Wookyoung Kim · January 23, 2026
How To Predict The Future
I happened to read an article written in 2012 that predicted human-level AI would appear in 2024. According to the piece, by 2030 we will be able to implant computers in the brain.
How could someone in 2012 predict when human-level AI would appear?
Today I translated William Hertling's How To Predict The Future, which explains how to predict when software technologies will appear by looking at trends in hardware development.
How to Predict the Future If there were a sure way to predict the future, everyone would want it. Maybe you are trying to decide which startup to invest in now, or where your company should focus its resources. Or maybe you are simply curious about how the world will change 10, 20, or 30 years from now.
There are many techniques for thinking logically about the future, discovering ideas, and predicting when new technologies will emerge.
I want to share one method I have used myself and found effective. It has proved accurate several times, and I used the same method when writing technothrillers set in the near future. The story goes back to 1994.
Predicting streaming video, and the birth of the spreadsheet
There seem to be two main perspectives for predicting the future of information technology: one that looks at software, and one that looks at hardware. In my view, looking at hardware trends is always simpler and more accurate.
What follows is a story about a spreadsheet I have maintained for nearly 20 years.
In the mid-1990s, I was in graduate school in computer science at the University of Arizona with my close friend Gene Kim, founder of Tripwire and author of When IT Fails: A Business Novel. One of the major technical challenges we were studying at the time was transmitting streaming video over a network. It was difficult because there was not enough bandwidth to send the data, and not enough processing power to compress and decompress video. Improvements were needed in both video compression and TCP/IP.
The funny thing was that no matter how many incremental improvements were made--and dozens of people were attacking the problem from different angles--streaming video always seemed as if it was just about to happen. For years, I heard some version of 'next year will be the year of video' over and over. But it did not happen.
Around this time, I started making a spreadsheet. I entered information about every computer I had owned: processing power, hard-drive capacity, RAM, and modems. Then I calculated the average annual growth rate for each item and extended it into the future.
I looked at the future projections for 'modem speed,' as I called it then; today we would call it internet connection speed or bandwidth. By this point, I was exhausted by claims that streaming video was right around the corner, and I gave up trying to predict advances in software compression. I decided to look only at hardware trends. According to those trends, internet connection speeds were continuing to rise, and by around 2005 they would be fast enough to stream video in real time without relying on extreme video compression or miracles in internet protocols. Gene Kim laughed when he heard my prediction.
Nine years later, in February 2005, YouTube appeared. Streaming video had finally become reality.
The same spreadsheet also predicted that music download services would appear in 1999 or 2000. Napster, the first P2P music-sharing service, launched in June 1999.
The data was surprisingly accurate over the long term as well. Using only two data points, the modem I used in 1986 and the modem I used in 1998, the spreadsheet predicted that I would have a 25 Mbps connection in 2012. I currently use a 30 Mbps connection, so for a 15-year prediction, that was quite accurate.
Why this method works, Part 1: linear vs. nonlinear
I did not fully understand the concept at the time, but looking back, what I was doing was this: using linear trends, gradual progress over time, to predict the timing of nonlinear events, or technological disruptions. I was calculating when the underlying hardware would open the way for a breakthrough. That is what I meant above when I said I 'gave up trying to predict software advances and decided to look only at hardware trends.'
Of course, you still need to imagine what kinds of advances might happen in the future, though trends can inspire ideas. What this technique does is let you map an idea to its underlying requirements and determine when it might become feasible.
For example, it can answer questions like these:
- When will magnetic-platter hard drives be produced for the last time? 2016. Around 2006, I charted the capacity growth trends of magnetic-platter hard drives and flash drives, and the chart showed flash surpassing magnetic media in 2016.
- When will a general-purpose computer become small enough to implant in the brain? 2030. According to the continuing miniaturization trend of computers, by around 2030 a computer will be the size of a pencil eraser, small enough that implantation will not be difficult.
- When will a general-purpose computer be able to simulate human-level intelligence? Between 2024 and 2050. It depends on how you estimate the complexity of human intelligence and how many computers you devote to the simulation.
Wait, human-level artificial intelligence in 2024? Gene Kim would laugh at that too. AI is an enormously difficult field. Haven't predictions that artificial intelligence is just around the corner been repeated for 40 years?
Why this method works, Part 2: crowdsourcing
When I participated in a panel at SXSW on the future of artificial intelligence, one of my fellow panelists objected. He could not agree with the idea that exponential growth in computing power alone would be enough to develop human-level intelligence on computers. Artificial intelligence has very difficult problems, he said, and each of those problems requires the efforts of extremely talented researchers.
That is not wrong. But the world is large, and there are many talented people. The principles of open source and crowdsourcing are already well understood. When enough talented people work on a problem, especially when they collaborate openly, progress comes quickly.
I once wrote an article for IEEE Spectrum titled The Future of Robotics and Artificial Intelligence is Open. In it, I analyzed how hobbyist developer communities were creating inexpensive autonomous flight hardware and software for unmanned aircraft. Something that once required skilled researchers in labs and $20,000 can now be built by hobbyists part-time for $500.
When hardware becomes powerful enough, invention becomes possible. Before that, it is impossible. It is like trying to build a car without a motor.
And when that powerful hardware becomes widely available, invention becomes inevitable, because it enters the realm of crowdsourcing. Hundreds or thousands of people can now contribute. Once enough people had enough bandwidth to share music, it was inevitable that someone, somewhere, would invent online music sharing. Napster was only the first.
IBM's Watson, which won the American quiz show Jeopardy, was built with $3 million worth of hardware and used 2,880 processing cores. When the same level of computing power becomes possible on personal computers, roughly around 2025, IBM's research team will not be the only group playing with advanced AI. Hundreds of thousands of AI enthusiasts around the world will contribute to open-source projects comparable to Watson. At that point, AI will truly take flight.
(If you doubt that so many people would be interested, remember that more than 100,000 people enrolled in Stanford's free AI course. A similar number enrolled in the machine learning / Google self-driving car course.)
Of course, this technique does not work for every kind of innovation. Wikipedia was a tremendous invention in the way knowledge is curated, and it depended on the invention of the wiki itself. But even in hindsight, it is hard to say that Wikipedia could have been predicted, much less when it would appear.
(If you had the idea of a publicly curated online knowledge system, you could have applied the litmus test of internet connection speed to estimate when enough contributors and users would be available. A document system like a wiki is useless if people have no way to access it. But that is a digression...)
Objection, Your Honor
The common objection is this: this kind of growth cannot continue forever; eventually it will hit fundamental limits. Just look at computer processing speed, people say. Aren't there limits in silicon manufacturing, heat dissipation, signal propagation, and so on?
I remember first reading this sort of argument in the mid-1980s, in an article about Intel's 80386 processor. It said that the chip was being manufactured with an 800 nm process, and that we would soon reach fundamental limits that would make further miniaturization difficult. In processor technology, smaller is faster.
But manufacturing technology has kept getting smaller. Limits are overcome, bypassed, or solved through technological transitions. For a long time, improvements in processing power came mostly from increases in clock speed. When that approach began to hit its limits, the industry shifted toward increasing speed through parallelization: more processing cores, and more execution threads per core. In the future, graphene processors or quantum processors may appear. But whatever the underlying technology, speed will likely continue to increase at roughly the same rate.
Why predicting the future is useful: prediction and verification
I use this technique in two ways. First, I use it as a seed for brainstorming. When you extend linear trends and get a clear sense of where a technology is headed, you become free to imagine what that technology will make possible.
For example, I had never seriously thought about neural implant technology until I looked at a chart of physical size trends and realized that neural implants would be feasible in the near future. And if something is technically possible, it is effectively inevitable.
What OS will they run? Which app store will neural apps come from? Who will sell advertising space in our brains? What else can we do with a coin-sized supercomputer?
Second, I use it to test other people's claims. There is a company called Lifenaut that says it will preserve people's data and provide personality simulations after death. It is an extremely compelling idea, but it is in a situation similar to video streaming in 1994. The hardware is not there yet. If human-level AI cannot arrive before 2024 at the earliest, and even then only on a cluster of more than 1,000 computers, it seems impossible for Lifenaut to provide realistic personality simulations before that.* On the other hand, if the project is willing to continue for 15 years, it may be in the best possible position when the required performance becomes available.
At a recent 'science fiction and scientific fact' panel, most of the other panelists and audience members thought strong AI was 50 years away and brain augmentation technology was still 100 years away. If those things are that far in the future, they become things we do not need to think about now. That seems like a dangerous attitude.
* A common objection is, 'A software implementation will be far more efficient than what nature implemented in the brain.' Well, I would bet on millions of years of evolution.