So ... can a computer think?

Comments on an ongoing philosophical controversy

Recently, Google engineer Blake Lemoine made media headlines when he claimed that a Google chat program called LaMDA was sentient – it could not only think but had gained some level of consciousness. For a very good overview of the situation and a critical look at the claims, I’d recommend reading this on The Convivial Society Substack: LaMDA, Lemoine, and the Allures of Digital Re-enchantment

More interesting, to me anyway, is the bigger discussion behind the headlines: it is possible for machines to think or to be conscious? I’ve recently found two (long) articles that predate this recent controversy and take interesting and critical positions at odds with the normal views you hear.

The most recently published is Reality Minus, by David Bentley Hart, and is an explicit reaction to the recent book Reality+ from David Chalmers. It is over 7000 words and Hart’s writing style can be dense, but it is worth the effort if you can manage it. The second is Can Machines have Common Sense?, by William Hassenberger. It is a much easier read and slightly shorter so I’d read this one if you only want to read one long article.

However, to save you the effort of reading 13,000 words of (at times dense) prose, I’ve picked out what I see as the core, and most interesting, features of their articles. Some reasons for my interest in these arguments that contrast machine to human approaches to thought and knowledge can be found in my previous posts expressing skepticism about algorithmic approaches to the world and knowledge.

Reality Minus

David Bentley Hart offers an interesting and direct critique of Chalmer’s arguments. Of note, however, is that his critique applies equally to a lot of other people, including Lemoine, all those who believe that conscious thinking machines are inevitable and the transhumanists who think that we will be able to ‘upload’ our conscious to a machine one day. If you aren’t aware, this is a common view in Silicon Valley and similar circles today.

Chalmers shares these views and, as a philosopher, is careful about his arguments. He explicitly bases it on a position he calls ‘structuralism’, which is common across many modern thinkers about intelligence and consciousness. To quote Hart’s summary of structuralism:

consciousness is the result not at all of the physical substance or substrate of the brain, but entirely of its structure of relations and functions.

Put differently, consciousness (and thought) do not depend on the type of stuff doing the thinking, but only on the way that stuff is structured and operates. This is a familiar model when we talk about computers, as software does the ‘thinking’ and can run on a wide range of hardwares. Or as Hart puts it:

the curious dualism that so easily separates mentality into, on the one hand, a kind of functional software and, on the other, the purely structural hardware or “platform” where it is realized.

One of Hart’s big criticisms is that this position is taken to be axiomatic - so isn’t justified - and he argues that it begs the question. If this structural conception is correct, then it is likely possible (eventually) for machines to consciously think. However:

As mind has never been discovered anywhere except in organisms, where it appears to be associated with brains and nervous systems and nerve tissues and organic cells, what precisely justifies the belief that mental activity resides entirely in purely electrical activity and in the relations of the circuitry that permit it?

Or further:

Would it not make more sense to assume that the brain’s capacity for mentality has some causal connection to the cells and tissues and enzymes, synapses and axons and myelin, composing it, as well as to its uniquely organic and ontogenic history of continuous growth, development, catalysis, regeneration, and neural routing, as well as to its complex relations with the complete neurology, biology, organs, and history of the body it belongs to?

To rephrase, given that we have only ever encountered thinking and consciousness in biological organisms, it would make empirical sense to assume that there is something intrinsically biological about it. At least, we should want strong evidence that it is possible to recreate it with computing power on machines.

While there is a lot in the article, Hart’s core philosophical criticism of Chalmers and others is that they have recreated a strong dualism between mind and body. Unlike a Cartesian dualism between soul and body, this is between the software that thinks and is conscious; and the hardware on which it runs.

Hart argues that this is an error of modern thought and philosophy. Before Descartes:

Very few would have thought it sensible to ask whether it was the soul or the organism to which mental acts belonged, not because the prevailing paradigm of human life was dualistic, but because it was not.

Even for Platonists, who are commonly assumed to have been dualists:

the body was not a machine merely animated by an extrinsic agency; it was itself already a reflection of an eternal form naturally disposed toward and instantiated by life.

In short, Hart argues that the modern approach to consciousness, and the transhumanist movement, depends on a reinstated dualism between mind and body. Hart is confident that this dualism is flawed philosphically and doesn’t exist. For that reason, he concludes that none of the confident predictions about thinking, conscious machines will ever come true.

Can Machines have Common Sense?

William Hassenberger’s article is, in part, a review of The Myth of Artificial Intelligence: Why Computers Can’t Think the Way We Do by Erik J Larson. While he agrees with Larson’s arguments, he doesn’t think they go far enough and expands on them. In short, Hassenberger advances a number of reasons why computers will never be able to think humanly or, to put it in the terms he tends to use, will never be truly intelligent.

The first reason is one that Larson strongly prosecutes in his book. Traditionally, logical reasoning was divided into two types: deductive and inductive. Deductive reasoning works from premises to conclusions while inductive reasoning makes generalisations and statistical inferences. Larson draws on CS Peirce’s third type of reasoning: abductive.

Abductive reasoning is where we come up with an explanation that makes the best sense of our evidence, and take that as good reasons for believing it. A detective story is typically a good example. The successful sleuth can see the same evidence as everyone else, but is able to make a creative leap to find an explanation that makes all the pieces fit together.

Abduction necessarily involves a creative leap and is taken by proponents to be very common:

Larson follows Peirce in claiming that it actually goes on constantly in human life. It is involved in identifying interesting problems and offering possible solutions; in understanding language, with all its ambiguities of meaning that depend on context; in understanding a complex story; and in being able to fruitfully converse with other human beings.

Perhaps more surprisingly:

Abduction, claim Peirce and Larson, is even involved at the core of ordinary visual perception: seeing an azalea as an azalea is really an abductive guess from the raw or uninterpreted sense data.

The key point is that we can, and do, code computers and machines to reason deductively and inductively. However:

abduction goes beyond regularity, and we simply do not know how to codify it into a formal set of instructions or a statistical model that can run on a machine.

If this is the case, then current machines simply cannot think in a manner analogous to humans, however closely they may mimic some features of it.

Hassenberger buys this argument but thinks it is an insufficient explanation for why machines will never be intelligent, at least in the way humans are. To start with:

Making sense of the real world, as we humans do, requires a flexible and mysterious capacity to grasp meaning and interpret what is relevant in unfolding situations, a kind of know-how that researchers gesture at with the vague umbrella term “common sense.”

To motivate it, he runs through an extended example involving his coffee mug. While a computer program can classify it as a “coffee cup”, as humans we could articulate a very long (practically infinite) list of facts about a coffee mug that we can come up with and apply in context.¹ He gives an example:

A draft of wind blows through my open window — and suddenly none of the facts I considered earlier about my mug is relevant, only the fact that it can serve as a paperweight (whereas the crumbs on the table cannot, and I don’t even consider them). Common sense is the ability to navigate this terrain, and spot what makes sense and what matters in real time, often without even noticing.

We can imagine many scenarios where we would find a coffee mug highly useful for something other than it’s intended purpose. In those scenarios we would often instinctively reach for the coffee mug without analysing options or properties. Hassenberger’s point is that humans can do this instinctively with everything they come across and have some ideas about. Computers and programs need things to be defined explicitly (either in coding or as part of the training in machine learning). These two ways of processing information are fundamentally different.

Another line of argument he develops is that human thinking, understanding and even conversation is heavily dependent on a range of “concepts, rich in emotions and values”. These concepts, like love, trust, betrayal, anxiety and anger, “combine thought, feeling, and value in particularly human-centered ways.” These rich concepts are ones that we pick up as little children but can only really be understood by being felt in some way. There is more to them than information about them. However, information about them is all that machines can compute.

To put it differently:

In our mental life we experience the world as mattering — even in the negative or depressive instance of seeing the world as meaningless, as mere “sound and fury.” An algorithm doesn’t even “give a damn” enough to experience things as empty and vain.

To draw these threads together, he argues that a machine can never have human intelligence. Human intelligence is, for Hassenberger, fluid, contextual, abductive and heavy on meaning. Artificial intelligence is precise, rule bound, narrow and formal.

Summary

This has been a very truncated summary of the two articles. Nevertheless, it is clear that both offer coherent arguments that machines are essentially something very different to humans, to the extent that it may be a category error to ascribe human concepts of consciousness and intelligence to them. Nevertheless, we often do and are increasingly relying on artificial intelligence for a wide range of applications.

One strong concern, shared by both authors, is summed up neatly in the opening Substack article on Lemoine. Apparently, the chat program only responded in a conscious like way to particular ways of prompting it. When people other than Lemoine tried, they didn’t get anything that seemed remotely conscious. As argued in the article:

It illustrated how our machines often work only to the degree that we learn to conform to their patterns. Their magic depends upon the willing suspension of full humanity.

Social media dynamics sometimes suggest that we are adapting our behaviours to the algorithms. And professional behaviours are increasingly governed by the rhythms set by Microsoft Office. I guess this raises the question: are computers becoming more like us or we becoming more like computers?

1. Hassenberger starts an illustrative list:

it has a hard surface, that it will break if I drop it from my window (but not onto the carpet), that it will hold enough liquid to satisfy my craving for hot coffee (but not so much that the coffee will be cold after I’ve had enough), that it serves just as well for drinking tea or hot cocoa (but not champagne), that it weighs less than the planet Saturn (but more than a speck of dust), that it will hold a handful of AAA batteries, that spilling it puts my laptop at risk, that it would not taste good if I tried to bite it, that my wife would not appreciate my giving it to her as a gift on her birthday, that it cannot be used for playing baseball, that it cannot be sad or frustrated, that it was not involved in the French Revolution, and on and on forever.