So we've been dreaming about talking to machines for a lot longer than most people realise, and we're lucky enough to live in a time when Aristotle's dream is becoming a reality.

Frankenstein, in case you weren't aware, is a pretty direct response to Galvani's frog-leg twitching experiments with electricity. "OMG, electricity is the secret force of life itself!". In a way, Michael Crichton's thought in Jurassic Park that we shouldn't use DNA to play god is the same as Shelley's response to electricity.

Neither electricity nor DNA has really played out as this all powerful omnipotent force. Big changes, sure, but not quite the secret of everything. If DNA and electricity were the secrets to life, perhaps we'd all be discussing our latest DNA changes on 23andme with our GPs, while they insert life-extending electrodes into various orifices / slash charging ports.

Instead we have e-scooters and even better vaccines. Revolutionary yes, but perhaps a little more modest than our early imaginations were predicting.

SO when these moments of discovery occur, they often result in thinking that is weirdly limited or overblown, at the same time.

What happened with microwaves was also instructive: they weren't imagined as an augmentation to your kitchen, they were imagined as a replacement for it. Early microwave cookbooks had recipes for everything in them, baking, grilling, you name it, anything was supposedly possible, and you didn't need an oven any more.

The truth is that any exciting trend like AI will become a boring part of the landscape, eventually.

You can see this misplaced excitement in the changing names of Silicon Valley startups. A few years ago you just had to have AI in the title. But over time, we see that it stops making sense. oh, you use AI, but doesn't everyone? It ceases to be interesting or unique to signpost that as anything special.

The history lesson here is that LLM-based AI will go from being a feature to being furniture - nothing terribly special. Or as a now discredited comedian once put it well: "Everything's amazing and nobody's happy"

If there is indeed a historical pattern here, our response to LLMs should perhaps be as follows:

What our use of language about AI demonstrates is an interplay between what we imagine it is for, and what we discover it can actually do. Often these things do not align. The term itself may be the primary source of all the confusion.

Let's have a look at a variety of ways people are thinking about LLMs, as revealed by the terms they use, and let's start with some positive ones. It's less interesting to ask whether they're accurate, but what they say about our imaginations.

And there's a compelling body of thought in linguistics on why we should take the metaphors we use more seriously.

In their seminal work, Metaphors We Live By, Lakoff and Johnson propose that humans navigate new things unconsciously, by means of metaphor. It’s easier to understand a new thing using something we already understand - which is why those of us who work with technology must be so mindful of them.

What do we call this? Hold up smartphone. Is it really a phone? I barely talk to anyone on it. My use of it is nothing like what came before. But we had to use the metaphor of a phone: we needed to give people a rough approximation of something they already understood. A better description might be "personal computer", but that one was already taken.

And this pattern is not new: it occurs over and over again, whenever new technology emerges.

Back when cars were new in the late 1800s, there were all sorts of funky designs being created to figure out what they should look like. And while we can laugh at the specific designs (like this one!), once again, their vision of the future is less interesting than what it says about the metaphors we use when we’re creating new things.

And so it begs the question, what metaphors do we use for AI? I’ve already pointed at a few, but let’s just look at the imagery that you see in the press (going back to Fritz Lang). Is this anyone’s experience of AI? And why is this so regressively sexist?

Firstly, let's just take the word 'artificial'. It really

does say nothing. There are times when making a distinction between something synthetic and something natural is useful, but this is not one of them, because what we are referring to is fundamentally different. We don't say that we eat with an "artificial spoon", nor do we call our hands "natural cutlery".

So using spoken or typed words to interact with a machine: it’s often what people have come to mean when they use the term 'AI'. We've been adapting ourselves to computers ever since they emerged (some of us here being old enough to remember when computing was done solely via a command line), and we'd prefer machines to communicate the way we do. Aristotle's dream.

We have had conversational interfaces for around half a century, without them being particularly intelligent on their own: they relied on the sticky-tape and plaster approach. Think terrible phone IVR systems. The reason they're traditionally awful is that designers had to hand-craft all of the behind the scenes logic to make them seem more intelligent than they actually were, but they were very brittle.

This is why products like Alexa were both fantastic and disappointing at the same time. So much more natural to interact with; but ultimately very limited intelligence, and therefore disappointing.

The second way we can think of AI that I've found useful is as a material for building things. This is an idea from Josh Clark and Veronika Kindred - stop thinking of AI as an end-point or interface, but as an ingredient, a building material. It has shape. It has a grain, like wood. It complements other things, but it is not for everything. Just like a microwave.

We must learn how best to use it, how it fits in, what it's good for. We should not think about it as an end, or a product in itself. More ‘database’, less ‘Terminator’.

In the near future, AI will just be... part of the plumbing - a material we use to build things.

So why these distinctions so important?

When we talk about AI in meetings, it often gives everyone the idea that there is shared understanding that does not really exist.

So we've established that the term 'AI' itself is confusing, and that we need new terms. But there's perhaps a deeper reason why AI is confusing, particularly in the West.

The best example I can give you of this comes from James Cameron. Have a think about the character of the Terminator - what does this symbolise - with its glowing red eyes, skeletal visage, unstoppable desire to punish Sarah Connor?

It’s a demon! I’m not sure if James Cameron is a Christian, and let’s ignore whether there’s any truth in his predictions about technology for a moment. What The Terminator franchise proposes, is that intelligent machines are demons, who will turn our planet into a literal hell on earth.

Bell's theory reveals that Christian-infused cultures are biased towards binary thinking. So if one culturally-Christian response to a revolutionary technology is to imagine hell, what’s the other end? Heaven!

And there’s no shortage of that from the Silicon Valley types. It’s going to solve all our problems! It will be your personal assistant! Life coach! Therapist! Do your laundry! (IMAGES) Weird that these are all female-coded roles, no? Heaven is your mother looking after you, for the rest of your life. I mean, that really one person’s heaven, another’s hell, isn’t it?

Back to that bubble again: this christian bias seems to be a contributing factor. The companies commercialising LLMs have made promises their bottom lines can't keep; investors are only now realising that their dream of making a holy money fountain from AI is simply unrealistic.

In a recent budget announcement, the president of the European Commission Ursula von der Leyen made the prediction that AI will achieve human-level reasoning not by 2050 as previously claimed, but next year.

Now, you could be forgiven for getting over-excited by the progress made in the field, but there are two problems: a. I think we'd all prefer our most powerful politicians to be a little more level-headed (especially in recent times!), but b. when people dug into the claim she made, it turned out it she was parroting marketing statements from Sam Altman at OpenAI and Jensen Huang of NVIDIA.

Thankfully, a group of concerned academics have written a strongly-worded complaint about this claim, so I fully expect the EU to course-correct their over-active imaginations immediately (!).

OK, this next one is really confusing.

If we were to personify this mix of smart and unreliable, we might say in today's world that LLMs are a "masking" technology: they hide their unreliability. Fundamentally, this technology is designed to guess the average, most-likely next word, not get the right answer. Not fraudulent, per-se, but trying so hard to people-please, it's hard to spot the problems. And as this article points out: companies like OpenAI have made it this way, because their incentives are not aligned to telling the truth.

In terms of modelling this, I think we could all be a lot bolder. Again, this is the way we tell ourselves a new story, instead of accepting the confusing, hyperbolic default one. And to show you a taste of this: we could learn a thing or two from the Dutch (thanks, Dad!). The Dutch have no problem cutting through any BS they hear, and the City of Amsterdam has done just this in relation to AI, when they published "Amsterdam’s vision for AI". It's principled, positive, and very pragmatic.

I won't recap too much of it, but just a few select points.

First of all, they have specifically outlined what they think it is, and is not, in a refreshingly Dutch way.

It is also very concerned about the dependence on big tech companies, in a very pragmatic way. They acknowledge that the dependence is not going away, but that the city should be prepared to invest in alternatives, and make explicit and conscious choices towards greater independence.

Do we have anyone from institutions who have a manifesto for AI? Please talk to me afterwards!

Another way we should be pragmatic, is to be very, very clear about the purpose and phase of any work involving AI. I think a lot of people think they're making products, when they are in fact experimenting. Or worse, they're trying to solve problems with AI, when we already have established solutions, that just aren't quite so sexy.

Speaking of sexy: To paraphrase Jeff Goldblum in Jurassic Park: "let's stop to think about whether we should, not just if we could."

Here are the broad phases and types of work I think are involved in adopting new general purpose technology.

The confusion about AI means that people are very unclear about which of these four things they're actually doing, and whether it's actually worth it.

Or when it comes to this phase, what we're seeing is that organisations are scaling or adopting this technology before applying any guidelines, or even having a notion of what 'safe' might mean. It's the classic Silicon Valley FAFO.

Imagine, for a moment, that a traditional health product (like a new drug) was having an effect on suicide rates. Regulatory oversight would (most of the time) have ensured that such risks would have been identified, and strict safety guidelines applied.

But when it comes to software and the startup world, we seem curiously timid when it comes to oversight. OpenAI have admitted that around a million people per day are discussing the possibility of ending their lives with ChatGPT.

As your institutions are serving large groups of young, often vulnerable people, I'd think that much stronger guidelines might be a useful counter-force to the libertarian values of Silicon Valley.

So our job is to be the voice of sanity here. Very often a question like, "Can't we use AI to do that?" (Point 4) which I'm sure many of you have faced, should actually be, "Is an LLM appropriate for solving this problem?" (Point 2, and also a moving target). Again, the excitement and overpromise has fried our brains a bit; we're forgetting to figure out what the problem is that we're actually solving.

So, to help people out here, we need to draw a hard distinction between experimentation, research and design, what measures we take around safety, before we try and scale a product.

One sign that you are mature with regards to products vs. experiments, is that most ideas or prototypes are disposed of. You're trying more things, to find a very small-subset of workable ideas. Who tracks this number, BTW?

This is Harry Brignull - an ex-colleague of mine. You might have heard of something he identified: “Dark Patterns”, or as we say now, “Deceptive Design”. Psychological tricks used in interfaces that businesses use to manipulate users. Harry collected dozens of these tactics back in 2009.

So Harry was raising attention about these tricks for years, trying to get governments to change the laws so that businesses can’t do this any more. But he gave up after a few years; he thought that he’d never get the kind of consumer protection into law. So it was quite a surprise when recently, he started getting calls from law-firms and legislators. Harry has appeared as a key expert in the EU parliament, advising them on what the law should be.

So the point here, it can take us YEARS to figure out how we should ethically, legally, and productively use a technology. How many decades did it take us to figure out that we shouldn’t be sharing smoke in indoor spaces?

And the only sensible approach, it seems to me, is to speed up the learning loop. Get to the answers about how we should use this thing faster, and place some guardrails around it. Now that’s really, really hard, when you’re still figuring it out - you might not be aware yet, of the effect its having (thinking about how hard 2nd and 3rd order consequences are to predict). but like the city of Amsterdam, it’s useful to plant a flag in the sand and say “this, but not this”. At least you have something that can be challenged, and have it evolve.

Now, you folks are the people in change management, I'd imagine this is an evergreen question: how might you speed up the response of an institution?

Anyway, before I close our, a quick summary of the main points today.

(RECAP) If you only took one thing from this talk, it might really be this final point.

We can be much more intentional with our words, to manage expectations, and forge sensible strategies.

Quick example: someone recently complained to me that no-one was reading their slack updates, here's what I said.

If it takes you an hour to write that update, from now on, spend 50% of that time writing the first headline, and 50% of the time writing the rest. Make sure that first sentence people read is compelling, clear, and concise.

In other words, don't assume the default language is the right approach.

We create meaning in the minds of others with the words we use. So if you want to be positive, principled, and pragmatic, start with a hard look at the words you're using.

Words are the primary medium we use to share our ideas, but so often we transmit the wrong framing when we use terms we've heard others use, in an attempt to make it easy to understand for all.

And on that...

The term "artificial intelligence" is perhaps so misleading, that one of the originators of the term, Marvin Minsky, effectively disowned it when he said that it's a "suitcase term" - i.e. people project anything they like into it. Or as I like to say, it needs a lot of unpacking.

To reduce the confusion around it so that we can figure out what it's for, and what it's not, we must change how we talk about it.

As I explored earlier, the academics dispute that what we call artificial 'intelligence' is actually intelligent at all.

Let's start by avoiding the term itself. Be more specific than the term 'artificial intelligence', where possible. Reframe it for others. You might have noticed that I haven't used it very much, I've tried to use the term 'LLM' most of the time... and for my final point, let's look at an example from someone much smarter than me.

Stewart Butterfield, the CEO of Slack, actually banned the team from using the word "chat" when they were creating it. He thought that it would be impossible to revolutionise a communication medium, if you used the old language.

His theory was that if the team used the word "chat", a word with strong associations to existing "chat" products like Skype and ICQ, it would both cheapen the product (chat is often a derogatory way of describing communication) and it would tie up their creative thinking with old ideas.

And that, everyone, is really my final point. The term "Artificial Intelligence" is an old, tired, unrealistic piece of hyperbole, that has ended up confusing a lot of people. Let's put it behind us, and forge new stories, with new terms to help us make better sense of what is, and what to do next.