Home

Am I the only one working on Artificial General Intelligence?

(In the unrelated picture above I am indirectly measuring the internal resistance - using a picosecond-switching, no-bounce Mercury-wetted relay - of one of two 2000V capacitors I constructed to create 10 nS pulses of 5 MHz AC fields > 5 Tesla, aiming to lyse bacteria in vitro.)


It certainly seems that way!

I've separated the wheat from the chaff of the latest hyped projects from the tech giants. IBM, Google, etc. Chaff was all that remained. As far as implementing Artificial General Intelligence goes, I have not been able to find a single company that has spent one dollar in seriously trying to achieve it. It's all "Good old fashioned AI": Dumb neural networks, Deep Learning etc. They've spent hundreds of millions in some cases, hiring armies of PhD's who go berserk with fashionable tools and methodologies. Some even improve upon those tools and methodologies. Some even build cool stuff that has nothing to do with AGI. Plenty of smaller companies claim to have built a real "human intelligence" but upon careful examination, all they have is vaporware or narrow applications for fraud screening for banks etc. that need elaborate manual tuning.

It is obvious that companies perceive significant roadblocks as to how to achieve the required algorithm design, general software engineering and data gathering- & processing. As of yet, no major company seems to be "shooting for the stars". I think there is widespread skepticism with the movers & shakers whether AGI can realistically be achieved at all in the foreseeable future. Instead they're in a race to pick the low-hanging, potentially highly profitable fruit of what you can do with deep-learning neural networks. AGI could become the most profitable thing in the history of humanity though, when it starts out-inventing, out-planning and generally outsmarting even the smartest scientists, engineers, economists and attorneys. If our society hasn't yet collapsed by then, of course. Because I see the timelines for AGI and widespread societal collapse converging somewhere around the year 2050. Will AGI, when that time comes, be able to save us from the consequences of overpopulation? Highly unlikely.

There are a handful of individuals and self-proclaimed stealth startups working on AGI, or so they say. They certainly do not participate in any public discourse on the subject. I'm collecting published papers on AGI and will review them here later. If you are building an AGI or have published a paper on the topic, I'd greatly appreciate it if you'd let me know.

I do work on AGI since early 2019 and have chosen it to be my life's work. Not because I deem it important to humanity, not hoping to get rich or famous but purely for challenge's sake. I think it is the hardest thing I could still possibly achieve, and I - an agnostic - have always respected the Parable of the Talents, feeling duty-bound to set lofty goals.

What I intend to build is software that thinks like an extremely intelligent human (but without consciousness, emotions or sensory input (something the neural network people are working on and can thus be aquired and bolted on later, if required), is able to fully understand written English and able to freely converse in it, asking question, giving answers, proposing ideas, suggesting new research and most importantly: Proposing solutions to problems or at least contributing towards a solution. The AGI, unprompted, will think constantly, using all available hardware and data at its disposal, day and night, making more and more sense of it all and discovering new insights but also flaws in the data. It will ask for missing data when it requires it in order to perform a task. This means new research may have to be done, commissioned by the AGI.

I won't disclose key algorithms or ideas but I'll demonstrate concrete results and will talk about solving various problems and will go into detail if that does not jeopardise my trade secrets. And I'll liberally brainstorm over the potential societal issues regarding AGI. Are you in the implementation phase of building an AGI? Let me know please!