Medium Article

June 7, 2024

How Apple can take a position of dominance in AI — if they want it

AI + Apple: thoughts on how Apple could change AI forever for the average consumer, redefine how AI is used, and deliver an experience that no other company could

article originally posted:

https://medium.com/@adamrtsmith/how-apple-can-take-a-position-of-dominance-in-ai-if-they-want-it-e0e69afd08c7

As we approach the 2024 WWDC keynote on June 10th, expectations on Apple, and what they will or will not announce regarding Artificial Intelligence is at a fever pitch. There are rumours of deals with OpenAI, and / or Google Gemini to power underlying features and capabilities. There has been research papers from Apple itself about on-board LLMs for speed and privacy. There’s the Vision Pro, which seems to have gone quiet in the US where it was initially released, but now the rumour mill is abuzz with the potential volume of units shipping around the world ahead of the expanded international launch of the device expected to be announced at WWDC. There’s the news beginning to surface of economists and pundits saying Apple is about to have another iPhone moment with AI, and that they will once again change the industry with what’s coming next week on the AI front. There’s the absolutely insane fact that the iPad got the next-gen M4 processors before their laptop and desktop counterparts, and there has been a number of pre-WWDC feature announcements from Apple directly — all of which are abnormal to say the least for Apple.

There’s no question something different is happening. Any one of the above alone could be considered as odd, or a rare moment of transparency from a company where secrecy around new products, and features, is paramount. All happening before WWDC this year… that’s really out of character. It certainly feels as if Apple is making room in their WWDC keynote to give ample time for a huge, game changing announcement — and not in their historical “one more thing…” way.

All signals indicate that this will be Apple’s entry into the AI space, and if they do it correctly, they can lead the AI space the same way they lead the smartphone space — with absolute ingenuity, simplicity, and incomprehensible dominance.

Historically, Apple’s always been a late entrant to new markets. Not because, as some would argue, that they’re slow to move, nor because they ignore markets until they can’t any longer, but because they only enter the markets when there is an established product group of products they can contrast their version against. It’s impossible to be better, when theres nothing to be better than. It’s genius.

Apple waits for competitors to establish early expectations, educate the market on the new product space, lets them spend the time and money and effort to explain the value, the virtues, the capabilities and the functionality of the product category, and build the hype. Then, and only then, does Apple step in with a cut above, and can sell on experience, simplicity, and magic. There’s no denying that the iPhone was mind bindingly different than other smartphones of the era, nor that the iPad was a whole other experience compared to existing tablets, or that the Vision Pro was a measure above other devices in terms of quality, build, design and OS experience.

This is how Apple dominates. And this, if I’m not wrong, is what they will do with AI.

Will Apple overpower the AI space the way they overpowered the smartphone space? I think they can, I hope they will. Let me tell you how and why:

Apple knows me better than anyone. It knows my family, my friends, my business contacts — how I talk, my sense of humour, the things I do, where I am, where I go, all my photos and people who are in them and the locations they were taken, what I sound like when I talk on the phone or to Siri, and how it differs from person to person, situation to situation, what I spend my free time on. With a little analysis, they can easily determine what I care about, and what’s important to me. They know it all — because it’s all in iCloud, or on one of my multitude of Apple devices. 2 decades worth of information about me, and by me. And they likely know the same about anyone I communicate with via phone, email, and text — especially if they too are Apple ecosystem users. Yeah, they know you too.

As a training model for humans, more specifically for individuals, for “you” and “me”, the Apple ecosystem is an unprecedented powerhouse without competition. No other company has as much real information at their disposal about me as a person, a friend, a businessperson, nor about my private life, than Apple.

Think about it: my devices and all their sensors, where they are physically, the things I write or say on those devices, everything I access, all of it can be used to understand me as a human being. Simple contextual web queries in the background about my current environment, locale, and well, anything those sensors are “aware” of, can be used with the plethora of data I have processed through the ecosystem over the years to have as much awareness of my real-world situation, and even how I behave in each as I do myself, if not more.

Throw ‘in-ecosystem-AI’ into the mix, which we know Apple has been working on thanks to their own research documents, and something truly magical happens. Something no other company is really in any position to do: create the first individualised and personal AI agents for each user on the ecosystem. This is unique, and has the power to truly create smart, aware, agentive experiences.

This doesn’t scare me, it’s going to happen anyway. I’m actually insanely excited about the possibilities. More on that later.

How’s this different than having a ChatGPT account, or using the generative AI features everyone and their brother is slapping, often haphazardly, into their products? These systems only know what you do with them, and nothing more. They’re pretty limited in terms of their exposure to you as an individual when you think about it.

In the simplest of examples, I have 3 or 4 disparate “write with AI” features across different email platforms alone that I actively use today (no, friends… of course I personally write all the emails I send you… honestly). I mean it’s cool to have AI assistants popping up in so many products, but based on how I use each app, each assistant is learning different things about me: different goals, objectives, personas, ways of speaking or writing, etc. Each of these independent, isolated, and completely disconnected AI-islands are learning different things about me — throw in ChatGPT, image generators, sales tools, coding assistants etc… each with their own learnings and assumptions about me, separated and disconnected from one another. Over time as these assistants proliferate that’s going to get really annoying, really quickly. I already find myself frustrated by the lack of shared knowledge in all the different AI tools I use, that I find myself using the “stick-on” AI tools less and less… and we’re still in the very, very early days.

These disparate AI assistants are rather dumb when you think about it. They have super constrained visibility beyond their own systems, learning equally from what every other user does as they do from me. Even if they prioritise my own limited interactions with them from a learning perspective, as opposed to applying equal weighting to the requests they get and the responses they generate from and for all their other users — they don’t really know ‘me’ in a comprehensive manner. They were trained on emails or designs or documents created by folks who are likely NOT very much like me, don’t TALK like me, or BEHAVE like me or THINK like me. AI islands like this will only push the creations they produce further and further into mediocrity — as they begin to learn from their own creations more and more — let’s call it AI reinforcement bias — and ultimately, everything will become an echo chamber which looks, sounds, reads and feels the same.

The last thing I want to do is work with 100 AI assistants, each knowing little about me, and get more generic results over time. I don’t want to remember which one remembers I’m funny, which one remembers I have a company called Tension Consulting, which one remembers where I’ve been, and which one remembers that once upon a time I had a social life. Individual AIs, all encapsulated in the apps they exist within, is exhausting, let alone concerning from a personal brand perspective.

Let’s get back to Apple and everything they know (or could know) about me, and why that matters so much in terms of AI. Let’s say, for arguments sake, that Apple has been building on-board AI, let’s say they have signed with OpenAI, and let’s say, putting privacy aside for the moment, that they can use any information I have ever stored in photos (let’s also ignore the recent reappearing deleted photos bug in iOS and what the heck that could mean), or messages, or email, or iCloud, or Music, or AppleTV… the list goes on and on. Let’s assume it’s all true. Let’s assume that’s what we’re getting a preview of in a week. Note: I have no insider information nor connections who know anything on the matter — but I do have an idea that could change the face of AI as we know it today.

Apple’s AI, with all it knows about me, talking to the AI’s that know very little about me, but know all there is to know about the systems they service, on my behalf, and in the background.

If they’re smart, Apple will introduce not only their own AI, but an AI to AI framework. Their own AI would leverage all my information, photos, messages, emails, contacts, web history, phone calls, FaceTime videos, notes, calendar events, reminders, documents, settings and preferences to create the first person-centric AI. The one that knows me, all I do, say and think. The Adam AI, or Leah AI, Logan AI, or Ryleigh AI. The AI version of you and me — as an assistant, as a worker bee, that does things for us, on our behalf, and at our request. This individualised AI will represent us, add context for us, and remember to say the things we forget to say, or prompt in this case.

The second part is where this gets really interesting: having my individualised AI, communicate with and instruct all the other dumb AI islands for me. By creating an AI-to-AI framework, Apple could essentially create the first widely available network of AIs, where Apple’s AI Agents, specialised in knowing me and my data over the last 20 some odd years, acting on my behalf, could talk to, inform, direct, manage, and work with the disparate AIs specialised in knowing the product or service they serve — augmenting their internal LLMs with my personal information from the Apple ecosystem.

AI to AI, whether through conversation, APIs, Agent to Agent interoperability, or simply by providing access to my Apple dataset to further understand ‘me’, changes everything.

Think of it as a personal assistant. When you have just hired a new assistant, you probably have to give a ton more information to contextualise each request you give them. For example, if you want to make a dinner reservation with a client, you’d have to give them the day, time, name of the client, what type of restaurant you want, food you like and don’t like, the vibe you prefer, ambiance, area of town etc. Once that assistant has been with your for years, you can likely just say “I want to have dinner with Anne H. this week” and leave it at that — your assistant will likely know when you’re free, what times, dates, areas of the city you prefer, food you like, and probably the preferences of your client too. If there’s a back and forth or negotiation between your assistant and your client’s assistant, they can probably work out the details on your behalf without troubling you with the details. That’s what Apple creating individualised AI Agents can mean — it’s not about “write this email for me”, it’s not about “summarise this document or meeting”. It’s about knowing you well enough to give the additional context required to get the outcome you want, without you having to say or do anything further. I would call it, if I was Apple, AgentKit.

With this in mind, my personalised Apple individualised AI Agent, becomes my broker to all other AIs. AIs I don’t have to see or interact with at all. They’re still there, in the background, doing what they currently do, and likely much more, given the additional context that Apple’s data can provide them — without me ever having to prompt them directly. I use my own IA Agent, which understands my prompts because it knows me, has all my history for context, and can communicate on my behalf better than I can write the appropriate prompt myself in the third party AI. It’s more akin to an API: me asking my AI, to tell the other AI, to do what I want — and filling in all the blanks, and knowing what I want better than I do myself.

Of course Apple’s AI plans also will include bringing Siri to what it always was intended to be. It will enhance built-in apps, allow me to generate text and images, enhance my listening, watching and messaging experiences, it will likely allow more predictive actions on device in all Apple apps in numerous ways. Apple will, perhaps eventually, allow access to the internal model for any app on the Apple ecosystem too, whether iOS, iPadOS, Mac OS, TVOS or VisionOS. But it will not be evident to the user that they are using these third party AIs — it will look, feel and sound like Apple’s AI — whether they call it Siri or not — it will be one AI interface that commands them all.

What I’m talking about here really, is the concept of nodal AI networks. Hundreds of thousands of interconnected, specialised microservice AIs that all service the commands of a handful of Individualised AI Agents from Apple, Google, and Microsoft. Gone will be the days where there’s one AI in this app, another AI in this other app, and oh, look this new app has a new AI too. It’s too many AIs, and most of them suck anyway because they’re trying to do too much.

This really is the only way AI assistants can work at scale. There’s no way, memory enabled or not, that any AI assistant will truly get to know me, as they are either locked exclusively into a property, a product or an app. They will never, ever have the exposure that Apple’s products have to me, nor should they try. It’s impossible, mostly. That being said, it could be possible for OpenAI perhaps, if they were granted access to Apple’s devices and software, all my data, and in the background formed partnerships with every single AI assistant provided by every product or service out there, tokenising “ME” and my personal LLM or GPT and ensuring that it had the ability to communicate with all other AI assistants on the market, powered by ChatGPT or not. I think perhaps that’s what the idea behind individual GPTs, and what the GPT store is setting the groundwork for — but it’s crazy ambitious, and a long shot. And if Apple plays its cards right, they will own this market, for Apple users at least.

Sure, Microsoft has Copilot (powered by ChatGPT), and Google has Gemini. Microsoft has Windows, Teams, Surface, all their enterprise tools, and xBox. Google has Android, Chromebooks, Gmail, and Google Workspace. Either of these companies can make a strong case for enterprise or business AIs. And Google might have a play for the consumer space too. But none, NONE, have the level of access and deep integrations with how I live, share, communicate and think as Apple. None.

One AI, my AI, to rule them all. Let the battle for Middle Earth begin.

see all news