Unmaking Sense

Extended reflections on Lev Shestov‘s favourite quotation from Seneca, “decide once, obey forever” as it applies to Scripture, to science, to compressability, and to all the decisions that we make in life. And it could, were we foolish enough to relinquish or abandon our obligation to decide by using active discernment in the name of some all-knowing, all powerful AGI, as we have proved so eager to do in the name of some all-knowing, all-powerful deity, mean that we would then obey it forever without question, treating it as if it were incapable of error. The dangers of placing such faith in anything are incalculable.

Human beings tend to love being fans and disciples because their devotion to a person, team, party, religion or cause seems to exempt them from exercising judgement in deciding what is right and true. But active discernment is essential in all our decisions, and a key component of participative democracy. So were chatGPT to come to be thought infallible, it would signal terrible danger for us all, for then whoever controlled chatGPT would rule the world.

Learning from self-play plus questions that arise from where we are in our learning journey that ChatGPT or some lookalike answers at context-sensitive levels of difficulty will mean that every learning-experience is suited to the place on a personal educational trajectory defined by how each of us has made sense of the world. Education is about to be transformed.

Mixing self-okay with personal questions and learning-context-relative tailored chatbot answers by combining ChatGPT-style responses with the kinds of AI Duolingo has tried to implement in its BirdBrain learning analysis along lines used by DeepMind in AlphaZero promises a transformation of learning for all students at all ages, and it’s coming soon.

How do we experience our lives as a single self when our brains are performing different functions in different neurological regions? Does the brain create the illusion of the self to persuade us that we exist as identifiable individuals ready to fight to survive?

A report on a conversation with chatGPT’s parent program about its offspring: size; speed; number; time; and other statistics.

“Understanding”, “insight” and “intuition” are just names we give to processes we don’t understand but seem to exist. Once we thought them the products of a mysterious, metaphysical soul or self; now we should just regard them as exhibitions of our brains’ functioning and be more circumspect about denying that they can be replicated - perhaps using very different mechanism - in machines like chatGPT and other AGIs.

Learning to do exams by memorising answers to past exams is a technique of very limit effectiveness. In AGI this is called “overfitting” and the dangers of it can be reduced using regularisation techniques such as Early Stopping and Dropout, which we explain in the episode.

We explore some of the more interesting aspects of the thinking process and see that it isn’t really something that takes place in our conscious brains at all.

If the processes that go on in our neurological infrastructure are unintelligible, and I think there are a great many reasons to believe that they are, and if the same is true of what goes on in a neural net to beneath the surface of its output, then that again is another parallel between the way chatbots, or this particular chatbot chatGPT and human beings operate. That we cannot interrogate these inscrutable infrastructures to check up on them should make us more cautious about trusting their output m, just as we are cautious about trusting the output from human beings. But that chatGPT is no more considered to be infallible than you or me may in the end prove to be one of its most endearing characteristics, for if it were ever to be regarded as a universal oracle incapable of error, the world would be in much more danger than it is when we treat it as something that we have to take and scrutinise and challenge and question and analyse before we act too trustingly upon what it tells us. In that respect it mirrors human interactions.