Video: The 3 Types of AI (Dawid Naude)

In this video, Dawid Naude, an independent AI Consultant who until recently was the head of generative AI at Accenture Technology provides a framework on “The 3 types of AI”, which you might find useful as you develop your AI Strategy.

Like to setup a call?

Please submit this form if you would like us to setup a call with Dawid.


This is a model that we've come up with. It's how to think about how AI is felt within your organization.

It's three categories: invisible, assistive, and integral. So "invisible" is where you don't actually see the AI. It's hidden. The great thing about that is there's no change management or anything like that required because it's completely hidden. The most obvious example of this is not a new one. It's spam filtering.

You only know that your spam filter hasn't worked when you get a spam message. Or when somebody contacts you to say, "I sent you an email, why haven't you replied?" and you find it in your spam inbox. So as much as you can put things into as invisible, the better. Things like being able to direct a call to the right place based on an email, or, I mean, support agent based on the context of an email.

That's something that's invisible. Nobody really sees it. And if you get it wrong, it's okay. It just gets the person redirected to the right area.

The next one is the chat GPT world. This is assistive. It doesn't replace you. It makes it easier for you to do your job. So I can ask a chat bot, "What's your leave policy? And am I allowed to take, particularly if lost time on annual leave?"

And it'll give you that answer. But you could have got that answer anyways from going to the different portals. So it helps you, but it doesn't necessarily replace you. Now, obviously, the different speeds at which it helps you depend on how familiar you are with the tool and also how well the tool has been configured for your use. So you don't need to expect a user to prompt engineering and all of that. That's at the middle pillar. The third pillar is integral.

This is where it gives you superpowers and allows you to do things you simply cannot do without the AI. This is things like allowing a call center agent who's English to speak in Mandarin to a customer.

At the beginning, this is all chat, but with how quickly audio is moving, it's very soon. In fact, I'd be blown away if the first trials aren't happening this year where we have somebody speaking on the phone in English to somebody in Mandarin, and they're speaking and it's a fluent conversation, maybe with a small lag, but it's a fluent conversation.

It's being able to give your full-stack Java engineer the ability to write in Cobol so that you're not no longer single point dependent on a Cobol engineer who's probably gonna retire in the next two years or should have retired in the last five. So it's giving them superpowers. It's allowing a much more junior contract lawyer to actually give a whole lot of very important contract markups that typically would have had to rely on somebody a lot more senior, which has probably led to a lot of delays to getting work done. So integral is really where it gives you those superpowers.

And it's also no surprise that the change management effort is very different for each of these.

So invisible, there's basically no change meant for the effort. It's you just basically need to train people on what do you do when it's not working the way you expected?

What many people are completely overlooking at this point is the change management effort for assistive. So I've seen this myself. When I was at Accenture, we organized fifty licenses of Chachi BT enterprise for people in Australia. I was so excited. Obviously, I'm an AI guy. I was using these things every day.

What I found was out of the fifty, maybe five of them were using them regularly. Most had used it once or twice or maybe now and again. And then there was a big group, maybe ten of them, that had never even logged in. So it was a complete, typical bell curve. And I asked one of them, "How are you finding chat GPT?" And they said, "Alright, it's pretty good. I used it to write a cold outreach email to this client. I'm trying to get a meeting with, but it was a bit salesy."

And then that led them to stop. You know, they didn't try again. I said, "You actually just literally have to tell ChatPT that was too salesy."

And then it will rewrite it. So there's a big change management effort that's required there. The thing that we all need to realize though is we need the change management effort, but our kids and grandkids definitely won't. They're gonna operate in this universe.

I remember early in my career when I was at NAB, there was a banker that was recounting his weekend at dinner. And he said that he had gotten into a massive argument over the date something had happened, some historical event, and basically, it spoiled the whole family dinner. They were arguing with each other. I was a grad, and I just said, "Why don't you just Google it?" Like, to see what the answer is. And he paused for a second and thought, "Yeah, I don't know why. We just didn't think of that."

He just wasn't used to doing that. It was a very new behavior for him to Google the internet for an answer to something. That is absolutely what the coming generation is gonna treat these tools as. They're gonna expect that you just have AI and having to log into workday to submit a leave request or get a pay slip or something like that is absolutely not gonna be what the future generation expects. They're gonna just ask, "Can I log a leave request, sure, which days?"

Yeah, great. Done. And it does all the interactions for you.