“Reasoning AI” and the path to AGI
Want to listen to this week’s edition of Product Faculty’s Newsletter? Scroll to the bottom of this post to tune in.
Last time, we looked at how artificial intelligence is essentially mimicking human intelligence, using different models to emulate how we sense things, understand our environment and even create.
Going a bit deeper, we can break down AI into three distinct categories.
Artificial Narrow Intelligence (ANI) refers to an AI designed to perform highly specialized tasks. As the name suggests, it has a narrow focus, excelling at its assignment but lacking the ability to perform other tasks.
Think of how AI is being used for web search, your Netflix recommendations or your smart speaker digital assistant. ANI has seen major advancements during the last 10 to 20 years, to the point where it’s now seen as part of our daily lives.
Generative AI refers to algorithms used to create new content. Generative AI has been making all the headlines recently, ever since ChatGPT’s release at the end of 2022. Since then, other models like Gemini and Claude have been released. The initial releases could only generate text, but today’s models can create photo-realistic images, audio and even video.
For many people (including the OpenAI team), Artificial General Intelligence (AGI) is the goal. The definition varies depending on who you ask, but typically AGI refers to a type of AI that does everything a human can, but better.
The road to AGI
Some believe that AGI is right around the corner.
On his X feed, Elon Musk has shared his belief that AI will likely be smarter than any single human next year, and that “by 2029, AI is probably smarter than all humans combined”.
Others are a bit more conservative in their estimations, suggesting we’re decades away from AGI.
In a recent blog post, Sam Altman said it was possible we will have superintelligent AI “in a few thousand days”, adding that it could take longer but he’s confident we’ll get there.
Either way, reaching AGI won’t happen overnight. Instead, it’ll follow a path, with five distinct steps.
Conversational AI - The initial release of ChatGPT marked the first step in the journey to AGI, enabling users to interact with AI technology in a conversational manner, without having to follow strict syntax and other rules.
Reasoning AI - The recent release of the o1 model is the next stage, introducing the ability to reason through complex tasks and problems.
Autonomous AI - At level three, AI agents will be capable of autonomously performing tasks for days, dramatically boosting business efficiency and productivity.
Innovating AI - Innovators will be AI systems that can independently improve and innovate processes, enhancing efficiency and effectiveness.
Organizational AI - Organizations is the final AI stage, performing all organizational functions autonomously. In other words, we won’t be looking at individual agents, but entire businesses comprised of different AIs working together.
What “Reasoning AI” means for the world
If you’re reading this newsletter, it’s safe to assume you’re familiar with conversational AI. However, the release of the o1-preview and o1-mini models has introduced reasoning AI, which not only represents another step towards AGI but also signals a change in how we approach AI, both as users but also as product managers.
An AI model that’s capable of reasoning has massive implications for math, science, healthcare and other similar fields.
As more people begin to understand what these models are capable of, it’s entirely likely that we’ll unlock new discoveries and new possibilities within this space.
These new models are operating at an unprecedented level, performing similarly to PhD students on benchmark tasks in biology, physics and chemistry. It’s also great at math; in a qualifying exam for the International Mathematics Olympiad (IMO), the reasoning model scored 83%, blowing GPT-4o out of the water (which only solved 13% of the problems).
Take a moment to think about the implications here. With a PhD, you’re essentially expanding the scope of human intelligence by 0.0001%. If we can start using AI at that level and in that space, then the scope of human knowledge starts to increase significantly more.
Anecdotally, people have told me how the o1 model has saved them months of research, or coded complex programs in a fraction of the time a human can. I was speaking to someone who’s currently carrying out research in cardiology, and in his view, the o1 model knows as much as he does.
What’s especially exciting is how these areas – healthcare, math science – where AI is starting to impact are also the areas that will have an outsized impact on the rest of the world.
What “Reasoning AI” means for product managers
Whether you’re trying to change the world or not, the advent of reasoning AI is going to change the way you interact with AI and the way you create products.
Up till now, a lot of product management work has been happening with a zero-shot approach. You give the AI a command, it gives you an answer and then you try to optimize against that command.
We’re going to start seeing more of an asynchronous approach – you give the AI a command, but then the model comes back with a response minutes, hours, maybe even days later.
That’s a completely different modality to start to think about. Product managers need to start focusing on what they can achieve with long-running tasks instead of thinking in terms of just short bursts.
The model can now take the time to think, create a plan and then execute against that plan. If your product is capable of taking time to carry out more complex actions and reasoning before responding, you’re in a prime position to take advantage of this latest technology.
For example, what would it look like to have an AI model complete your taxes and then let you know when that’s done? Would you want to sit there and watch it work through every line, or would you prefer it to let you know when it’s done and give you a summary of the required actions?
Think about the interactions you have with people in these types of industries today.
When you go to a financial planner, they ask you a bunch of questions and then come back with a financial plan to review with you.
When you go to a doctor, they listen to your symptoms, they go through their tests and then they give you their diagnosis.
It’s time to start thinking in the same way about AI products.
Slowing down for better results
This might all sound a bit counter-intuitive. After all, why would you want to wait for something if you don’t have to?
Mental models have been slowly changing over the last couple of decades, and people have been trained to think that faster is always better.
For example, do you remember the joys of connecting to the internet through a dial-up modem? After listening to that crazy dial tone for a bit, you’d have to sit and wait for anything to happen. If you wanted to download a song, then you’d have to wait ten minutes or more.
That time has come down so significantly that it’s become irrelevant. For most people, you can click play and listen to a song instantly.
People have become accustomed to that immediate result. That’s one reason why AI only really began to capture mainstream attention when tools like ChatGPT made it possible to type in a command and get that instant response.
However, it’s now time to change the way we approach AI.
We need to start showing our customers that, when they’re willing to wait a little longer, they can get a better result.
This might mean taking a lesson from video game designers. When you’re in the middle of an exciting game, nobody wants to take a break from the action and stare at a loading screen. As a result, designers have found a variety of ways to hide the loading screen, such as replacing it with a mini-game or cutscene.
Similarly, as product managers, we need to find ways to continue giving our users value, even while the larger action that’ll provide the real value takes place in the background.
People have grown accustomed to instant results, but I don't believe instantaneous responses will be the future. Instead, we’ll be able to achieve more with a model where you can set it and forget it.
The back-and-forth interaction model that we’ve all grown used to over the last two years is going to likely fade away and the product managers who can think bigger will benefit the most.
What the future holds for product managers
We’ve covered a lot of ground in this edition of the newsletter, looking at the latest developments in reasoning AI and how that represents yet another step along the path to AGI.
So, what does this all mean for product managers?
Well, if we get to that fifth level, organizational AI, we’re all probably going to be looking for different jobs (or at least different ways to work at that point).
But, before anyone starts to panic, product managers are still going to play an important role for the foreseeable future.
We’re still going to need product managers – people just like you – who can take a step back and see the bigger picture of how these models can be positioned to solve the problems our customers are facing.
At the top level, product managers need to:
Determine the problem space
Figure out the strategy
Understand the model capabilities
Understand how the user will traverse through the system
Understand the specific stages within the use case where the models could play a factor
All of this work is essential and is going to remain essential for a long time to come.
I also believe that we’ll start to see product managers go a step deeper and start getting into the technical weeds of what the systems are doing and where the AI flow matters most in those systems.
As a result, the best thing you can do right now to improve your career prospects is to go beneath the surface level and commit to mastering the art of AI in product management
Today’s challenge
We’ve talked a lot about the new o1 model and what it means for product managers.
However, I want you to go a step deeper.
OpenAI has published the o1 System Card, giving users and developers an overview of the o1 model series, its features, capabilities and more. It lets you know what it’s good for and where it’s not so good. It lets you know how to use it and how to get the most out of it.
It’s 41 pages long. It’s very detailed.
As a result, not many people are going to read through the whole thing.
I want you to be the exception.
Set aside an hour of uninterrupted time and go through the entire document. Don’t skim it. Don’t ask ChatGPT to summarize it for you. Sit down and read through it. Think about the products you’re working on and ask yourself how exactly this model could benefit you and your customers.
By making the effort to truly familiarize yourself with o1 and its capabilities, you’ll have a significant advantage over other product managers and be in a position to get the most out of the new models.
Perhaps more importantly, you’ll be building a valuable habit, one that separates you from the rest – diving deeply into foundational resources to understand new technology at its core.
This isn’t just about the o1 model; it’s about training yourself to be fully informed and proactive, so you can leverage AI innovations as soon as they emerge. By developing this habit, you’ll enhance your strategic insight, making you a more effective and adaptable product manager in the long run.
When you’re finished, we want you to hit reply and let us know what your biggest insight was. We look forward to seeing your responses.
P.S. Want to take your product management career to the next level with the latest in AI knowledge? Check out our #1 rated AI Product Management certificated course on Maven. You’ll get the latest insights on how AI is affecting product management, direct access to us for questions and feedback, and an active community of like-minded product managers.
Enjoy your newsletter-exclusive $250 discount with this link.
Voiceover
What did you think of today’s email?
- 😊 Loved it 🧠🧠🧠
- 😐 It was ok 🧠🧠
- 😞 Terrible 🧠
 | A guest post by
|