Finally, we’ve invented the sci-fi technology of the future! And what do we do? Make tech support chatbots and check insurance claims…
A few years ago, I was prepared for dazzling AI futurism, but the market continues to lag behind my expectations. I want fun AI. I want a Star Trek replicator that spawns any food on command and a robot dog.
Personally, the first thing I did when I got access to deep-learning software was to create “Parm,” an algorithm that could identify types of cheese from pictures. It couldn’t tell gouda from sharp cheddar, but I loved it like it was my own baby.
Unfortunately, there wasn’t any reason for Parm to exist. The backend alone would have cost tens of thousands to put in production, not to mention continuous maintenance requirements. And who was going to use it? Olive Garden? Ultimately, I sunset Parm to make space on my hard drive for icky medical imaging data of tumors. I might have shed a tear.
The reality of AI today is that deep learning is too expensive and too gatekept to be fun. AI products have to justify their existence by generating a significant amount of income, so applications end up restricted to uninspiring corporate tasks.
However, fun and useful AI does exist. Some examples are well-known, like Google’s Perfect Cookie AI, medical applications that screen for covid or cancer, and the Wildlife Insights photo inventory, but I’ve found a few extremely underreported use cases from my time in the industry that deserve far more attention, like these garden-building robots.
I hope that these unique AI applications help you feel a little more optimistic about a glamourous, hi-tech future. I’ve also included some links to free and accessible machine learning software at the end, just in case you feel inspired to make a Parm of your own.
Selfishly, I put this one first because I worked on the technology behind it, but that doesn’t make it any less amazing.
A team of Japanese researchers collected 3D scans of land around Nasca Pampa, Peru with lasers, generating an unapproachable amount of landform data. To tackle data processing tasks that would take humans decades, they trained an AI model to find intentional shapes in the mess of natural formations, specifically looking for a type of ancient art called biomorphic geoglyphs.
The biomorphic geoglyphs are giant pictograms carved into the landscape by the ancient indigenous Nascan people. They depict detailed fire-breathing monsters and vengeful gods, typically ranging in size from 150 to 330 feet across. Hundreds of these beasts were hand-dug deep enough into the soil to still be visible after 2000 years of erosion, indicating that they must have been significant to the people who made them. Scientists suspect the figures were components of religious worship, but their purpose remains unknown.
My home city of Austin, Texas happens to also be the fastest growing city in the U.S.. I know the architects are doing their best to keep up, but their designs often don’t address the issues I need them to fix. However, this might change thanks to AI like Scout. Their system for modeling infinite iterations of architectural features without technical know-how invites anyone to contribute their expertise to design more effective, inclusive spaces.
Scout is an easy-to-use data modeling AI tool that runs criteria like street width, sunlight, urban density, and even ocean views to generate optimized construction plans. Instead of one architectural firm running a handful of designs, now a whole team can collaborate to run thousands of models and select the best compromise to meet everyone’s needs. For example, a social worker could make recommendations on public services which could then be plugged into the model to ideate solutions for public spaces and transportation. Yes, I’m talking about the end of rush-hour traffic.
Another project that took off from my work was the Mayflower Autonomous Ship, an unmanned AI-driven boat charted to retrace Columbus’s course from Plymouth, UK to Plymouth, USA. This time, however, the boat won’t bring any white invaders. Instead, the boat will collect data about ocean health that can be used to combat threats like microplastics and chemical pollution.
The technology isn’t so different from what you might find in a self-driving car, but with an extra spatial dimension and an understanding of maritime traffic laws instead of stop signs. Not having humans onboard also frees up space for cargo — such as advanced computing equipment for tracking endangered whales — and offers huge potential for cheaper worldwide shipping. It’s not far from Hyperion’s iconic self-driving Windwagon!
You might be familiar with AI beating humans at our own games such as chess, Jeopardy, Atari, and perhaps most infuriating of all, poker, but I’ve noticed an astounding detail about many of these victories that remains totally underrated: these bots are often self-trained. They might get an overview of the game’s basic rules, but the rest of their education involves zero human input.
These AI models, like Google’s AlphaGo, teach themselves winning game strategy over millions of iterations. AlphaGo ran 30 million games in its first 40 days. Here’s where it gets really crazy: during that time, it independently recreated the same gameplay trends that humankind came up with starting at the game’s invention 2500 years ago. It matched the chronology of our progress refining game strategy, then, after it discovered the methods used by today’s greatest champions, it kept going. It beat the 18-time world champion in 2016.
Today, modern Go players are copying novel techniques invented by the machine and ushering in a new era of gameplay. It’s only a matter of time until this process, called reinforcement learning, is successfully applied beyond games, perhaps iterating on the best techniques to fly to Mars or evolve the human genome. Someday, one of these models will match our work and then eclipse it, taking us along on an accelerated fast-track to progress.
The Scythians of archaic Russia, Mongolia, and western China buried their dead in carefully-arranged “kurgan” mounds, filling the graves with valuables and, unfortunately, leaving visible piles of dirt aboveground. These circumstances have made the kurgans popular targets for tomb raiders since the 3rd century B.C.
Modern tomb raiders got a new edge with the release of Google maps, which, for the first time, offered detailed topographical information to anyone with a laptop. The visible patterns of burial grounds were easy to spot in satellite images and nearly every ancient gravesite was pilfered by tech-savvy thieves.
Only a few unbothered graves remain, but AI is helping archeologists locate and protect these sites before the bandits can strike. A trained model sifts through satellite mapping data and tags potential gravesites, alerting research teams to get a move on when it finds a viable hit. This innovation has aided archeologists in discoveries of gold, glory, and even mummies like the Siberian Ice Maiden, who was so beautifully preserved that her tattoos still remain intact.
Feeling inspired? Machine learning and deep learning are still emerging technologies, but there are a few entry-level platforms that you can use to get started. No coding required.