How mind bootstraps itself

Notes from the book "From Bacteria To Bach and Back" by Daniel Dennett

Daniel Dennett, who recently passed away, is an inspiration for people like me who love grand ideas that connect many different fields at a deep level (and be rigorous about that; well as rigorous as grand theories can be). His thinking is difficult to categorize: part philosopher, part scientist, he had many good ideas on topics ranging from evolution to consciousness to God.

The book From Bacteria To Bach And Back had been lying on my shelf for many years and I recently picked it up because I’ve been thinking a lot about evolutionary constraints on brain and consciousness.

The book is hard to summarize because it spans so many topics, but perhaps the central idea is this: competence without comprehension is not only possible, but is the norm; but when comprehension slowly develops, it leads to an explosion of artifacts and culture that we see around us. Dennett argues that there’s a difference between performing a behavior and having a (manipulable) representation in the mind of that behavior such that you can “think” and “reflect” on that behavior to fix a particular context or to make it better.

For example, a bird making an intricate nest displays a startling case of competence but is debatable whether she understands what it’s doing. The bird nest works well because it is an adaptation to the environment the bird finds itself in, so there’s a rationale behind design of the bird nest but nobody designed it. This is what Dennet calls as a design without a designer where rationale for the designed objects are free-floating. So reasons exists but not in someone’s minds.

Humans, on the other hand, seem to understand what they’re doing as they can flexibly change their behavior depending on context. Our entire society and technology is a proof of us transcending beyond our evolved instincts.

How did that happen?

Dennett argues that this transition was gradual and links it to languages (and memes, in general). Our ancestors, like birds, would have instincts for performing various behaviours without understanding them. Imagine if one such biased behavior is copying the elders or copying the most successful. This kind of copying behavior could initially bootstrap rudimentary form of language as a group settles on some sort of shared signals. It also bootstraps knowledge-sharing (aka culture) as copying allows spread of successful discoveries much faster than what genetic reproduction cycles would allow.

Once rudimentary culture and language is in place, the benefits of having these capacities would allow for co-evolutionary process where the brain and biology of humans could change in order to absorb language and culture faster. However, this is still competence without comprehension. One can imagine sophisticated behavior in a primitive group of homnids but nobody knowing what they’re doing. Even proto-language (like most animal calls) could have been “mindless” in the sense of it eliciting the right behavioral responses, but individuals not comprehending what they’re doing.

So how did competence then led to comprehension? The road to comprehension wasn’t a step change, though. There’s a continuum of instinct to flexible behavior (which is a hallmark of comprehension). You could have behaviors that work well but you only sort-of understand why (for example, the impulse to do art or driving a car well).

Dennett suggests that gradually comprehension could have become better and better due to the need to justify our actions to others (or deceive others) in a social settings. Once a shared proto-language has taken hold in a group, everyone has an incentive to take advantage over others by giving false information. This creates a pressure to justify and convince through giving reasons and that required having a representation of reasons in the mind. Such representations of “why am I doing this” were glimmers of comprehension which became stronger thanks to the arms-race between deceivers and questioners.

This dynamic also explains where a sense of self comes from. In order to decieve others, you need to model others and their behavior in the situation at hand. And that model includes how you yourself will behave in response to their responses. As you can see here, self emerges within the theory of mind because social dynamics require identification of distinct selves including oneself. (This does make me wonder if solitary animals have an explicit sense of self that they feel as strongly as we feel.)

The last part of the book is about consciousness. Dennett is an illusionist, which means he believes that there’s no hard problem of consciousness to be solved. He says that our experienced reality is like a user-interface constructed by evolution on top of the raw reality. The user interface exists because it is beneficial to the organism’s survival but we must not mistake it for “truth”, which only a scientific investigation can reveal (and not introspection).

Over time, I’ve grown sympathetic to the illusionist position but haven’t yet come around to embracing it fully. My current feeling is that the hard problem will not be solved but will gradually dissolve as neuroscience progresses (just like the problem of “what is life” got dissolved as we made advances in biology).

There’s definitely an air of mystery around consciousness, but is it because evolution hides complexity of brain processes that compose it from us (which science will eventually reveal) or is it because consciousness is in some way fundamental in the universe?

Dennet argues it is the former and he calls this as the Cartesian Gravity. Our inner experience is so vivid that we mistake it for truth but we find it mysterious because evolution had no incentive to represent how it’s built. The more we probe internally, the more we find it as a given, which seems unexplainable and baffling. This also prevents us from being scientifically objective about consciousness as we keep gravitating to first-person experience (which could have holes that we’d never see unless we study it from a third person point of view). This is why Dennet advances his approach of studying consciousness, heterophenomenology, which treats first-person reports of consciousness as a third-person scientific object of analysis.

---

I have a lot of more notes from the book (in partly random order). Reproducing them below, in case you find them interesting.

On comprehension

• On the topic of comprehension, two open questions (in my mind):

  • What does comprehending something mean?

    • It’s representing a topic such that it can be explored or manipulated

  • Is precise understanding of anything even possible? What does that mean?

• Learning and plasticity emerges in organisms if environment keeps throwing novel situations at them, otherwise evolution hardwires behavior in them

  • When evolution can’t predict, it gives organisms a general problem solving ability

• How do we test for comprehension? Think about what comprehension enables:

  • Context specific modulation of behaviors

  • World model of the situation where you can do a variety of actions (in imagination)

  • “Sophisticated” behaviours during repeated interactions / observations, grounded in success or increasing odds of success

    • Sophistication → success of goal achievement under varying conditions

  • Ultimately, comprehension should lead to a behavior which is deemed successful (for the agent).

• Comprehension / understanding / sophisticated behavior → matter of degree, not binary. Agents can be categorized into four levels

  • Darwinians. Pre-baked behaviors in genetics.

  • Skinnerians. Try random behaviour (in the world) and reinforce successful ones

  • Popperians. World modelers. Simulate trajectories in the head and perform ones that seem successful

  • Gregorians. Agents with Thinking Tools. Reflecting tinkers of thinking tools. Improve thinking process itself (including world modeling)

• Semantic information is a bit of information that’s expected or could be useful in the future

• Value of information is relative

  • The more someone knows about a domain, the less they have to remember to transmit same amount of knowledge / information.

• Tools compound if representation of tools are there

• The social constraint for humans had a lot of question answering going around (for justification and convincing); and this was one step away from asking questions yourself

• Internal monologue helps by simulating a conversation which enables brain access information from parts of brain that wasn’t available before as reasoning requires integration of information

• Understanding is representing a topic such that it can be explored or manipulated

On evolution

• Abstracted / Generalized Darwinian principle: whatever persists is expected to have elements of persistence.

• Blind evolution v/s intelligent design: not a binary but a continumm

On memes

• Memes are ways of doing something that can be copied

  • Words are memes that can be pronounced

• Memes may have no author, like a baseball cap worn backwards, they may start as a mistake but because they can be recognized by others who may copy it (and tweak it slightly randomly or by mistake)

  • This is why ideas can be copied without knowing why they work

  • Memes are like a cold virus; we can catch them from one another unconsciously and they can take hold in us to spread further

• What gets adopted doesn’t have to be beneficial to the host.

  • Viruses can be beneficial, neutral or even harmful. (And so can the memes)

• Memes speed up evolution as they allow horizontal transmission of behaviors (good or bad) with good providing benefits and hence helping genetic machinery co-evolve

• Some memes become thinking tools that enable more memes by virtue of intelligent and comprehending design

• LLMs may be modern examples of competence without comprehension — like evolution, gradient descent slowly fashions adjustable parts until the system fits the environment to a great degree

On language and culture

• Symbols / digital transmission helps remove noise as you snap signals to their respective categories

• Is language necessary for cultural transmission — like tending a fire?

• Once rudimentary copying of dispositions (like copying elders) is in place, memes can have a fertile ground to evolve

• How could our pre-language ancestors could have bootstrapped culture? Do a thought experiment on what would it take for parents to teach their kids without language:

  • Joint attention

  • Copying

  • Conttext understanding

• Words could have been instinctual inflections at first but later evolved for accuracy

  • Our ancestors could have started using words without realizing what they were doing, like an antelope stotting around — competence without comprehension

• Language led to culture and that led to thinking tools installed in our brain, which led to further culture

• Language evolved to fit the brain (and vice versa).

  • The better some words could get pronounced, the faster they replicated.

• Words are like DNA but for cultural evolution

• Words are affordances that our brains are designed to pick up.

On consciousness

• There must be something like to be a human since so many of us talk about it in books, poems, journals, etc. — whether it is an illusion or not, it’s an objective fact

  • (anyway, what is an illusion and how would you differentiate it from a non-illusion)

• What would a Martian scientist studying life on Earth say about consciousness

• Feelings like cuteness are (Bayesian) expectations of what we could do next, projected out there in the world

  • This is not merely valence projected, but our behavioral tendencies / expectations / affordances

• Visual illusions are an example of expectations during subjective experience

  • Reality (how it is) differs here most clearly from experienced reality

• Just like sweetness / red doesn’t exist out there, there’s no internal object that gives off those properties — there’s no homunculus observing these properties, there can’t be because of infinite regress

• What does qualia do that cannot be done without qualia

  • (This is a question I find myself asking again and again because if the answer is not-much or nothing at all, then we could subsume qualia into a functionalist account of mind and the question would shift to the meta problem of consciousness which asks why do we report there’s a mystery to be explained).

• Problem with introspection is that this channel was not designed for scientific investigation but for survival

  • Evolution has given us a gift that sacrifices literal truth for utility