Smart (part 1) - how did things get to be smart?

I seriously do not enjoy reading the endless articles on Artificial Intelligence (AI) that turn up in my social media. Not because they are too technical, nor for fear that AI might revolutionise our lives in unpredictable and scary ways.

It is because the people who are driving the AI race¹ are really, really sloppy with the way they use important words. 

I'm starting to suspect that some of the fantastical claims about the future of AI are based on a pile of rubbish word use. It's a massive pile. It irritates me so much, I felt compelled to explore the many words and concepts used in writing about AI.

Why do AI articles now refer to all manner of devices and machines as ‘smart’? Does it really mean the same as when I say a friend is ‘smart’? What does it mean to claim machines will soon ‘surpass’ human intelligence? And why do some of these articles suggest that if a machine can be smart like a human, is can also be ‘conscious’?

To explore these questions I will dig into what ‘smart’ means, what human intelligence is and what artificial intelligence is supposed to be. In a second post, I will also explore how intelligence relates to language, and how a device supposedly demonstrates to us that it is intelligent. And finally, I will ponder why so many claims about artificial intelligence rest on considerable human ignorance.² 

Follow me on a wordly exploration of the use - or more accurately - the misuse of words about AI!

Definition of smart for humans

Smart is a familiar word, but let’s check in with the formal definition anyway. The dictionary defines smart as: 

♦️ having or showing a high degree of mental ability: intelligent, bright. 

Interestingly, the origin of the word links a person’s ability with words and language to the meaning of smart as intelligent. (This comes up again later.)

Etymology Online says smart is used: 

♦️ in reference to persons, "quick, active, intelligent, clever," 1620s, perhaps from the notion of "cutting" wit, words, etc., or else "keen in bargaining." Of speech or words, "harsh, injurious, unpleasant," c. 1300; thus "pert, impudent; on the impertinent side of witty" (by 1630s).

So, when a human is described as smart it means ‘intelligent’.³ 

In fact smart used to be a word we reserved solely for humans: only humans could be smart or intelligent. It was thought our intelligence made us special, put us at the 'top of creation'! For some centuries, we have used a bit of circular logic to say that intelligence is what made us human, by insisting that other animals (or other life forms) didn’t have it. Only recently, we have had to face the fact that they do indeed (viz the octopus, the crow, etc.) Oops!

This issue comes up again later, but for now, let’s leave non-human animal intelligence in the too-hard basket now while we focus on the new breed of machines that are supposedly ‘smart’ like humans.

A brief history of ‘things’ getting smart

Back in the 1970s, we heard about the ‘smart bomb’. It is the first ‘smart’ device I can find. Smart in this early use had the sense of ‘behaving as though guided by intelligence’; this guiding was provided by a human, somewhere down the line.

Now, more commonly, when we hear about devices being described as smart, it is things like phones, TVs and appliances.

When the first ‘smart’ phone was publicly released in 1994, it wasn’t called a smartphone. It was called IBM Simon Personal Communicator, and Simon had a touchscreen, faxing, emailing, calculator, notepad, and a dynamic contacts list. Over the next few years, other companies released their competitor models of phone, with various enhancements (e.g. a camera).

In 1997, the generic term ‘smartphone’ entered the language, courtesy of the marketing department of the Swedish company, Ericsson, when it launched its newest phone.

Smart meant phones could do more than what the ‘dumb’ phones could do, which is just make calls.

Then in 2007, Apple released the iPhone with a touch screen, GPS, camera, iPod, various apps - and most importantly - internet access. Android phones with internet access were released the following year, and suddenly smartphones were everywhere.

And with them, smart developed the new meaning of connected to the internet.⁴

The consumer appetite for the smartphone influenced a transformation of the television. The first Smart TVs were released in 2008 providing a combination of television, computer with internet access and digital media player. The word smart was selected as the best option (over others) by the Samsung marketing people.

And we all obligingly upgraded from our ‘dumb’ TVs.

After that, it’s just been ‘smart’ things everywhere - the Internet of Things. Smart doorbells, smart washing machines, smart fridges, all communicating with your smart television to provide you with various ‘benefits’ (e.g. sending you ads for what you’ve run out of in your fridge). These smart things have a built-in microprocessor for automatic operation, processing of data, greater versatility, and of course for communicating often via the internet.

So, the word smart came to refer to the integration of computing and telecommunication technology with other technology, all connected to the internet (or Bluetooth or other).

While their capacities are incredible, these devices are not actually ‘smart’ in the sense we use that word for people. It’s a marketing term used to appeal to consumers: ‘smart’ devices and appliances are marketed as convenient, individualisable, connected and flexible. No one was making claims to intelligence.

A leap in the meaning of smart

If you’ve tried to buy furniture, coffee, or insurance, etc., online over the last 5 to 10 years, you’ve possibly interacted with a not-very-smart-at-all chatbot. These chatbots answer basic customer questions or collect basic data. For example, the large fast food chains use chatbots to take food and beverage orders from customers online. These bots are trained to perform a narrow set of core tasks.

I doubt anyone ever described this type of chatbot as ‘smart’! In fact, the most frequent description of them that I've heard is ‘dumb’ and ‘annoying’ (and not just my personal experience!).

That was, until the release in November 2022 of ChatGPT by Open AI, quickly followed by several model upgrades, and products from many other companies looking for market share.⁵

ChatGPT (and its ilk) was triumphantly announced as a ‘smart chatbot’.

Ye olde marketing againe? But nae!

Now, when these machines are called smart it means ‘intelligent’, like it does when we use it to describe people. But this type of intelligence is artificial.

What is artificial intelligence?

Smart chatbots differ from their (dumb) predecessors by being based on artificial intelligence (AI). This is a term that has caught on readily as it seems to make sense: artificial – made by people based on a natural model, and intelligence – well we know that means ‘smart’, like us! 

Alan Turing

AI has a long history. After WWII, a number of people independently started to work on intelligent machines. One of these, Alan Turing, gave a landmark lecture in 1947, which set the scene for researching AI by programming computers rather than by building machines. He also proposed the standards for when a program or machine could be considered ‘intelligent’. The term artificial intelligence was coined in 1956 by computer scientist John McCarthy. It was an optimistic time when the scientific community assumed it would soon unpack the secrets of human cognition, and AI would be achieved within years.

When techy people work on artificial intelligence, they focus on increasing computational speed, increasing computational power, and feeding in more source information to a device/system/machine.⁶  To be considered AI, a machine must ‘learn’ and ‘create’ – making new content or connections. 

Two categories of AI are now used: artificial narrow intelligence and artificial general intelligence.

Is ‘narrow intelligence’ actually intelligent?

Most often, what is casually referred to as AI is in the category of artificial narrow intelligence (ANI). We have all been interacting with ANI for some time. Google Translate has been using ANI for years; it is pretty competent at direct sentence translation, although misses the nuance of native speakers. The technology that music platforms use to customise music playlists is a type of ANI; mostly helpful but also sometimes restrictive. 

A Herculaneum scroll from Vesuvius challenge

In more technical applications, ANI refers to programs that work on a specific narrowly-defined problem: for example Alpha Fold which identifies previously unidentified proteins and ‘creates’ (maps) new proteins); monitoring whale migration; or deciphering inaccessible ancient texts for the Vesuvius Challenge.⁷ Such technology can be incredibly effective in the specific area in which it has been programmed: facial recognition, pattern recognition, etc. ANI can integrate sources of information and generate new content (‘learn’) within its programmed area but cannot apply this to new uses or contexts.

Machines using ANI can solve real problems, can overcome the limits of human physiology, and can achieve things that humans cannot. No wonder people are excited about the applications!

These and other ANI applications are often described as ‘smarter than humans’. However, they would  more accurately be described as ‘better than humans at doing this very specific x activity’ such as pattern matching for whale fluke identification and tracking. They are powerful and sophisticated tools. But this narrow skill is not necessarily ‘intelligence’.

Calling them smart or intelligent seems to be just excitement about the incredible variety of useful applications. Sure they are ‘artificial’ being made by humans, and some incredibly smart humans at that. But the machines themselves are not really smart like a human in important ways.

The holy grail: artificial general intelligence

In contrast, the idea is that artificial general intelligence (AGI) would function like human intelligence does – integrating existing and new information and applying this to completely new uses and contexts and even creating entirely novel ideas and concepts. To date, AGI does not exist.

And that’s where smart chatbots comes in. They use AI-based software to conduct conversations, interacting via voice or text in a human-like manner. They use technology known as ‘neural networks’, extremely complex programming with billions (yes!) of parameters (algorithms), and ‘learn’ from vast amounts of information accessed from the web combined with training and feedback (more on this later).

And they can seem pretty smart – interactions with e.g. ChatGPT can appear like a conversation with a person, but they can equally seem very strange at times. 

Smart chatbots are held up in support of claims that computer scientists are nearing artificial general intelligence. And if not quite yet, then it’s only a matter of time until these machines are close to human intelligence, and soon after more intelligent than all of humanity.

But what does ‘close to human intelligence’ even mean, when most inconveniently, we continue to struggle to define intelligence in humans?

What is intelligence?

Before we can ask how machines can be smart, let’s explore how we know a person is smart. A pivotal question is what exactly is ‘intelligence’? 

However, what might seem like a simple question is not at all. In practice, there is no single shared definition of ‘intelligence’. In the field of psychology, the study of human intelligence features varying and competing ways of conceptualising and measuring intelligence.⁸

Take this one: the APA defines intelligence as:

♦️ the ability to derive information, learn from experience, adapt to the environment, understand, and correctly utilize thought and reason.

I don’t need to tell you that I am fussy about definitions. The immediately obvious problem with this definition is that it lists various uses and results of intelligence, without saying what it is.

But that’s the big clue! It doesn’t say what ‘intelligence’ is because intelligence is not a thing. Intelligence is a construct. Construct words represent ideas, concepts and groups of concepts. 

 
Humans use the word intelligence to encompass a range of cognitive and other capacities that contribute to problem-solving abilities and creativity to achieve goals in the world.

However, even after more than a century of study, we cannot precisely describe and agree on the kinds of cognitive abilities to use to delineate intelligence. It involves multiple mechanisms and capacities and coordination. The concept can also encompass adaptability, emotional skills, street smarts, and the ability to navigate the complexities of life. We understand some aspects of intelligence and not others. Possibly, some abilities or coordinating functions may exist that we don’t even know we don’t know about!

As a word, intelligence refers to a multifaceted and not fully understand concept. We use the term commonly, so might think we know what it is (or someone does!), until we look more closely.

So, if intelligence is not a clear cut ‘thing’ in a human, it’s hard to image how it can be a ‘thing’ in a machine.

Intelligence is a human idea about humans

Further to this, our ideas and descriptions of intelligence are based on what humans can and cannot do. Until fairly recently, we had asserted that intelligence and reason was what distinguished us from animals, plants, etc. (a claim that is rapidly falling apart and forcing all sorts of re-defining of the word).

IQ test item in the style of a Raven's Progressive Matrices test.

Following from this assertion, certain normal human physiological attributes and limitations are central to how we measure intelligence, including the speed at which a person can perceive and manage new information, how much they can hold in short-term memory, and how stable and accurate their long-term memories are. For this reason, digit span, or how many numbers a person can repeat back to an examiner, correlates well with general measures of human intelligence.⁹

Thus, our way of understanding and describing intelligence is intrinsically bound up in human capacities, normal limitations, specific academic activities, and our ideas about what humans are.

So, in answer to the question ‘what is intelligence?’ the main take-aways are that intelligence is a complex and abstract construct and not a single ‘thing’, a person cannot simply be described as intelligent or not (it’s not yes/no except in our colloquial use); measures of human intelligence are constructed around human physiology and human purposes (e.g. academia); and ALL approaches to measurement are disputed. 

Thus, it's a massive leap to begin to consider intelligence in non-humans – animals or machines.

Let's just ignore all those complications

In AI publications, you can read concerns about the restriction on the development of machine intelligence because the cognitive sciences have not yet determined exactly what the abilities are that make up human intelligence. When we understand that intelligence is a construct (complex compound idea) and not a thing, we could probably say that cognitive sciences will never be able to determine them exactly! 

Assuming complex ideas are ‘things’ can create stumbling blocks for all sorts of human endeavour. 

But in terms of developing AGI, the technology world just ignores all these complications and poorly understood concepts, assumes ‘intelligence’ is a unitary and uncontested thing, and continues on with the project of trying to build ‘intelligence’ into machines, even hoping it might happen by accident!

So, I have isolated one major source of my irritation when reading AI articles. Discussions about creating ‘artificial intelligence’ use words that refer to complex and disputed concepts but ignore (or do not even realise) that they are using laughingly simplistic ideas instead.

 
I suspect one can only aim to build ‘intelligence’ in a machine using an impoverished or rather vague idea of what intelligence is.

Can’t a machine just be gobsmackingly sophisticated without claims of intelligence?

Machines have been surpassing certain human capacities for a long time. We are comfortable that technology like trucks and pulley systems can surpass our physical abilities. But we have some blind spots when we talk about technology that can surpass some of our specific cognitive abilities.

The normal human limitations important in measuring human ‘intelligence’ mentioned above (like digit span) are irrelevant in describing machine ‘intelligence’. In fact, very fact rapid processing of new data, holding long strings of information (like digits), and storing accurate ‘memories’ are the easiest actions for machines. That’s why we use computers. So machines have surpassed some human cognitive abilities long ago. But we don't call that 'intelligent'.

But we do tend to label the application of these capacities all together by machines as intelligent. For example, computer scientists have long viewed chess as the ultimate test of human intelligence and strategic thinking, and thus a great indicator of artificial intelligence. Various chess AIs from Deep Blue to Alpha Zero now win at grandmaster level. 

But this achievement is not based on anything like intelligence. Chess AI programs - really ANIs - work with limited cognitive mechanisms compared to those used by a human chess player. A human player observes a position and mentally divides it into a collection of sub-positions which are analysed separately, then by their interaction, a simultaneous multi-layer analysis. In contrast, chess programs consider the board position as a single unit. Instead of the multi-layered analysis, chess programs like Deep Blue (20 years ago now!), substitute vast amounts of rapid computation (i.e. analysing millions of possible moves). 

However, scientists, marketers and media like to describe a machine (really a specific program) that can beat a chess master as ‘smarter than a human’, foreshadowing the day when human intelligence is left behind by machines. (And as soon as the word intelligence comes up in an AI article, the question of consciousness follows soon after. I always groan then!)

I think we have a blind spot here in our assumptions about ourselves. This hierarchical use of the word smart or intelligent is the outcome of our long-held and self-absorbed belief that only humans are intelligent and this intelligence is what makes us the ‘peak’ of existence (or formerly, of creation). The logic goes: if we are the peak of existence, and we alone are intelligent, then anything that can outdo humans MUST be even more intelligent.

That logic is actually pretty dumb, right?

In contrast to AI, human intelligence works from a limited set of information, aimed at finding the most plausible explanation and meaning from patterns. While knowing a lot of facts or having high levels of analysis is sometimes described as intelligent in humans, it is also sometimes described as a disability when it is not incorporated with other expected human capacities. 

So, the fact a machine can extract, hold, analyse and ‘compute’ many more ‘bits of information’ than the average human is not particularly convincing evidence of artificial intelligence to me.

I don’t want to downplay how amazing, sophisticated and potentially revolutionary what has been called AI - artificial narrow intelligence - seems to be. As a tool to assist people to achieve new things, it is already proving its capacities. I'm happy to say it is gobsmackingly amazing. But I think intelligent is the wrong word.

The word intelligent might get the public’s attention and draw the research funding. But, so far at least, it seems that we are responding less to intelligent machines and more to programming and marketing by some very smart humans (to be explored more in part 3).

How would we know a machine is intelligent?

SMBC

So, let’s put the complexity about what intelligence is aside, just like the tech world does. Let's leave behind claims that outdoing specific narrow human capacities makes a machine intelligent. Let’s approach it from the other angle – how would a machine demonstrate to you and I that it is intelligent?

In the continuing pursuit of AGI, some computer scientists suggest it could be achieved through machine networks, structures and functions very different from those of humans. But then, would it be intelligence as we currently conceptualise it? And how would we know that the machine was intelligent? How could we tell?

Since the 1950s, Alan Turing’s ‘Turing test’ or ‘Imitation game’ has been the yardstick for the achievement of AGI. Turing proposed that if a machine could successfully convince a knowledgeable human observer of its intelligence, then it should be considered to be intelligent. 

Mmm, so how would you go about convincing humans that a machine is intelligent when you don’t know quite what human intelligence is, or how it works? Would having remonstrating hands do it, as SMBC suggests? 😁

In the case of the smart chatbot, the search for AGI has chosen the proxy that humans use most commonly to ascribe intelligence: language. And that raises a whole lot of wordly concerns for me!

In part 2, I explore how well (or poorly) smart chatbots use language, and how they can convince a human of their intelligence. (In the third part of this post, I’m going to explore how computer scientists have focused more on the ‘convincing’ rather than the ‘intelligence’.)

Back to more comfortable territory for wordly explorations!

Footnotes

1. Yes, it’s a race, a race for profits and power; never forget that! 
2. Unsurprisingly, to explore all this will take me at least 3 posts. I always assume I am starting with a simple question and then when I take the lid off, there is so much in the way of assumptions and problems with words. 
3. It also can mean witty, dressy, snappy, etc., but these other meanings are not my concern today. No one is making claims that AI is a snappy dresser!
4. One might think the ‘i’ in all the Apple products stood for 'internet', but Steve Jobs had earlier said that letter represents the five words: internet, individual, instruct, inform and inspire; technically the ‘i’ doesn’t have an official meaning or abbreviation status.
5. Any attempt to list the players in the AI chatbot 'market' will be out of date almost immediately, but here are two lists to give you a taste, from The Verge and Forbes.
6. I’ll just consistently use machine for easier writing, but it's not always the best term.
7. This story is truly amazing. More than 800 scrolls known as the Herculaneum papyri were carbonized by the eruption of Mount Vesuvius in 79 C.E. Researchers discovered the trove of texts in the 18th century, but attempts to read them proved futile: unrolling them by hand only caused them to fall apart. As part of the Vesuvius Challenge, researchers around the world have been competing to decipher scans of one of the scrolls without ever actually touching it. A group successfully used artificial intelligence to read passages from one of the ancient papyrus scrolls early in 2024.
8. Another source of irritation for me on this topic is how techies ignore the fact there are various and competing theories about human intelligence; those working on AI pick or inherit one and don’t necessarily even know there are other theories. 
9. You might be wondering about IQ or intelligence quotient - the score you get from intelligence test or tests. These scores are estimates or indicators. They are not direct measurements, in contrast with direct measurements like distance or weight, for example. A concrete, specific and direct measure of intelligence is not possible, because of the abstract nature of the construct of intelligence. IQ tests tap into some aspects of what we have called intelligence, particularly academic applications, but they don’t provide a good indication of other aspects, for example creativity or social skills that can be important for problem solving. The illusion that intelligence  and thus IQ is a ‘thing’ is reinforced by this single score that (while useful for some professionals), is also potentially misused. Famously, Stephen Jay Gould devoted an entire book to criticising the use of IQ test results because they are based on the idea that intelligence is a thing, when it is not.

Images, used under Creative Commons license where provided

• Female Scavengers dig through pile of smoking garbage by Philip McMaster, World Sustainability Project https://www.flickr.com/photos/dragonpreneur/3409144522/ [CC BY-NC]
• Crows can count out loud, clipped from social media [Fair dealing]
• History of smart phones by Jojhnjoy, Public domain, via Wikimedia Common https://upload.wikimedia.org/wikipedia/commons/8/8d/Mobile_Phone_Evolution_1992_-_2014.jpg [CC BY-SA] 
• Chat bot interface image from the FRIDA (FReedom of Information Digital Assistant) from the government of the Republic of the Philippines. https://en.wikipedia.org/wiki/Chatbot#/media/File:FRIDA_(FReedom_of_Information_Digital_Assistant)_welcome_interface_(May_2023).png  [Public domain]
• Alan Turing https://commons.wikimedia.org/wiki/File:Alan_Turing_(1912-1954)_in_1936_at_Princeton_University.jpg [Public domain]
• The Greek word for purple extracted from a Herculaneum scroll from Vesuvius challenge https://scrollprize.org/ [Public domain]
• Musk quote made by the author from text from an article at The Guardian 
• IQ test item in the style of a Raven's Progressive Matrices test by Life of Riley https://en.wikipedia.org/wiki/Intelligence_quotient#/media/File:Raven_Matrix.svg [CC BY-SA] 
• Meghan O'Gieblyn quote made by the author adapting text from God, Human, Animal, Machine by Meghan O'Gieblyn, 2022
• Chomsky quote clipped from social media [Fair dealing]
• Hands toon SMBC: https://www.smbc-comics.com/comic/hands-down [Fair dealing]