A Model to Predict AI’s Impact
Introduction
How might you predict AI’s impact on edtech or construction? Could a model be developed that might predict AI’s impact on any industry?
If so, it could help guide product or service development, institutional asset allocation, positioning strategy, etc.
To attempt to predict how AI will impact an industry or submarket, I’ve developed what I call “the Density Model of Artificial Intelligence.”
Essentially, it says that AI will increase the “density” of intelligence. Stated differently, AI will increase intelligence per unit of time, space, and currency of a given situation. Further, AI will substantially improve situations where multiple AI systems can communicate and where increasing the number and diversity of sensors is possible (e.g., infrared + visible light vs. only visible light).
I appreciate your time reading this. If I’ve done a poor job of explaining these concepts, the model isn’t clear, or you have additional ideas, I’d appreciate hearing from you. Thanks Sean.
Greater Intelligence
Why didn't the guy 10 cars ahead of you tell you about that pothole?
The first component of the model is what I call: “greater intelligence,” but before I get too far, I need to provide a definition for “intelligence.”
For the sake of this model, "intelligence" is the ability to integrate data from multiple sources, create an accurate model of the world, and generate decisions that achieve a goal.
Humans are incredible model builders. We can enter a relatively new domain and, with little data, build good-enough models that allow us to make effective decisions. Currently, but not for very long, AI systems are poor at this. They tend to require a massive amounts of data to reach parity with our abilities.
However, where AI has an edge and will increasingly have an edge is in the integration of new data to update their models. Our cognitive biases tend to fix our mental models, even when contradictory information would rationally invalidate them. In general, it is only when there is a massive disparity between reality and model that we change and even then, we change slowly.
This is not true for AI systems. Why?
First, the typical "costs" associated with an error that might come from erroneously changing a world model, are far lower for AI systems. They can easily duplicate and fork with one system staying on the previous model and the other trying one or more new models. If one "dies" due to an error, the cost is low.
Additionally, their inner concepts are less "fuzzy" than ours allowing them to better sense anomalies that may precede a substantial change to the world.
Further, there are less constraints on them scaling. This scalability gives them the opportunity to more easily integrate new sensors. Self-driving cars are already attempting to integrate LiDAR and whether or not that is the path toward full autonomy it speaks to how additional sensation can be powerful in specific situations. For example, security applications benefit from the ability to seamlessly integrate both visible and infrared camera data.
Additionally, this scalability allows them to more effectively integrate input from other AI systems.
While we get annoyed when someone tells us how to drive, self-driving AI systems can easily integrate both data from the cars around them and all cars in a city (and they don’t roll their eyes). This gives AI-powered cars not just the ability to avoid traffic and potholes, but also automatically alert the city, and do a whole host of things we haven't imagined yet.
The Implications of Greater Intelligence
It isn’t groundbreaking to suggest that increased intelligence should mean better decisions which will increase the likelihood of achieving a goal. Further, it should also be fairly obvious that greater intelligence means achieving that goal using a more effective approach, i.e., less resources: time, money, energy, etc. But, what may not be obvious are the possibilities created by the new types of intelligence AI offers - for example, the above mentioned, collective intelligence pothole scenario.
However, artificial intelligence doesn't just mean new forms of increased intelligence, but a greater density/distribution of intelligence.
Greater Intelligence in Time/Cost or...More Intelligence per Unit of Time
Why don't you go to the doctor every day?
Presumably, if you went to the doctor every day you would extend your life. Seeing you every day would allow them to discover and treat potential issues early enough for them to avoid being serious.
But, a doctor exchanges their time for money. So access to their intelligence is limited and therefore pricey.
Our brains are incredibly fast at some activities, unfortunately, making intelligent decisions is not one of those activities. As the complexity, stakes, and number of decisions increases, the time it takes our brains to make a decision increases significantly. Further, under these conditions (e.g., stress), additional cognitive biases kick-in and our decisions can become less intelligent.
Due to the speed of computation of non-organic systems, artificial Intelligence has the potential to integrate more data and make reliable, intelligent decisions in less time. And as computation technology improves, the speed of this intelligence continues to increase.
Additionally, because AI systems can duplicate themselves, the cost of these decisions is far less than similar human-systems (e.g., doctors, lawyers, marketers).
Finally, since the speed and cost of intelligence is lower, the speed and cost of shared artificial intelligence is significantly less.
The Implications of Greater Intelligence Time-Density
Having more collective & individual intelligence, in less time, per dollar has the potential to redefine pretty much every aspect of our lives.
We are entering a world where your customized AI can both make real-time decisions to improve your life, while getting increasingly better at those decisions over time as it both learns more about you and learns from the experiences of other AI systems learning about their humans.
It is easy to imagine a scenario where your AI doctor will monitor you on a sub-second basis, using hundreds of sensors. It will automatically communicate with your AI chef about how to amend your diet in real time, testing the food for pathogens, and learning from AI chefs around the world.
Greater Intelligence in Spatial Distribution/Cost
In addition, to an increased intelligence per unit time, per dollar, AI allows for increased intelligence per unit of space.
Implications of Greater Intelligence Space-Density
Physical
This increased density of intelligence is more familiar to us. We have smartphones, smart watches, and smart air conditioner units. We are pushing intelligence into more and more of the physical spaces of our lives. What we have is increased control, what AI will provide is increased intelligent autonomy.
In the example I gave above of your AI doctor, your smart fridge is now able to make reorder decisions based on its communication with your AI primary and the various local warehouses or even local farms.
Virtual
I believe an increase in the spatial-density of intelligence will have an even more significant impact in virtual space. Experiences like the metaverse and websites will gain a whole new level of interactivity and depth because they are infinitely and immediately configurable and individually customizable.
Gamers are well used to this. Many games live or get abandoned based on the quality of their AI.
We are already seeing comment sections that self-regulate. We will increasingly see blockchains and AI combine. Imagine your Charity AI, automatically evaluating a number of decentralized autonomous charitable organizations (DAOs) and autonomously optimizing the allocation of your charitable giving in alignment with your values.
Physical-Virtual Intersection
We will also see AI speed and empower the adoption of physical-to-virtual experiences.
For example, I can envision the following scenario: your AI Stylist learns your preferences from your activities in the virtual world. It then communicates those preferences to a brand’s AI Stylist. When you walk into that brand’s brick & mortar store their AI Stylist knows how to design clothing that will appeal to you and does so in real time. You will try these on in their AR mirror (a technology that is already in use). Then, as you sip complimentary champagne, their robot tailor makes your outfit.
Summary
In short, we are able to cram intelligence into more areas of our lives because figuratively intelligence is getting smaller, faster, cheaper and more connected.
Applications of Intelligence Density - Decision Map
One potential way to apply this model is to map out all of the decision points in a customer’s journey with a product (or industry) and see if there aren’t ways to add great collective intelligence or alternative sensors.
For example, a product lead might select a complex, goal-base scenario (e.g., diagnosing a disease or making a marketing decision), determine all the decisions needed to achieve the goal, and then asking a set of clarifying questions to determine what would happen if intelligent decisions could be made more frequently, cheaper, at different locations, or using different types of data. Here are a few question examples:
What might happen if we can make those decisions more quickly?
What would be possible if we added different types of sensors?
What if there were two or more decisions to be made simultaneously?
I’m sure you can think of others. If so, please share in the comments.
A Quick Note on Timing - Artificial Intelligence Pace Will Likely Outpace Expectations
It was only a few years ago pundits were suggesting that what is currently possible, was potentially decades into the future. The speed of transition from AlphaGo to AlphaZero to AlphaFold and from GPT to GPT3 to ChatGPT, and soon GPT4 should suggest that under estimating the pace of AI is dangerous.
My hope is that my density model aids these adjustments.
Here are a few forces that I believe will continue to accelerate the pace of AI development
AI will be Recursive
AI is helping us program.
You can tell ChatGPT to create a python app for you and CoPilot intelligently aids programmers.
We are currently developing artificial systems ourselves. As a result, development, while blisteringly fast, still moves at “programmers speeds.”
When AI learns to effectively build artificial systems, AI development speed will increase by several orders of magnitude.
AGI is not needed to outperform us
You don't need to be human-intelligent to outperform humans.
A few years ago, I would hear AI pundits suggest that humans still had decades of breathing room because Artificial General Intelligence (AGI) was far off and the science community wasn't even sure human-level intelligence was possible in a machine.
While I would argue that it certainly is possible, you don't even need to engage in that debate to see that AI systems will significantly outperform humans in most tasks. In most applications, being hyper-connected and adaptive in the way AI systems can be, is enough.
For example, how many times have you hit a pothole?
Calling back to my above example, an AI enabled car communicating with other AI enabled cars looking at the road can let every car know right where that pothole is and proactively avoid it. Then, it can "call" the city to request a fix.
Quantum Computing
Not all tasks benefit from quantum computing, but many of the matrix optimizations that underlie AI systems and likely many that we haven't imagined yet will. As quantum computers develop (and as AI helps them develop faster) they will reduce training time and decision time. As fast as current AI systems are at making intelligent decisions using current silicon, it still takes seconds or minutes. Using quantum computing it will be done instantly.
In combination with recursion, connection, increased sensation, and wide distribution, we will eventually hit an inflection point where AI performance is unrecognizable.