The differences between programming instinct and true intelligence

For Lorenz there is an insurmountable difference in the type of instinctive behaviour that we see in simple insects and the type of intelligent behaviour that human beings display in learning; he is arguing that intelligence does not and cannot emerge out of instinct by any form of evolutionary process.

No example of this is more extreme than the European (common) cuckoo. The common cuckoo parasitizes hundreds of species of bird, of which the Eurasian reed warbler is amongst the smallest. A reed warbler weighs around 13g, but feeds up a cuckoo to a weight of around 130g without any indication of realisation that the cuckoo is not its own offspring. From our perspective, the reed warbler is clearly vastly smaller than the cuckoo, with different patterning and morphology. But the reed warbler is driven by its misdirected instinct to feed whatever chick is in the nest. For all the benefits of its instinctive behaviour being able to be driven by a small brain, the reed warbler accepts a cost from nest parasites.

von Frisch was working at the opposite end of the spectrum, showing how simple instincts can combine to yield complex behaviour in honey bees... Most astonishingly, in the waggle dance, a worker bee can pass on information about a food source to other works by an intricate and enchanting display. von Frisch laid the way for subsequent research to manipulate a bee’s senses to trick a recruited workers into flying the wrong way. For all the complexity, the instincts that underlie the behaviour are still fixed and incapable of intuitive or intelligent use of sensory information in behaviour.

A classic (but not always helpful) distinction which can help to break through the various possible sources of behavioural software is hailing from either genes or environment as two ends of a broad spectrum. Instinct is toward one extreme of genetically-sourced behaviour, which is consistently and rigidly determined by fixed patterns. Human-like learning is closer to the opposite end of the spectrum, where behaviour is almost entirely determined by flexible rules that are learnt from environmental sources like experience, other organisms and stored information.

Both genetic and environmental sources could, in principle, be equivalently ‘intelligent’, but in general greater flexibility of behaviour has given rise to greater intelligence... Instead of learning a response per sentence, with flexible components or otherwise, it is staggeringly more efficient to learn to how the rules of language which can equip you to respond to any question in an intelligent manner. For the most part, this is what intelligence has come to mean — the ability to adaptively respond to a stimulus without prior experience.

The ability to respond to a stimulus does not equate intelligence. All living things respond to their environments, but only a select few are considered intelligent. The ability of a machine to change in response to the environment does imply it is anything like biological learning... Machine learning is competence without comprehension, or simply responsiveness not learning, unless machines are given sentience. Sentience enables learning without a predefined utility function.

No matter how many signal-response pathways are coded into a computer, the machine would not display human-like intelligence. Human intelligence is reached at the extreme of environmental influence on behaviour, where individuals learn behaviour from self-discovery through direct mimicry to abstract rules. To reach this extreme of learning, a machine would need enormous freedom of expression which no machine is currently afforded because it is not known how to do it.
https://becominghuman.ai/from-instinct-to-intelligence-has-ai-taken-a-wrong-turning-e84582dc4600