A lot…
But two notable progressions stand out. The first will be personalized bots. The second will be personalized autonomous AI systems. Allow me to extrapolate on both.
This post was inspired by Lex Fridman’s interview with (God knows what he is) George Hotz.
It is 2023 and the hot new job is prompt engineering. Chat GPT, an AI chatbot, is the new thing and has enthralled millions of us with its current capabilities but also its future possibilities. You interact with it by writing a question or command, and it returns magic. But how do you interact with it? The basics we get, like asking a simple question, or giving a task like write me a recipe for lemon garlic pasta… but… more complicated interactions require some knowledge and skill. Who has such knowledge and skill? A prompt engineer. They can get more complex and precise results from a chat bot. The interaction between person and computer is a request and response pattern, and better requests can mean better responses.
If we look briefly at the past 25 years of request and response patterns, we see the following progression:
- Coding is the way to talk to a dumb computer. It doesn’t know you, doesn’t care about you, and needs exact instructions. It
’s output is very predictable. - The first internet search engines also didn’t know you; they only knew about the millions of web pages on the internet. When you filled in keywords, they returned a matching set of web pages, the same set for everyone, and you had to sift through them to see which one you preferred. The response is not so predictable because there was so much data to sift through and potentially infinite ways to filter them.
- Eventually search engines started collecting information on you, like location, past searches, etc, and could tailor the result set to you personally.
- Voice assistants like Alexa got to know a little more about you and when you asked for information, they skipped the manual step of you having to sift through a result set and gave you a specific answer. If you didn’t like the answer, there was not much you could do about it. They took the possibilities a step further and could perform basic personal digital tasks like send emails or play music.
- The current set of chat bots, such as Chat GPT and Bard, go a step further with personalization. For one, they let you interact with them as you would another person – a step far removed from coding. As Hotz explains, coding was “do what I say” and these proper AI systems “do what I mean”. What’s more, the conversational request and response can be iterative. That is, it maintains the conversation in memory and any corrections you want performed can be based on the previous tries. They too can send emails and play music, but they can also create information like articles, code, images, videos, spreadsheets, and the like.
Each step of the way, we are getting systems that are better at knowing what you want and more powerful in their ability to give it to you.
The progression along the way has been getting more intelligent, but it is the current batch of chat bots that we consider to be full-fledged AI. They are a technological milestone comparable to the printing press, the combustible engine, airplanes, the internet, and smartphones and our fascination and fear are warranted. They will transform human society. But to put icing on the cake, these proper AI systems need one important thing: a proper data set on you.
Up until now, tech companies have been collecting data on you through osmosis and piecemeal data mining. Their motivation has been to predict what you want to spend money on so that they can manipulate you ever so slightly while you are doing the things you actually want to do. You look up Thai restaurants in the Baltimore area and it makes note of the fact that you like Asian cuisine and may live in the Baltimore area. But soon that data gathering process will change from “them mining your data for sales” to “you forming a data set for your own digital representation,” and when it does, we will see the real leap AI presents.
You will need to teach AI systems about you. About your personal life, your health, your career, your (relation to your) kids, your (relation to your) partner, your family, your finances. I put relation in parentheses because it will be focused on you, so with regards to other people, it will translate that to “your relation to”. But of course, it will need to have an understanding of those other people too. How that would work would be a whole other article.
Personalized Chat bots – Tell me a little bit about yourself
For me, this point on the horizon in now visible since Chat GPT came out. Imagine, you spent hours telling an AI system about yourself in a structured way. You answered long questionnaires. You ranked your favorite films and music, and said why you liked what you liked. You detailed your daily activity for weeks – everything you did and why you did it. Or why you think you did it. What you ate, what you drank. With whom. How you felt about it all. For a moment, ignore the work involved in doing this, and whether or not you would actually like to do it. Just imagine you did it, and now you ask Chat GPT to do something. Or ask it where you should go on vacation. If you should quit your job, and if so, to go do what. Ask it to play your new favorite song. Ask it to write you a song. Can you imagine?
This personalized integration with AI will be here very soon… but that is just a peek into the future. One that is a bit scary, but imaginable and not really mind blowing. We’re already wearing heart monitors, and step counters, liking songs on Spotify and getting recommendations on what to listen to. But the systems that provide these services own your data, not you. And they don’t share that data with each other (well, not completely). What’s coming is companies that will gather and store your personal data for you, like DropBox stores your personal files. And they will make this data available and consumable for various AI systems.
Sounds exciting and scary… But… In this scenario, you are still explicitly telling AI what to do. You want to unleash the powers of AI, but there is a bottle neck: you.
But further down the road new AI infrastructure will unclog the tunnels. This future of AI will be transformational to human society, and though it seems a little distant, I believe it’s in the cards for us since it already happened hundreds of millions of years ago.
Autonomous AI systems
In short, we will do with AI the same thing our genes did with us. I could spend a few pages and days trying to get the exact specifics and wording down, but neither of us is up for that. What essentially happened was that genes (the basic unit of natural selection) banded together with other genes to create DNA, which was used to create multi-cellular life forms. Life is about self-replicating molecules. RNA is self-replicating, but limited. So, making the step to DNA – a combination of many genes that culminate into a set of instructions on creating a completely new organism — , allowed for a more robust mechanism for gene proliferation. And now we have plants and animals hundreds of millions of years later. Their sole motivation is still gene proliferation. Do they always do it? No, but they try and statistically are successful. The genes wrote the code for these plants and animals telling them what to do and how to do it. Through DNA, they created multi-cellular life as robots to execute their goal. It wasn’t one gene and one robot; it was groups of genes forming a set of instructions that could be spawned into millions of robots. Those instructions said, “This is what the world looks like, so you’ll need to do A, B, and C to deal with it. And this is your goal: replicate me as much as you can. ” We humans are still bound by this arrangement. We have instinct and an innate sense of our environment, and we try to pass on our genes based on that information stored in our biology – information created by the genes themselves.
Do the genes sit there are worry, “I hope they don’t get out of control”? Of course they do. They set us up with a specific goal in mind. They are very disappointed when they see us smoking meth and masturbating. On later iterations, they tried to protect against such behavior by building in instructions for guilt and remorse.
Now, let’s look at our turn in the driver’s seat. In the 21st century, we, robots created by genes, are now the ones who will be creating robots.
In 2023 we can dedicate a few minutes writing one command to send a lemon garlic recipe to the five friends. But imagine a system that knows us so well that takes its own initiative and runs this task without asking. The more we fuse with AI the closer this possibility will become. We’ve all thought about what it would be like to have a clone, and soon we all will, albeit a digital clone. AI systems will become our autonomous digital representation in the world. We won’t have to be bothered giving it explicit commands, since it will know us so well as to know what we want, and have the inherent scalability of computers, so it will be out there in the world doing what we want 24/7.
This is AI truly doing what we mean. And just what do you mean?
At some point, we will train our AI by telling it: This is me, this is my general goal in life, these are the boundaries and environment you will be working in, now go make these goals happen. It will know what you mean. At least, it will have a good idea, and though it will be saving us time and bolstering our output, we will have to spend time correcting it along the way. This AI training will be a never-ending part of your life to make your robots more and more what you want them to be.
This is the work our genes did millions of years ago. For example, they said to fish: It’s really wet, so you’ll need to get oxygen from the water. Gravity plays a minimal role in movement, so get around by flapping your fins. Primary focus is mating and creating baby fish. The training came once a generation when the fish did or did not pass on their genes in their lifetime. The ones that did not were blindly tweaked in hopes they’d do better next time. This is what comes after prompt engineering: a database of information which will act as your personal instructions for your robots. Eventually, it will guide your robots even when you are not there. Even when you are dead. It will go a step further than your DNA. It will not only represent your genetic makeup, but also your personal circumstances. Whereas DNA contains instructions for your nature, this data set will be your nature and your nurture. Each of us will build up our own database over time. Initially, we will still be the ones to initiate AI actions, but eventually, it will know us so well and we will be so content with these actions, that we will start to hand the steering wheel and the gas pedal over to AI systems.
When you make this last step, you will be able to employ multiple AI systems to do your bidding.
And to clarify, these systems will come in various forms. Some will be simple apps on a smartphone, some more complex systems with complex interfaces. And others will be actual, mobile robots physically interacting with the world. Your data will be the common element they share.
At some point, everyone in the connected world will have at least one personal AI system. Each AI system will cost money, so of course, as with all things, the rich will have more than the rest of us. Like the plants and animals of today, these systems will also struggle against each other in a world of limited resources. Eventually, our personal systems will form groups and work together, just as individual genes grouped with others to form DNA. You will create a system for music, and probably do so within a federation of other systems that share similar musical tastes. You’ll have one for investing, and set it up within a federation with other systems that also have similar investing goals and means as you. The systems will share costs, learn from each other, and work together on the common goal. Since your personal AI systems will operate in groups, they won’t get a completely personalized result, but the net effect will be better than if each system was out on its own. Sound familiar? Yup, just like us humans and our group lives. Just like genes and their collective in the form of a plant or animal.
Sound too abstract? If you’re like me, you want specifics. I mean I have never let an AI bot search for recipes and send them out to friends. I had to imagine that example, but it was imaginable.
What the hell will these next-gen AI systems be like? What will they do?
Let’s see… They will:
- find mates for us
- find entertainment for us
- communicate on behalf of us – to family, friends, and
social media followers
- create entertainment for us
- buy things for us
- find and summarize work related trainings and courses
- be our own personal physicians
- manage money for us
- help us with parenting
- help with legal issues
- summarize insurance policies and other contracts for us
- settle on our mortgages
- sell our houses for us
- buy and manage our cars
- manage our houses
This is just a quick list. I could spend 20 minutes on each point alone.
Most notable, they will help us with our jobs. Your company will certainly be using AI systems, but each of its employees will be bringing their own AI systems to work to help them with their jobs. Just as your unique qualities are the reason you were chosen for a specific job, your AI, as an extension of you, will have unique qualities that your employer will also want on board. The better AI you have set up in your life, the more employable you will be.
So… a compelling vision of the future. It’ll be here soon. It will create a dramatic change to how we live our daily lives, and it will have statistically significant effects on society, much like the internet and smart phones did. Unfortunately, it probably won’t make us happier. But if you’ve been paying attention, you already understand that happiness was never the point of technology. AI and all its manifestations will take us even further away from our ancestral roots. But that’s the deal: our subconscious is in control and deep down all it wants to do is pass on our genes. AI’s raison d’etre is to help us do that. Many people will dispute this point, and I urge those people to go read Quanology for a more complete explanation of human motivation and technology’s role therein. But for now, the key point is that genes are the puppet master in our subconscious and they want to be proliferated, so they create technologies to help us proliferate them better.
As with the internet and smartphones, new technologies are both tools and drugs. Our subconscious, created long ago, can’t easily tell the difference and likes them for both these reasons. The successful people in the world will harness their power as tools and the less successful people will use them as drugs. AI will be the same. The list above provides examples of how they will work as tools. I’ll have to make another list of how they will work as drugs. But maybe you can imagine a few AI drugs already, having witnessed porn on the internet and mindless video feeds on Tik Tok.
As always, it’s hard to prepare our brains, let alone society, for the changing times. The best we can do is nurture awareness in ourselves as to which one of the two uses we are engaging in: the tools or the drugs. And, taking breaks every now and again from the modern world to regress as much as possible to that time in history when the only things we had were time, nature, and each other.