible a decade ago. Anybody that thinks that there’s a wall or a plateau or something, show it to me in the graphs, because it’s just not there” 

On the other hand in light of recent developments the authors of AI 2027 have delayed their estimate of when the Singularity will occur, they had said it would occur in late 2027, but progress has been slower than they expected so now they think it will occur between mid-2028 and mid-2030.

John K Clark    See what's on my list at  Extropolis

2,<

[John Clark]

On Thu, Apr 23, 2026 at 7:38 AM John Clark <johnk...@gmail.com> wrote:

I ran across an interesting graph and quotation in Tuesday's issue of the journal Nature, one AI researcher is quoted as saying:  

"AI systems are doing things that seemed impossible a decade ago. Anybody that thinks that there’s a wall or a plateau or something, show it to me in the graphs, because it’s just not there” 

[Russell Standish]

2-4 years for AI self-improvement to become mainstream (rather than
the few limited experiments to date that indicate it might be
possible), for SGI to emerge and for AI to take over the means of
production (ie companies like Intel and TSMC run by AIs, not
humans). That's a tall order, even assuming here are no impediments
along the way. And solving the energy crunch is just one such
impediment to overcome - these AIs are currently very expensive to
run.

I'm still betting the singularity is at least a decade, if not two
away. Kurzweil's original prediction is still 19 years away.

Maybe there's some conflation between the existence of AGI (which I do
think exists now) and the Singularity (AI self-improving on a
hyperbolic growth path towards a an infinity)?

>
>  John K Clark    See what's on my list at  Extropolis
>
>
> 2,< 
>

> --
> You received this message because you are subscribed to the Google Groups
> "Everything List" group.
> To unsubscribe from this group and stop receiving emails from it, send an email
> to everything-li...@googlegroups.com.
> To view this discussion visit https://groups.google.com/d/msgid/everything-list
> /CAJPayv0icsSSajCPVt3QQKa6qmtwp%3DL3cPMbCUSQ48QxaMO5MA%40mail.gmail.com.

--

----------------------------------------------------------------------------
Dr Russell Standish Phone 0425 253119 (mobile)
Principal, High Performance Coders hpc...@hpcoders.com.au
http://www.hpcoders.com.au
----------------------------------------------------------------------------

[John Clark]

On Fri, Apr 24, 2026 at 6:43 PM Russell Standish <li...@hpcoders.com.au> wrote:

> Me: in light of recent developments the authors of AI 2027 have delayed their estimate of when the Singularity will occur, they had said it would occur in late 2027, but progress has been slower than they expected so now they think it will occur between mid-2028 and mid-2030.

I should've added that the AI 2027 people made their observation of slower than expected progress before Anthropic decided not to release its newest AI model called "Mythos" to the general public because it was so smart it would create massive security risks; and before Open AI released GPT 5.5 which can produce answers that are just as smart or smarter than the previous model while needing considerably less computational power and thus less energy. 

> I'm still betting the singularity is at least a decade, if not two away.

Maybe, but it's difficult to make predictions when a specific AI milestone will be passed because iterated processes in positive feedback loops tend to grow exponentially or super exponentially. And we have already reached the point where AIs are building their successors. 

 

> Kurzweil's original prediction is still 19 years away.

Yes but in 1999 Kurzweil predicted that a computer would not pass the Turing Test until 2029, but that milestone was clearly passed by 2023, and an argument could be made that it was passed even earlier. Today the only giveaway that you're talking with a machine and not a human is that it's too smart, too fast, too knowledgeable, and too polite to be human. 

 John K Clark    See what's on my list at  Extropolis

6;s