University of Birmingham > Talks@bham > IRLab Seminars: Robotics, Computer Vision & AI > Recently hatched ideas about hatching and intelligence, using very low energy physics and chemistry at "normal" temperatures in egg-laying vertebrates

Recently hatched ideas about hatching and intelligence, using very low energy physics and chemistry at "normal" temperatures in egg-laying vertebrates

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Martin Rudorfer.

Many researchers investigating intelligence ignore the deep observations of Immanuel Kant (around 240 years ago) about both ancient forms of mathematical intelligence and everyday spatial intelligence in humans and other animals, e.g. knowledge concerning what is impossible or necessarily the case. E.g. it is necessarily true that if S1 S2 and S3 are spatial regions, and S1 contains S2, and S2 contains S3, then (necessarily) S1 contains S3. Another example: our natural number system depends on the fact that the relation of one-to-one correspondence is necessarily symmetric and transitive: exceptions are impossible. Why?

Many current researchers believe (mistakenly) that such forms of intelligence result from learning processes in which neural networks collect statistical data and derive probabilities. Such neural learning theories are deeply misguided, in part because they cannot explain how ancient mathematicians made discoveries (including proofs of Pythagoras’ theorem several centuries before Pythagoras was born) demonstrating that something is necessarily true, or impossible. Such conclusions cannot be derived from statistical evidence: necessity and impossibility are not extreme degrees of probability.

Moreover, the ancient concept of number (cardinality of a collection) used in connection with counting processes, is based on the fact that the one-to-one correspondence relation is necessarily transitive and symmetric: exceptions are impossible, not merely improbable. The work of Piaget in 1952 showed that young children generally do not grasp that transitivity until aged about six years. What brain mechanisms can explain such non-statistical, non-probabilistic mathematical discoveries? Kant thought that these were “mysteries that will forever lie concealed in the depths of the human soul”. I suggest that they have chemical explanations.

As Kant noted, discoveries about necessity and impossibility cannot be explained by mechanisms that operate by collecting statistical data and deriving probabilities (using ratios found in the data) since necessity and impossibility are not extremes on a probability scale.

So currently highly fashionable neural nets cannot make discoveries in geometry and topology about impossibilities and necessary connections, some of which go back centuries before well known ancient mathematicians such as Pythagoras, Archimedes and Euclid. For example several proofs of Pythagoras’ theorem about right angled triangles had been found long before Pythagoras was born. How? The ancient mathematicians could not have used modern logic-based reasoning mechanisms that were not developed until the 19th and 20th centuries. There’s no evidence suggesting that spatially intelligent non-human animals, or human 6-year olds use sophisticated symbolic logic.

What sorts of mechanism can detect facts about spatial necessity and impossibility, and make use of them in selecting or controlling actions, for instance crows or weaver birds building safe-to-use nests in trees, or a female orangutan holding an infant in one arm, using the other to rearrange branches to form a safe structure to sleep in? Can any current robot do anything like that?

I’ll relate these questions to the task of explaining how eggs can produce surprisingly competent vertebrates!

Eggs of vertebrates, such as chickens, swans, turtles, crocodiles, alligators, and many more, start off with large amounts of a few kinds of relatively undifferentiated matter, and then in a remarkably short time (weeks not months) they (mostly) produce new hatchlings that not only have extremely complex species-specific physiological systems and structures—far more complex and intricately structured than any machine of comparable size produced by human engineers—but also have sophisticated spatial competences that do not need to be learnt, as they are ready for use very shortly after hatching, like the competences in the young avocets in this 35 second videoclip from BBC springwatch last June: (the whole Springwatch programme is on Youtube).

What are the alternatives to neural networks? In eggs, the only alternatives seem to be chemical information processing mechanisms (on which I think Alan Turing was working shortly before he died). I’ll present some clues about such mechanisms, but complete answers are a long way off.

Ontogeny cannot recapitulate phylogeny. But the processes of construction of mechanisms of ontogeny in eggs (i.e. construction of chemical assembly mechanisms in eggs) may correspond systematically to the evolutionary history of the species—phylogeny.

The talk will sketch a collection of ideas about evolution, development, chemistry-based reasoning and control processes that I think contradict widely accepted, but mistaken, theories about how neural networks can explain intelligence, including mathematical intelligence, including ideas that are still only a few weeks old—and in need of further development.

The new ideas are also relevant to non-egg-laying species, e.g. mammals, but their developmental mechanisms are more complex and more varied. The problems and solutions are less constrained when there isn’t an eggshell limiting resources and mechanisms relevant to the developing embryo.

There are still major gaps in these ideas, some of which may involve deep features of quantum physics that I don’t yet understand!

One of the questions arising is whether previously unnoticed facts about biological assembly processes in eggs also have previously unnoticed implications for fundamental features of physics, inclduing features that make possible extraordinarily complex assembly mechanisms that operate at animal body temperatures, apparently using relatively little energy?

No current human-designed processes for manufacturing complex physical machines are capable of being compressed to the size of an egg or capable of going through processes of repeated self-extension with no external influences apart from absorption of a small amount of thermal energy and small amounts of atmospheric gases. And no human-designed machine produced so far can construct a machine with the competences of a baby avocet, chicken, duckling, turtle, alligator, python, etc.

I hope the talk will provoke suggestions from researchers with much deeper knowlege of biochemistry and physical theory, including quantum physics, than I have.

Extended notes for the talk, including additional references, can be found here (still being revised):

This is a revised, extended version of a presentation to the Centre for Computational Biology on 23rd March 2022.

This talk is part of the IRLab Seminars: Robotics, Computer Vision & AI series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


Talks@bham, University of Birmingham. Contact Us | Help and Documentation | Privacy and Publicity.
talks@bham is based on from the University of Cambridge.