Steering Blindfolded toward Intelligence

A motivating and optimistic account facing all the work ahead.

The idea to evolve minds is gaining momentum, and I would like to expose a challenge in this post, together with an attitude in moving forward. The challenge is what I have come to call the “steering blindfolded problem”. It stems from premises of the mind evolution argument, on where intelligence comes from, and how it has evolved into us (among other intelligence forms).

Several research endeavours state intelligence coincides with life. Morowitz, Reber and Perez independently walk us through the cosmic eons, with solid indicators intelligence and life are inseparable. Many others have drawn similar parallels, perhaps more focused on Physics or Biology, like Tegmark and Marcus.

Such intelligence-life position (ILP) is motivated in part by our inability to progress on the “real AI” question (now AGI, at some point strong AI, but ultimately what we really expect in an AI). A common method to build something is to split the initial problem into smaller problems, and continue splitting them in turn until we recognize pieces we already know how to solve. We refer to it as the “top-down path” (∇P) in the remainder. This approach has been serving us for millennia, on a daily basis. Yet we have after all still little to show on the intelligence problem.

Endeavours accumulate evidence Darwinian evolution plays a key role in creating our minds, but is not sufficient. In Empathy Emergence, Perez explores missing mechanisms, as standard evolution models come short at explaining several observations. I particularly like Perez questioning why Nature has come up with very similar neuron cells across wildly different organisms, including us, mollusks and insects.

Going back to the origin of life, mollusks and us do have a common ancestor. Going forward along evolutionary branches so far apart, how does Nature converge to neurons here and there? Coincidence or not, we lack an explanation on such observations. We simply do not know yet the way up.

A minority of deep neural networks are ΔP by nature. Art-related constructs can be creative explorations. Unrestricted “network architecture search” like AutoML Zero can converge to innovative network solutions, at times overlooked by our best human endeavours. Some also mention “generative-adversarial networks” (GAN) in this category (my candid justification is the opposing network pattern is the key here to be in ΔP, as a resonating trial-error loop). The potential of ΔP exists in Deep Learning, but there seems to be many low-hanging fruits from the ∇P mindset hiding it.

The good news is the popularity of Deep Learning provides us with a powerful tool to explore ΔP, if we care to look.

Thinking in ΔP about AI starts getting momentum. The key problem to solve is understanding the mechanisms to climb the intelligence-life ladder, hopefully up to what most recognize as “real AI” — (almost) us. Yet a set of ancillary problems may prove more difficult to deal with. One of them is the “steering blindfolded problem” (SBP).

By extension and more anthropomorphism, we see here too what the AI doom camp refers to by the “end of us” or “summoning the demon”. Steering blindfolded does not guarantee any result, and even less any desired result. And this a problem we need to address, not only by “just doing it”. We may believe following the Homo sapiens evolution groove, while we take a wrong turn along the way, ending up with unfathomable forms of intelligence, or worse (or better?).

So what to do against SBP ? The problem exists yet not a stumbling block. Keeping it in mind may turn to be the most important point. ΔP might not lead to AI, but if it does get us closer and we want guarantees like forms of ethics, then ∇P analysis tools are our strength to make sense, monitor and “control by circumscription” any ΔP design that comes about, and before “release”.

Quite an undramatic conclusion: We need to think both upward and downward, although there is much to do upward at this point. Purist may see cheating here.

Related posts:

The lighting that you choose for your room can either make your game or break it. a genuine choice of lighting could as per a trustworthy perspective whenever play considers for your room’s…

Everything communicates. As a traditional definition, we look at marketing as the actions of communicators to invoke an action or belief upon the consumer. A blend of science and art instills a form…

Some dark patterns and cycles of negative reactivity have taken hold of our relationship, like toxic black mold that blooms unseen in the walls of your house — and that will likely get worse over…