AI is about to change how you use computers
gatesnotes.com
gatesnotes.com
Good, that's how I would like it to be!
1. Homomorphic Encryption
2. Self-Hosted AI Systems
Homomorphic encryption is the idea that you can perform ML operations on encrypted data -- so that the providers of powerful AI need not know precisely what data you have.
Self-Hosted AI is just what it sounds like -- what if your Alexa processed on-device, and never left your home? What if your mobile devices could "phone home" to it from wherever you were, using an encrypted tunnel?
The second point is possible today, but the issue is keeping your local model to pace with the cloud offerings.
I guess I'm suggesting that Mr. Gates is probably right, even if it makes us here on HN uncomfortable.
You can use alternatives but you have to maintain them meticulously. If any of their websites or services can get a bit in, you're fucked.
Welcome to the world of quantum computers, pseudo-AGI and corporate interests.
Maintaining my own alternatives isn't a big deal. I'm largely doing that anyway.
Oh yeah, don't forget they are sharing this information with an ad agency who covered the installation cost of not only your grocery store, but also your bank and gas station. All of those prefernces, tied together in a neat little digital folder, ready to sell to the highest bidding government agency.
On your topic, I'm not totally helpless, though. I can't do anything about store surveillance except go to stores that don't do that (while they still exist).
That said... I can and so put my phone into airplane mode before I enter stores, and I avoid using credit cards in them, specifically to help protect against those kinds of spying.
"The next generation of interesting software will be done on the Macintosh, not the IBM PC" (1984)
"I see little commercial potential for the internet for the next 10 years" (1994)
"Today's Internet is not the information highway I imagine, although you can think of it as the beginning of the highway" (1995)
"There are no significant bugs in our released software that any significant number of users want fixed" (1995)
"One thing we have got to change in our strategy - allowing Office documents to be rendered very well by other peoples browsers is one of the most destructive things we could do to the company. We have to stop putting any effort into this and make sure that Office documents very well depends on PROPRIETARY IE capabilities" (1998)
Under oath: "I don't recall" 6x, "I don't remember" 14x, "I don't know" 22x (1998)
"[E-mail] spam will be a thing of the past in two years' time" (2004)
That understanding will belong to some company whose interests and goals are not yours.
This will not bother most people. Most people are willing to use GMail, even though it snoops on their private emails and uses that info for advertising purposes.
While I agree with your larger point and am no fan of Google's privacy practices, they stopped this specific behavior in 2017
https://www.npr.org/sections/thetwo-way/2017/06/26/534451513...
"The activity information that we collect may include:
Terms that you search for
Videos that you watch
Views and interactions with content and ads
Voice and audio information
Purchase activity
People with whom you communicate or share content"
[1] https://policies.google.com/privacy?hl=en-GB#infocollect
Another take on this topic: It is reality right now, but only for poor people, so we can't see it. This will change, we will all suffer from automation.
But we're a long, long, long, long, long way away from agents. What we have now aren't even precursors to agents, more like very distant cousins to an ancestor.
In order to make what he's talking about, you'd need true general artificial intelligence operating at a human's level or above.
And if we have that, the entire world is going to change regardless - personal assistants will be insignificant in comparison.
It's difficult to predict when a breakthrough like that would happen, but I doubt it'll be within the next 200 years.
What we have now isn't even a stepping stone in the right direction, it's just smoke and mirrors that looks like it is if you only look at it on a surface level.
Slack lets your conversation kinda go.
What happens when slack starts remembering tasks someone asked you to do... follows up with you to make sure you did it? And the person that originally asked you?
Right now you arrive at work and you have a few things to do, but your brain only remembers 1 of them, and barely at that. With AI it will have a few iterations on emails ready for you.
This is all possible right now with an LLM of today's caliber. He's not talking about general AI.
See here:
"Agents are smarter. They’re proactive—capable of making suggestions before you ask for them. They accomplish tasks across applications. They improve over time because they remember your activities and recognize intent and patterns in your behavior."
None of that is possible with today's LLMs - not in the way he's describing it.
> They’re proactive—capable of making suggestions before you ask for them.
An LLM can capture things that are "todos" very easily now. If someone messages you and says "can you collect data on the Brenda client" it can know that this is a request.
Further, if you give it access to notion or company wiki, it could be tuned to understand your clients, work and much much more.. The LLM can, in today's work: open an email, or a VSCode instance with some pre-filled connection to your analytics database - maybe even a sample query with the Brenda client pre-filled.
Across applications - this is more-or-less something that will be coming soon. Within a few months you will see pretty much every major OS (linux, Mac, Windows) announce OS level APIs to interact with a basic LLM, and paid for LLMs. This will enable applications to start communicating with each other because they will generally understand what you're typing and output function calls..
When applications can ubiquitously call "predict(...)" at the OS level, and probably many other AI APIs that will be introduced, you will see a HUGE change in interactivity.
Now that being said, it sounds like you're picturing Tom Cruise in Minority Report, or Will Smith in I-Robot.
There's a huge in-between you're extending his conversation to that's totally unnecessary.
Worse, you'd need an AGI that was relentless, creative, able to lie and deceive, to mimic, to charm and flatter, to subtly threaten or act subservient, to be taciturn and know when to be silent, and generally socially engineer, understand rules and break them.
And that's just me dealing with a bank on a normal day with someone who actually wants to help me but is hamstrung by policies, checklists, scripts and service non-interoperability.
Wait until your "agent" is dealing with another "agent" purposed by the company (adversary) to also dodge, lie, trick you into agreement, procrastinate, time-waste, discombobulate and deliberately annoy.
The moment Bill Gates' "Agent" meets its counterpart from Verizon it will be like that first time you set your mail auto-responder to respond to their auto-responder.... they will instantaneously consume all the computing resources on the planet while locked in a battle of corporate shitfuckery to the death.
This is the endgame for "online computing services".
So, Bing Chat?
Yeah could you f**ng not? I pay taxes and insurance to get proper medical guidance, not answers from a bot. Public healthcare in the uk is nearly dead anyway and their patch fixes are to use “virtual” nonsense, while private healthcare is riddled by “ai” that does nothing but annoy customers. BabylonHealth, a former uk ai “healthcare” darling, rightfully went under because among many things it was gibberish.
Use of GPT-4 to Analyze Medical Records of Patients With Extensive Investigations and Delayed Diagnosis
"Six patients 65 years or older (2 women and 4 men) were included in the analysis. The accuracy of the primary diagnoses made by GPT-4, clinicians, and Isabel DDx Companion was 4 of 6 patients (66.7%), 2 of 6 patients (33.3%), and 0 patients, respectively. If including differential diagnoses, the accuracy was 5 of 6 (83.3%) for GPT-4, 3 of 6 (50.0%) for clinicians, and 2 of 6 (33.3%) for Isabel DDx Companion"
AI is not intelligence. It lacks empathy. It lacks just about everything a doctor does except regurgitate diagnosis based on data -- data gathered by doctors.
But this alone invalidates the paper:
“AI relies on clinical imaging.1 In low-income countries, where specialist care may be lacking”
Have they been to a british hospital recently? Specialist care scarcity is fast approaching that of low income countries. The only option left is the poor man’s choice - chatbots.
Still a highly non trivial result.
>But this alone invalidates the paper:
“AI relies on clinical imaging.1 In low-income countries, where specialist care may be lacking”
How does that invalidate the paper ?
This kind of thinking, to me, smacks of the kind of ignorant projections people used to make in the early 20th century about technology in the 21st century.
“Robotics will advance, so everyone will have their own personal robot assistant!” Nope, though we do have robots doing things that are commensurate with their cost and the value they provide.
“Advances in aviation mean everyone will have a flying car.” Nope again. Turns out the added utility of being able to fly directly to places vs the cost doesn’t trump a car in most cases, and when those cases don’t apply, airplanes are plenty economical.
“You’ll just tell your device what you want to do.” No again. In some cases this is useful (I already tell my device to remind me about something or to schedule some sort of event on my calendar, where the time to input far exceeds speaking it), but in most cases, I can do it faster myself with a good UI, or maybe I don’t want everyone around me hearing me blather into my phone.
It’s like, just imagine for a second what it would be like to have to dictate everything you do on your phone. It would be utterly exhausting.
https://arxiv.org/abs/2311.00871
The question I have is, what if these agents are really good at doing things they learned about, but not in new things? How will that impact people's actions?
Will we end up with a static world of people doing only things the AI can do? A kind of 1999 Matrix stasis world of 'the best we can do now'?
What will happen to what Arthur Kessler called 'blue' thoughts in such an AI world?
Because IMO, the purpose of computers is to make models and see what happens. Test ideas in a powerful way via computation. If the AI can ONLY select existing models and not make news ones, or even prevent you from making new ones, then I don't see how these AI agents are anything but a step backwards.
Alan Kay gave a great talk called The Best Way to Predict the Future is to Invent it taht touches on some of these ideas.
will your agent be programmed to tell you about one fast food vendor over another? will that change over time? will you be told?
of course things look rosy up there in that ivory tower.
but how is the foundation going to hold up?
Kansas City Shuffles inside Kansas City Shuffles. We refuse to deal with the truth.
We need to trust our kind or we will have to take down our past, the ways we lived and worked.
It's a generational conflict inside each and every one of us.
You know how to code. Decode.
What is changing about how we use computers that is genuinely getting us more in touch, more aware, more capable as computer users?
I see the introduction of AI to the SWE workforce as roughly equivalent to the introduction of digital kitchen scales, thermometers, etc. to professional kitchens. It's going to help us up our game, which is awesome! It's going to make a lot of the boring, mechanical stuff simpler so we can focus on the creative aspects of our jobs. But a burger flipping robot isn't replacing a michelin star chef anytime soon, and AI isn't replacing you anytime soon.
There are some fairly intractable problems in the "AI writes software" space. I don't think they're getting solved any time soon.
If software engineers have been replaced, it won't be any of us telling the AI anything, it would be a manager. Ask a manager about Kafka, I recon more would respond with something about Franz than anything about Apache.
> And if you tell it at a higher level than this, how will it know what tools it has available? How will it deploy stuff?
Step 1. Ask it to come up with a plan, after giving it read access to your corporate documents.
Step 2. Ask it to execute the plan it came up with in step 1, after giving it write access.
Of course right now there's a step 3: "Get real engineer to read and then perform business continuity/disaster recovery plan because current AI is about as good at this as someone fresh out of university, what on earth were you thinking, …" etc.
I don't know how long we've got in these jobs, but based on the rate of change (and the training cost for the larger models and supply shortage for the chips used to train the better models) I'm thinking at least 2 and no more than 15 years — though I'm hoping more towards the latter.
The real reason chefs won't be released by robots sooner than AI and software devs is a lot more boring. Sensorimotor and physical perception is harder than reasoning and "creativity".
It doesn't help that we don't know how to digitize taste so any model with a good sense of taste will have to develop it indirectly incentivized by something else (eg a language model training on recipes).
GPT is a predictor. It will just continue to reduce loss until it has modelled the data entirely.
They're overflowing with ideas and knowledge and passion, and "all" you have to do is point them in the right direction. Except that when you review their code, you find that they didn't consider about a thousand different edge cases. Oh, and they didn't follow the style guide. Oh, and they are way over-focused on the wrong parts of the problem, prematurely optimizing performance in places where it doesn't matter. Oh, and their code is nigh-on unreadable. Oh, and they wrote a bunch of code that is already provided in libraries, but they just didn't know about it, and their implementation is probably full of subtle bugs so they should just use the library version. Oh, and they forgot to update the CI because now they need to pull in a new dependency to run their tests. Oh, and... the list goes on.
I don't ever see AI progressing past the regurgitation stage; you can give it as much knowledge as you want, and it can rearrange and reproduce and restate that knowledge in a thousand different ways; but we're so far from AI being able to handle all the details of our work, that as I said above, fully specifying the problem is going to be just as much work as doing the work yourself. And you'll still need expertise to do so, because our software systems are complex and full of nuance that can't be easily communicated to a machine.
AI doomers always strike me as people who have never had to try to corral interns; maybe that's an overly specific life experience to expect someone to have, but it's a really useful proxy for how much "more productive" we're going to get with AI. A really good intern can solve simple problems given exact constraints, but anything requiring lateral thinking will take them a lot of coaxing to get to the right solution. And that's okay! People can learn and get better. But I don't see AI getting better "enough" to take on anything beyond that first rung of the ladder of complexity.
So? Assuming you're right, GPT-3 was not at the level of intern. GPT-2 could not even write coherent text.
I bet you didn't expect any of those developments either.
It's interesting how much people struggle to look forward. I guess we never needed to for nearly all of our evolutionary history. Like very few people genuinely think they'll be replaced right before they are.
Your argument basically boils down to "I have a hunch language models will stop improving".
Your hunch is unfounded, backed by nothing except vague assertions that "surely this time" these goals won't be reached.
Assertions that many people parroted just a few years ago and have continued to be proven wrong.
The text is data. Gradient descent and prediction will strive the model the data. The data will be modelled. That's really all there is to it.
To be clear, I'm referring to LLM. Obviously ML is used for a lot of analytics, but I also think that most of the hype around ML has died down as companies realized that not every problem has a ML solution. In a few years the same will happen with LLM.
There's some delicious irony in this: tech-savvy HN posters are so bullish on AI replacing everyone's jobs, yet there's been so many posts here about how people have lost their Google/Apple/bank/whatever accounts due to AI accidentally flagging, then being unable to recover because they can't get in touch with an actual human, with begging for help on social media as their last resort.
It seems like a variant of Gell-Mann amnesia[0]: people believe AI is capable of replacing humans in so many aspects of daily life, except for the aspects of daily life which they personally experience.
Also, there have been substantial new developments and discoveries about what transformer models do (both internally and in terms of capacity) every week or two for most of this year, so why do you think there's no known path forward?
Thinking of logic, I just tried the following with gpt-3.5, gpt-4, and gpt-4-1106-preview. The newest model spotted the trick (and then still got it wrong), the older two didn't even spot the trick. Can you spot the trick?
-
A person is in Nairobi. They board a plane, fly 9000 km north, then 1000 km east, then 9000 km south, then 1000 km west. Where are they now?
I don't know about you but my team's backlog is virtually endless. If AI/LLM/whatever can automate away a ton of dev work, that doesn't mean we're going to fire 90% of our team. It means we might have a glimpse of being able to actually make a dent in all the work that we want to do but can't.
Whether customers prefer that or not may or may not matter to the business decisions that are made (given that it's not a perfect customer-driven market.)
Have you been to McDonald's? They really want you to use a touch screen in-person, or an app even though you're in your car, talking to a person.
Perhaps an underground movement can maintain a grass roots database of "real human service" locations, after the revolution.
Hospitality will always be a thing. There's a reason why the Enterprise-D still has a human bartender that makes real drinks, even though it's completely superfluous. People will always want to be served by other humans, even if AI Android tech became perfect.
Looking back at them, the taking points about AI are basically the same now as in each episode of TNG, DS9, and VOY that is focused on AI, in the form of Data, Moriarty, The Doctor (and if he has copyright over his holonovel), that Irish village in Voyager, Wesley Crusher's nanites or the ship's main computer becoming sentient…
Fair, but the point still stands. A better example would be Quark's from DS9. Even in a world of perfect replication and perfect androids, people are always going to want a human touch for these things.
Doesn't really matter though. By the time a LLM can replace all programmers, all white collar work will be gone. That's a lot of workers.
Even if we have something competent enough to replace say half of programmers, still a lot of people.
This in turn may or may not be sufficient for work once the issues with bioprinting and/or tissue/organ culture get solved and someone prints/grows a brainless body, connects its organic lower nervous system to a silicon chip, and has the higher functions all performed by an AI on that chip.
And by "someone" I mainly mean The Thought Emporium given they're already trying to do exactly that.
That's called life.
Coincidence...?
> Imagine that you want to plan a trip. A travel bot will identify hotels that fit your budget. An agent will know what time of year you’ll be traveling and, based on its knowledge about whether you always try a new destination or like to return to the same place repeatedly, it will be able to suggest locations. When asked, it will recommend things to do based on your interests and propensity for adventure, and it will book reservations at the types of restaurants you would enjoy.
So sort of like AirBnB, but for the entire process of booking a vacation? Sounds neat at first, though I'm skeptical that an AI will ever match my ability to choose destinations, events, and travel plans that encompass all of my minute preferences (many of which have taken me years to discover about myself). I'm sure that this "agent" won't start recommending shitty and overpriced vacations when it achieves critical market share and the time comes to turn the profit screws!
This gets much, much, much, much scarier in the next section:
> The real shift will come when agents can help patients do basic triage, get advice about how to deal with health problems, and decide whether they need to seek treatment.
Just what I want: our private healthcare industry defining how I should seek medical attention using a faceless algorithm. Nurses and doctors, naturally, have empathy for other human beings and occasionally work in the grey boundaries of the system to help people. But that costs profit. Fortunately, the AI will optimize that empathy away!
> AI agents that are well trained in mental health will make therapy much more affordable and easier to get.
Do mental health experts truly believe that an AI agent can replace an actual therapist? What a scary, dystopian world, where Bill Gates can afford daily therapy sessions with a human but the lowly Microsoft code monkeys will only be able to speak to a machine...
The rest of the post is an absolute horror show of the kind of out-of-touch nonsense I've come to expect from billionaires like Gates:
> liberating teachers from paperwork and other tasks so they can spend more time on the most important parts of the job
Paperwork, sure. What are these "other tasks", though? Lesson planning? I'm pretty sure optimizing away a small portion of busywork is not the most pressing issue in the teaching world. And naturally this task's dark side is a massive yearly cost extracted from our school system, and oodles of collected personal data about every child, starting in pre-K. Think of the efficiency when we can start advertising to these kids based on what subjects they know best!
> If a tutoring agent knows that a kid likes Minecraft and Taylor Swift, it will use Minecraft to teach them about calculating the volume and area of shapes, and Taylor’s lyrics to teach them about storytelling and rhyme schemes.
The next level of "fellow kids".
> If you have an idea for a business, an agent will help you write up a business plan, create a presentation for it, and even generate images of what your product might look like.
If AI can do all of that, what's the human business owner even doing?
> If your friend just had surgery, your agent will offer to send flowers and be able to order them for you.
Isn't the whole point of that kind of statement that a human being thought of another human being and reached out? But I guess Gates just sees the opportunity to take a cut of flower orders. At least most of your friends won't be able to afford surgery when their insurance AI agents use watch data to deny them healthcare.
> Spotify has an AI-powered DJ that not only plays songs based on your preferences but talks to you and can even call you by name.
Wow, AI that _knows my name_? We're really living in the future.
> If you want to buy a camera, you’ll have your agent read all the reviews for you, summarize them, make a recommendation, and place an order for it once you’ve made a decision.
At what point does a human being manage to outsource hobbies to AI? Because if a friend came to me asking about cameras today, but didn't want to do any of the research or learn anything about the space at all to inform a decision, I would simply respond "don't buy a camera, you won't have any clue how to use it".
> To create a new app or service, you won’t need to know how to write code or do graphic design. You’ll just tell your agent what you want.
Interesting. Didn't Gates open this blog post claiming that we wouldn't use apps at all any more? Why would I use an AI agent to create an app in a post-app world?
> Although some agents will be free to use (and supported by ads), I think you’ll pay for most of them, which means companies will have an incentive to make agents work on your behalf and not an advertiser’s.
If there's one thing smart TVs, Spotify, iOS, macOS, Android, and Windows have taught me, it's that paying does not matter. If a company can make more money by collecting data and inserting ads, they will. You'd think Gates would know that when Windows literally advertises right on the desktop.
> Businesses that are separate today—search advertising, social networking with advertising, shopping, productivity software—will become one business.
Aaaaaaaaaaaaaaaand now we've come to Gates' main point: he apparently wants to live in communist china, where a single entity owns literally everything. At the rate of monopolisation we've seen in the USA, at least I can believe this prediction.
https://www.cnbc.com/2018/05/07/bill-gates-i-would-short-bit...
Bitcoin has quadrupled since then.
One might want to take this into account when evaluating his ability to predict the future.
“As an asset class, you’re not producing anything and so you shouldn’t expect it to go up. It’s kind of a pure ‘greater fool theory’ type of investment,” Gates said on CNBC’s “Squawk Box.”
“I agree I would short it if there was an easy way to do it,” he said.
Anyone who thinks the entire "value" of crypto isn't a giant pyramid scheme (or pump and dump) is deluding themselves